refactor(desktop): tighten right-rail tab close API

Promote closeRightRailTab/closeActiveRightRailTab as the single
public entry point. Drops the activeTabRef + handleCloseDocument
indirection in ChatPreviewRail, the unused $rightRailHasContent
atom, and the legacy dismissFilePreviewTarget alias. -70 LOC.

.dockerignore

@@ -5,7 +5,15 @@
 
 # Dependencies
 node_modules
+**/node_modules
 .venv
+**/.venv
+
+# Built artifacts that are regenerated inside the image.  Excluded so local
+# rebuilds on the developer's machine don't invalidate the npm-install layer
+# that now depends on the full ui-tui/packages/hermes-ink/ tree being present.
+ui-tui/dist/
+ui-tui/packages/hermes-ink/dist/
 
 # CI/CD
 .github
@@ -17,3 +25,7 @@ node_modules
 
 # Runtime data (bind-mounted at /opt/data; must not leak into build context)
 data/
+
+# Compose/profile runtime state (bind-mounted; avoid ownership/secret issues)
+hermes-config/
+runtime/

.env.example

@@ -384,9 +384,9 @@ IMAGE_TOOLS_DEBUG=false
 # Default STT provider is "local" (faster-whisper) — runs on your machine, no API key needed.
 # Install with: pip install faster-whisper
 # Model downloads automatically on first use (~150 MB for "base").
-# To use cloud providers instead, set GROQ_API_KEY or VOICE_TOOLS_OPENAI_KEY above.
-# Provider priority: local > groq > openai
-# Configure in config.yaml: stt.provider: local | groq | openai
+# To use cloud providers instead, set GROQ_API_KEY, VOICE_TOOLS_OPENAI_KEY, or ELEVENLABS_API_KEY above.
+# Provider priority: local > groq > openai > mistral > xai > elevenlabs
+# Configure in config.yaml: stt.provider: local | groq | openai | mistral | xai | elevenlabs
 
 # =============================================================================
 # STT ADVANCED OVERRIDES (optional)
@@ -394,7 +394,25 @@ IMAGE_TOOLS_DEBUG=false
 # Override default STT models per provider (normally set via stt.model in config.yaml)
 # STT_GROQ_MODEL=whisper-large-v3-turbo
 # STT_OPENAI_MODEL=whisper-1
+# STT_ELEVENLABS_MODEL=scribe_v2
 
 # Override STT provider endpoints (for proxies or self-hosted instances)
 # GROQ_BASE_URL=https://api.groq.com/openai/v1
 # STT_OPENAI_BASE_URL=https://api.openai.com/v1
+# ELEVENLABS_STT_BASE_URL=https://api.elevenlabs.io/v1
+
+# =============================================================================
+# MICROSOFT TEAMS INTEGRATION
+# =============================================================================
+# Register a Bot in Azure: https://dev.botframework.com/ → "Register a bot"
+# Or use Azure Portal: Azure Active Directory → App registrations → New registration
+# Then add the bot to Teams via the Bot Framework or App Studio.
+#
+# TEAMS_CLIENT_ID=                     # Azure AD App (client) ID
+# TEAMS_CLIENT_SECRET=                 # Azure AD client secret value
+# TEAMS_TENANT_ID=                     # Azure AD tenant ID (or "common" for multi-tenant)
+# TEAMS_ALLOWED_USERS=                 # Comma-separated AAD object IDs or UPNs
+# TEAMS_ALLOW_ALL_USERS=false          # Set true to skip the allowlist
+# TEAMS_HOME_CHANNEL=                  # Default channel/chat ID for cron delivery
+# TEAMS_HOME_CHANNEL_NAME=             # Display name for the home channel
+# TEAMS_PORT=3978                      # Webhook listen port (Bot Framework default)

.github/actions/nix-setup/action.yml

@@ -1,8 +1,18 @@
 name: 'Setup Nix'
-description: 'Install Nix with DeterminateSystems and enable magic-nix-cache'
+description: 'Install Nix and configure Cachix binary cache'
+
+inputs:
+  cachix-auth-token:
+    description: 'Cachix auth token (enables push). Omit for read-only.'
+    required: false
+    default: ''
 
 runs:
   using: composite
   steps:
     - uses: DeterminateSystems/nix-installer-action@ef8a148080ab6020fd15196c2084a2eea5ff2d25 # v22
-    - uses: DeterminateSystems/magic-nix-cache-action@565684385bcd71bad329742eefe8d12f2e765b39 # v13
+    - uses: cachix/cachix-action@1eb2ef646ac0255473d23a5907ad7b04ce94065c # v17
+      with:
+        name: hermes-agent
+        authToken: ${{ inputs.cachix-auth-token }}
+      continue-on-error: true

.github/dependabot.yml

@@ -0,0 +1,44 @@
+# Dependabot configuration for hermes-agent.
+#
+# Deliberately scoped to github-actions only.
+#
+# We do NOT enable Dependabot for pip / npm / any source-dependency ecosystem
+# because we pin source dependencies exactly (uv.lock, package-lock.json) as
+# part of our supply-chain posture. Automatic version-bump PRs against those
+# pins would undermine the strategy — pins are moved deliberately, after
+# review, not on a schedule.
+#
+# github-actions is the exception: action pins (we use full commit SHAs per
+# supply-chain policy) must be updated when upstream actions publish
+# patches — usually themselves security fixes. Dependabot opens a PR with
+# the new SHA and release notes; we review and merge like any other PR.
+#
+# Security-update PRs for source dependencies (opened ONLY when a CVE is
+# published affecting a currently-pinned version) are enabled separately
+# via the repo's Dependabot security updates setting
+# (Settings → Code security → Dependabot → Dependabot security updates).
+# Those are CVE-only, not schedule-driven, and do not conflict with our
+# pinning strategy — they fire when a pinned version becomes known-bad,
+# which is exactly when we want to move the pin.
+
+version: 2
+updates:
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+      day: "monday"
+    open-pull-requests-limit: 5
+    labels:
+      - "dependencies"
+      - "github-actions"
+    commit-message:
+      prefix: "chore(actions)"
+      include: "scope"
+    groups:
+      # Batch routine action bumps into one PR per week to reduce noise.
+      # Security updates still open individually and bypass grouping.
+      actions-minor-patch:
+        update-types:
+          - "minor"
+          - "patch"

.github/workflows/deploy-site.yml

@@ -53,6 +53,9 @@ jobs:
       - name: Extract skill metadata for dashboard
         run: python3 website/scripts/extract-skills.py
 
+      - name: Regenerate per-skill docs pages + catalogs
+        run: python3 website/scripts/generate-skill-docs.py
+
       - name: Build skills index (if not already present)
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -73,6 +76,16 @@ jobs:
         run: |
           mkdir -p _site/docs
           cp -r website/build/* _site/docs/
+          # llms.txt / llms-full.txt are also published at the site root
+          # (https://hermes-agent.nousresearch.com/llms.txt) because some
+          # agents and IDE plugins probe the classic root-level path rather
+          # than /docs/llms.txt. Same file, two URLs, one source of truth.
+          if [ -f website/build/llms.txt ]; then
+            cp website/build/llms.txt _site/llms.txt
+          fi
+          if [ -f website/build/llms-full.txt ]; then
+            cp website/build/llms-full.txt _site/llms-full.txt
+          fi
 
       - name: Upload artifact
         uses: actions/upload-pages-artifact@56afc609e74202658d3ffba0e8f6dda462b719fa  # v3

.github/workflows/desktop-release.yml

@@ -0,0 +1,341 @@
+name: Desktop Release
+
+on:
+  push:
+    branches: [main]
+  release:
+    types: [published]
+  workflow_dispatch:
+    inputs:
+      channel:
+        description: Release channel to build
+        required: true
+        default: nightly
+        type: choice
+        options:
+          - nightly
+          - stable
+      release_tag:
+        description: "Required when channel=stable (example: v2026.5.5)"
+        required: false
+        type: string
+
+permissions:
+  contents: write
+
+concurrency:
+  group: desktop-release-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  prepare:
+    if: github.repository == 'NousResearch/hermes-agent'
+    runs-on: ubuntu-latest
+    outputs:
+      channel: ${{ steps.meta.outputs.channel }}
+      release_name: ${{ steps.meta.outputs.release_name }}
+      release_tag: ${{ steps.meta.outputs.release_tag }}
+      version: ${{ steps.meta.outputs.version }}
+      is_stable: ${{ steps.meta.outputs.is_stable }}
+    steps:
+      - id: meta
+        env:
+          EVENT_NAME: ${{ github.event_name }}
+          INPUT_CHANNEL: ${{ github.event.inputs.channel }}
+          INPUT_RELEASE_TAG: ${{ github.event.inputs.release_tag }}
+          RELEASE_TAG_FROM_EVENT: ${{ github.event.release.tag_name }}
+          GITHUB_SHA: ${{ github.sha }}
+        run: |
+          set -euo pipefail
+
+          channel="nightly"
+          release_tag="desktop-nightly"
+          is_stable="false"
+
+          if [[ "$EVENT_NAME" == "release" ]]; then
+            channel="stable"
+            release_tag="$RELEASE_TAG_FROM_EVENT"
+            is_stable="true"
+          elif [[ "$EVENT_NAME" == "workflow_dispatch" && "$INPUT_CHANNEL" == "stable" ]]; then
+            channel="stable"
+            release_tag="$INPUT_RELEASE_TAG"
+            is_stable="true"
+          fi
+
+          if [[ "$channel" == "stable" ]]; then
+            if [[ -z "$release_tag" ]]; then
+              echo "Stable desktop releases require a release tag." >&2
+              exit 1
+            fi
+
+            version="${release_tag#v}"
+            release_name="Hermes Desktop ${release_tag}"
+          else
+            stamp="$(date -u +%Y%m%d)"
+            short_sha="${GITHUB_SHA::7}"
+            version="0.0.0-nightly.${stamp}.${short_sha}"
+            release_name="Hermes Desktop Nightly ${stamp}-${short_sha}"
+          fi
+
+          {
+            echo "channel=$channel"
+            echo "release_name=$release_name"
+            echo "release_tag=$release_tag"
+            echo "version=$version"
+            echo "is_stable=$is_stable"
+          } >> "$GITHUB_OUTPUT"
+
+  build:
+    if: github.repository == 'NousResearch/hermes-agent'
+    needs: prepare
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - platform: mac
+            runner: macos-latest
+            build_args: --mac dmg zip
+          - platform: win
+            runner: windows-latest
+            build_args: --win nsis msi
+    runs-on: ${{ matrix.runner }}
+    env:
+      DESKTOP_CHANNEL: ${{ needs.prepare.outputs.channel }}
+      DESKTOP_VERSION: ${{ needs.prepare.outputs.version }}
+      MAC_CSC_LINK: ${{ secrets.CSC_LINK }}
+      MAC_CSC_KEY_PASSWORD: ${{ secrets.CSC_KEY_PASSWORD }}
+      APPLE_API_KEY: ${{ secrets.APPLE_API_KEY }}
+      APPLE_API_KEY_ID: ${{ secrets.APPLE_API_KEY_ID }}
+      APPLE_API_ISSUER: ${{ secrets.APPLE_API_ISSUER }}
+      WIN_CSC_LINK: ${{ secrets.WIN_CSC_LINK }}
+      WIN_CSC_KEY_PASSWORD: ${{ secrets.WIN_CSC_KEY_PASSWORD }}
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020  # v4
+        with:
+          node-version: 20
+          cache: npm
+          cache-dependency-path: package-lock.json
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5
+        with:
+          python-version: "3.11"
+
+      - name: Enforce signing gates for stable releases
+        if: needs.prepare.outputs.is_stable == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+          missing=()
+
+          if [[ "${{ matrix.platform }}" == "mac" ]]; then
+            [[ -z "${MAC_CSC_LINK:-}" ]] && missing+=("CSC_LINK")
+            [[ -z "${MAC_CSC_KEY_PASSWORD:-}" ]] && missing+=("CSC_KEY_PASSWORD")
+            [[ -z "${APPLE_API_KEY:-}" ]] && missing+=("APPLE_API_KEY")
+            [[ -z "${APPLE_API_KEY_ID:-}" ]] && missing+=("APPLE_API_KEY_ID")
+            [[ -z "${APPLE_API_ISSUER:-}" ]] && missing+=("APPLE_API_ISSUER")
+          else
+            [[ -z "${WIN_CSC_LINK:-}" ]] && missing+=("WIN_CSC_LINK")
+            [[ -z "${WIN_CSC_KEY_PASSWORD:-}" ]] && missing+=("WIN_CSC_KEY_PASSWORD")
+          fi
+
+          if (( ${#missing[@]} > 0 )); then
+            echo "::error::Stable desktop release missing required secrets: ${missing[*]}"
+            exit 1
+          fi
+
+      - name: Install workspace dependencies
+        run: npm ci
+
+      - name: Build bundled TUI payload
+        run: npm --prefix ui-tui run build
+
+      - name: Build desktop renderer
+        run: npm --prefix apps/desktop run build
+
+      - name: Stage Hermes payload
+        run: npm --prefix apps/desktop run stage:hermes
+
+      - name: Map macOS signing credentials
+        if: matrix.platform == 'mac'
+        shell: bash
+        run: |
+          set -euo pipefail
+          has_link=0
+          has_pass=0
+          [[ -n "${MAC_CSC_LINK:-}" ]] && has_link=1
+          [[ -n "${MAC_CSC_KEY_PASSWORD:-}" ]] && has_pass=1
+
+          if [[ $has_link -eq 1 && $has_pass -eq 1 ]]; then
+            echo "CSC_LINK=${MAC_CSC_LINK}" >> "$GITHUB_ENV"
+            echo "CSC_KEY_PASSWORD=${MAC_CSC_KEY_PASSWORD}" >> "$GITHUB_ENV"
+          elif [[ $has_link -eq 1 || $has_pass -eq 1 ]]; then
+            echo "::error::macOS signing secrets are partially configured. Set both CSC_LINK and CSC_KEY_PASSWORD."
+            exit 1
+          fi
+
+      - name: Map Windows signing credentials
+        if: matrix.platform == 'win'
+        shell: bash
+        run: |
+          set -euo pipefail
+          has_link=0
+          has_pass=0
+          [[ -n "${WIN_CSC_LINK:-}" ]] && has_link=1
+          [[ -n "${WIN_CSC_KEY_PASSWORD:-}" ]] && has_pass=1
+
+          if [[ $has_link -eq 1 && $has_pass -eq 1 ]]; then
+            echo "CSC_LINK=${WIN_CSC_LINK}" >> "$GITHUB_ENV"
+            echo "CSC_KEY_PASSWORD=${WIN_CSC_KEY_PASSWORD}" >> "$GITHUB_ENV"
+            echo "CSC_FOR_PULL_REQUEST=true" >> "$GITHUB_ENV"
+          elif [[ $has_link -eq 1 || $has_pass -eq 1 ]]; then
+            echo "::error::Windows signing secrets are partially configured. Set both WIN_CSC_LINK and WIN_CSC_KEY_PASSWORD."
+            exit 1
+          fi
+
+      - name: Build desktop installers
+        shell: bash
+        run: |
+          set -euo pipefail
+          npm --prefix apps/desktop exec electron-builder -- \
+            ${{ matrix.build_args }} \
+            --publish never \
+            --config.extraMetadata.version="${DESKTOP_VERSION}" \
+            --config.extraMetadata.desktopChannel="${DESKTOP_CHANNEL}" \
+            '--config.artifactName=Hermes-${version}-${env.DESKTOP_CHANNEL}-${os}-${arch}.${ext}'
+
+      - name: Notarize and staple macOS DMG
+        if: matrix.platform == 'mac' && needs.prepare.outputs.is_stable == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+          dmg_path="$(ls apps/desktop/release/*.dmg | head -n 1)"
+          node apps/desktop/scripts/notarize-artifact.cjs "$dmg_path"
+
+      - name: Validate macOS notarization and Gatekeeper trust
+        if: matrix.platform == 'mac' && needs.prepare.outputs.is_stable == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+          app_path="$(ls -d apps/desktop/release/mac*/Hermes.app | head -n 1)"
+          dmg_path="$(ls apps/desktop/release/*.dmg | head -n 1)"
+          xcrun stapler validate "$app_path"
+          xcrun stapler validate "$dmg_path"
+          spctl --assess --type execute --verbose=4 "$app_path"
+
+      - name: Generate desktop checksums
+        shell: bash
+        run: |
+          set -euo pipefail
+          node <<'EOF'
+          const crypto = require('node:crypto')
+          const fs = require('node:fs')
+          const path = require('node:path')
+
+          const releaseDir = path.resolve('apps/desktop/release')
+          const platform = process.env.PLATFORM
+          const extensions = platform === 'mac' ? ['.dmg', '.zip'] : ['.exe', '.msi']
+          const files = fs
+            .readdirSync(releaseDir)
+            .filter(name => extensions.some(ext => name.endsWith(ext)))
+            .sort()
+
+          if (!files.length) {
+            throw new Error(`No release artifacts were produced for ${platform}`)
+          }
+
+          const lines = files.map(name => {
+            const full = path.join(releaseDir, name)
+            const hash = crypto.createHash('sha256').update(fs.readFileSync(full)).digest('hex')
+            return `${hash}  ${name}`
+          })
+          fs.writeFileSync(path.join(releaseDir, `SHA256SUMS-${platform}.txt`), `${lines.join('\n')}\n`)
+          EOF
+        env:
+          PLATFORM: ${{ matrix.platform }}
+
+      - name: Upload packaged desktop artifacts
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02  # v4
+        with:
+          name: desktop-${{ matrix.platform }}
+          path: |
+            apps/desktop/release/*.dmg
+            apps/desktop/release/*.zip
+            apps/desktop/release/*.exe
+            apps/desktop/release/*.msi
+            apps/desktop/release/SHA256SUMS-${{ matrix.platform }}.txt
+          if-no-files-found: error
+
+  publish:
+    if: github.repository == 'NousResearch/hermes-agent'
+    needs: [prepare, build]
+    runs-on: ubuntu-latest
+    env:
+      GH_TOKEN: ${{ github.token }}
+      CHANNEL: ${{ needs.prepare.outputs.channel }}
+      RELEASE_NAME: ${{ needs.prepare.outputs.release_name }}
+      RELEASE_TAG: ${{ needs.prepare.outputs.release_tag }}
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+        with:
+          fetch-depth: 0
+
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093  # v4
+        with:
+          pattern: desktop-*
+          merge-multiple: true
+          path: dist/desktop
+
+      - name: Publish desktop assets to GitHub release
+        shell: bash
+        run: |
+          set -euo pipefail
+          shopt -s globstar nullglob
+
+          files=(
+            dist/desktop/**/*.dmg
+            dist/desktop/**/*.zip
+            dist/desktop/**/*.exe
+            dist/desktop/**/*.msi
+            dist/desktop/**/SHA256SUMS-*.txt
+          )
+
+          if (( ${#files[@]} == 0 )); then
+            echo "No desktop artifacts were downloaded for publishing." >&2
+            exit 1
+          fi
+
+          if [[ "$CHANNEL" == "nightly" ]]; then
+            git tag -f "$RELEASE_TAG" "$GITHUB_SHA"
+            git push origin "refs/tags/$RELEASE_TAG" --force
+
+            notes="Automated nightly desktop build from main. This prerelease is replaced on each new run."
+
+            if gh release view "$RELEASE_TAG" >/dev/null 2>&1; then
+              while IFS= read -r asset_name; do
+                gh release delete-asset "$RELEASE_TAG" "$asset_name" --yes
+              done < <(gh release view "$RELEASE_TAG" --json assets -q '.assets[].name')
+
+              gh release edit "$RELEASE_TAG" \
+                --title "$RELEASE_NAME" \
+                --prerelease \
+                --notes "$notes"
+            else
+              gh release create "$RELEASE_TAG" \
+                --target "$GITHUB_SHA" \
+                --title "$RELEASE_NAME" \
+                --notes "$notes" \
+                --prerelease
+            fi
+          else
+            if ! gh release view "$RELEASE_TAG" >/dev/null 2>&1; then
+              notes="Automated desktop artifacts attached by desktop-release workflow."
+              gh release create "$RELEASE_TAG" \
+                --target "$GITHUB_SHA" \
+                --title "$RELEASE_NAME" \
+                --notes "$notes"
+            fi
+          fi
+
+          gh release upload "$RELEASE_TAG" "${files[@]}" --clobber

.github/workflows/docs-site-checks.yml

@@ -36,6 +36,9 @@ jobs:
       - name: Extract skill metadata for dashboard
         run: python3 website/scripts/extract-skills.py
 
+      - name: Regenerate per-skill docs pages + catalogs
+        run: python3 website/scripts/generate-skill-docs.py
+
       - name: Lint docs diagrams
         run: npm run lint:diagrams
         working-directory: website

.github/workflows/nix-lockfile-check.yml

@@ -1,68 +0,0 @@
-name: Nix Lockfile Check
-
-on:
-  pull_request:
-  workflow_dispatch:
-
-permissions:
-  contents: read
-  pull-requests: write
-
-concurrency:
-  group: nix-lockfile-check-${{ github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  check:
-    runs-on: ubuntu-latest
-    timeout-minutes: 20
-    steps:
-      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
-
-      - uses: ./.github/actions/nix-setup
-
-      - name: Resolve head SHA
-        id: sha
-        shell: bash
-        run: |
-          FULL="${{ github.event.pull_request.head.sha || github.sha }}"
-          echo "full=$FULL" >> "$GITHUB_OUTPUT"
-          echo "short=${FULL:0:7}" >> "$GITHUB_OUTPUT"
-
-      - name: Check lockfile hashes
-        id: check
-        continue-on-error: true
-        env:
-          LINK_SHA: ${{ steps.sha.outputs.full }}
-        run: nix run .#fix-lockfiles -- --check
-
-      - name: Post sticky PR comment (stale)
-        if: steps.check.outputs.stale == 'true' && github.event_name == 'pull_request'
-        uses: marocchino/sticky-pull-request-comment@52423e01640425a022ef5fd42c6fb5f633a02728  # v2.9.1
-        with:
-          header: nix-lockfile-check
-          message: |
-            ### ⚠️ npm lockfile hash out of date
-
-            Checked against commit [`${{ steps.sha.outputs.short }}`](${{ github.server_url }}/${{ github.repository }}/commit/${{ steps.sha.outputs.full }}) (PR head at check time).
-
-            The `hash = "sha256-..."` line in these nix files no longer matches the committed `package-lock.json`:
-
-            ${{ steps.check.outputs.report }}
-
-            #### Apply the fix
-
-            - [ ] **Apply lockfile fix** — tick to push a commit with the correct hashes to this PR branch
-            - Or [run the Nix Lockfile Fix workflow](${{ github.server_url }}/${{ github.repository }}/actions/workflows/nix-lockfile-fix.yml) manually (pass PR `#${{ github.event.pull_request.number }}`)
-            - Or locally: `nix run .#fix-lockfiles -- --apply` and commit the diff
-
-      - name: Clear sticky PR comment (resolved)
-        if: steps.check.outputs.stale == 'false' && github.event_name == 'pull_request'
-        uses: marocchino/sticky-pull-request-comment@52423e01640425a022ef5fd42c6fb5f633a02728  # v2.9.1
-        with:
-          header: nix-lockfile-check
-          delete: true
-
-      - name: Fail if stale
-        if: steps.check.outputs.stale == 'true'
-        run: exit 1

.github/workflows/nix-lockfile-fix.yml

@@ -1,6 +1,13 @@
 name: Nix Lockfile Fix
 
 on:
+  push:
+    branches: [main]
+    paths:
+      - 'ui-tui/package-lock.json'
+      - 'ui-tui/package.json'
+      - 'apps/dashboard/package-lock.json'
+      - 'apps/dashboard/package.json'
   workflow_dispatch:
     inputs:
       pr_number:
@@ -19,9 +26,105 @@ concurrency:
   cancel-in-progress: false
 
 jobs:
+  # ── Auto-fix on main ───────────────────────────────────────────────
+  # Fires when a push to main touches package.json or package-lock.json
+  # in ui-tui/ or apps/dashboard/. Runs fix-lockfiles and pushes the hash
+  # update commit directly to main so Nix builds never stay broken.
+  #
+  # Safety invariants:
+  #   1. The fix commit only touches nix/*.nix files, which are NOT in
+  #      the paths filter above, so this cannot re-trigger itself.
+  #   2. An explicit file-whitelist check before commit aborts if
+  #      fix-lockfiles ever modifies unexpected files.
+  #   3. Job-level concurrency with cancel-in-progress: true ensures
+  #      back-to-back pushes collapse to the newest; ref: main checkout
+  #      always operates on the latest branch state.
+  #   4. Uses a GitHub App token (not GITHUB_TOKEN) so the fix commit
+  #      triggers downstream nix.yml verification.
+  auto-fix-main:
+    if: github.event_name == 'push'
+    runs-on: ubuntu-latest
+    timeout-minutes: 25
+    concurrency:
+      group: auto-fix-main
+      cancel-in-progress: true
+    steps:
+      - name: Generate GitHub App token
+        id: app-token
+        uses: actions/create-github-app-token@7bfa3a4717ef143a604ee0a99d859b8886a96d00  # v1.9.3
+        with:
+          app-id: ${{ secrets.APP_ID }}
+          private-key: ${{ secrets.APP_PRIVATE_KEY }}
+
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+        with:
+          ref: main
+          token: ${{ steps.app-token.outputs.token }}
+
+      - uses: ./.github/actions/nix-setup
+        with:
+          cachix-auth-token: ${{ secrets.CACHIX_AUTH_TOKEN }}
+
+      - name: Apply lockfile hashes
+        id: apply
+        run: nix run .#fix-lockfiles -- --apply
+
+      - name: Commit & push
+        if: steps.apply.outputs.changed == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          # Ensure only nix files were modified — prevents accidental
+          # self-triggering if fix-lockfiles ever touches package files.
+          unexpected="$(git diff --name-only | grep -Ev '^nix/(tui|web)\.nix$' || true)"
+          if [ -n "$unexpected" ]; then
+            echo "::error::Unexpected modified files: $unexpected"
+            exit 1
+          fi
+
+          # Record the base SHA before committing — used to detect package
+          # file changes if we need to rebase after a non-fast-forward push.
+          BASE_SHA="$(git rev-parse HEAD)"
+
+          git config user.name 'github-actions[bot]'
+          git config user.email '41898282+github-actions[bot]@users.noreply.github.com'
+          git add nix/tui.nix nix/web.nix
+          git commit -m "fix(nix): auto-refresh npm lockfile hashes" \
+            -m "Source: $GITHUB_SHA" \
+            -m "Run: $GITHUB_SERVER_URL/$GITHUB_REPOSITORY/actions/runs/$GITHUB_RUN_ID"
+
+          # Retry push with rebase in case main advanced with an unrelated
+          # commit during the nix build. Without this, a non-fast-forward
+          # rejection silently loses the fix. If package files changed during
+          # the rebase, abort — a fresh auto-fix run will handle the new state.
+          for attempt in 1 2 3; do
+            if git push origin HEAD:main; then
+              exit 0
+            fi
+            echo "::warning::Push attempt $attempt failed (non-fast-forward?), rebasing…"
+            git fetch origin main
+
+            # If package files changed between our base and the new main,
+            # our computed hashes are stale. Abort and let the next triggered
+            # run recompute from the correct package-lock state.
+            pkg_changed="$(git diff --name-only "$BASE_SHA"..origin/main -- \
+              'ui-tui/package-lock.json' 'ui-tui/package.json' \
+              'apps/dashboard/package-lock.json' 'apps/dashboard/package.json' || true)"
+            if [ -n "$pkg_changed" ]; then
+              echo "::warning::Package files changed since hash computation — aborting; a fresh run will recompute"
+              exit 0
+            fi
+
+            git rebase origin/main
+          done
+          echo "::error::Failed to push after 3 rebase attempts"
+          exit 1
+
+  # ── PR fix (manual / checkbox) ─────────────────────────────────────
+  # Existing behavior: run on manual dispatch OR when a task-list
+  # checkbox in the sticky lockfile-check comment flips from [ ] to [x].
   fix:
-    # Run on manual dispatch OR when a task-list checkbox in the sticky
-    # lockfile-check comment flips from `[ ]` to `[x]`.
     if: |
       github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'issue_comment'
@@ -99,10 +202,12 @@ jobs:
           fetch-depth: 0
 
       - uses: ./.github/actions/nix-setup
+        with:
+          cachix-auth-token: ${{ secrets.CACHIX_AUTH_TOKEN }}
 
       - name: Apply lockfile hashes
         id: apply
-        run: nix run .#fix-lockfiles -- --apply
+        run: nix run .#fix-lockfiles
 
       - name: Commit & push
         if: steps.apply.outputs.changed == 'true'

.github/workflows/nix.yml

@@ -7,6 +7,7 @@ on:
 
 permissions:
   contents: read
+  pull-requests: write
 
 concurrency:
   group: nix-${{ github.ref }}
@@ -22,12 +23,95 @@ jobs:
     steps:
       - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
       - uses: ./.github/actions/nix-setup
+        with:
+          cachix-auth-token: ${{ secrets.CACHIX_AUTH_TOKEN }}
+
+      - name: Resolve head SHA
+        if: github.event_name == 'pull_request'
+        id: sha
+        shell: bash
+        run: |
+          FULL="${{ github.event.pull_request.head.sha || github.sha }}"
+          echo "full=$FULL" >> "$GITHUB_OUTPUT"
+          echo "short=${FULL:0:7}" >> "$GITHUB_OUTPUT"
+
       - name: Check flake
+        id: flake
         if: runner.os == 'Linux'
+        continue-on-error: true
         run: nix flake check --print-build-logs
+
       - name: Build package
+        id: build
         if: runner.os == 'Linux'
+        continue-on-error: true
         run: nix build --print-build-logs
+
+      # When the real Nix build fails, run a targeted diagnostic to see if
+      # the failure is specifically a stale npm lockfile hash in one of the
+      # known npm subpackages (tui / web).  This avoids surfacing a generic
+      # "build failed" message when the fix is a single known command.
+      - name: Diagnose npm lockfile hashes
+        id: hash_check
+        if: (steps.flake.outcome == 'failure' || steps.build.outcome == 'failure') && runner.os == 'Linux'
+        continue-on-error: true
+        env:
+          LINK_SHA: ${{ steps.sha.outputs.full }}
+        run: nix run .#fix-lockfiles -- --check
+
+      # If fix-lockfiles itself crashes (infrastructure blip, cache throttle,
+      # etc.) it won't set stale=true/false.  Treat that as a distinct failure
+      # mode rather than silently ignoring it.
+      - name: Fail if hash check crashed without reporting
+        if: steps.hash_check.outcome == 'failure' && steps.hash_check.outputs.stale != 'true' && steps.hash_check.outputs.stale != 'false'
+        run: |
+          echo "::error::fix-lockfiles exited without reporting stale status — likely an infrastructure or script failure"
+          exit 1
+
+      - name: Post sticky PR comment (stale hashes)
+        if: steps.hash_check.outputs.stale == 'true' && github.event_name == 'pull_request'
+        uses: marocchino/sticky-pull-request-comment@52423e01640425a022ef5fd42c6fb5f633a02728  # v2.9.1
+        with:
+          header: nix-lockfile-check
+          message: |
+            ### ⚠️ npm lockfile hash out of date
+
+            Checked against commit [`${{ steps.sha.outputs.short }}`](${{ github.server_url }}/${{ github.repository }}/commit/${{ steps.sha.outputs.full }}) (PR head at check time).
+
+            The `hash = "sha256-..."` line in these nix files no longer matches the committed `package-lock.json`:
+
+            ${{ steps.hash_check.outputs.report }}
+
+            #### Apply the fix
+
+            - [ ] **Apply lockfile fix** — tick to push a commit with the correct hashes to this PR branch
+            - Or [run the Nix Lockfile Fix workflow](${{ github.server_url }}/${{ github.repository }}/actions/workflows/nix-lockfile-fix.yml) manually (pass PR `#${{ github.event.pull_request.number }}`)
+            - Or locally: `nix run .#fix-lockfiles` and commit the diff
+
+      # Clear the sticky comment when either the build passed outright (no
+      # hash check needed) or the hash check explicitly returned stale=false
+      # (build failed for a non-hash reason).
+      - name: Clear sticky PR comment (resolved)
+        if: |
+          github.event_name == 'pull_request' &&
+          runner.os == 'Linux' &&
+          (steps.hash_check.outputs.stale == 'false' ||
+           (steps.flake.outcome == 'success' && steps.build.outcome == 'success'))
+        uses: marocchino/sticky-pull-request-comment@52423e01640425a022ef5fd42c6fb5f633a02728  # v2.9.1
+        with:
+          header: nix-lockfile-check
+          delete: true
+
+      - name: Final fail if build or flake failed
+        if: steps.flake.outcome == 'failure' || steps.build.outcome == 'failure'
+        run: |
+          if [ "${{ steps.hash_check.outputs.stale }}" == "true" ]; then
+            echo "::error::Nix build failed due to stale npm lockfile hash. Run: nix run .#fix-lockfiles"
+          else
+            echo "::error::Nix build/flake check failed. See logs above."
+          fi
+          exit 1
+
       - name: Evaluate flake (macOS)
         if: runner.os == 'macOS'
         run: nix flake show --json > /dev/null

.github/workflows/osv-scanner.yml

@@ -0,0 +1,67 @@
+name: OSV-Scanner
+
+# Scans lockfiles (uv.lock, package-lock.json) against the OSV vulnerability
+# database. Runs on every PR that touches a lockfile and on a weekly schedule
+# against main.
+#
+# This is detection-only — OSV-Scanner does NOT open PRs or modify pins.
+# It reports known CVEs in currently-pinned dependency versions so we can
+# decide when and how to patch on our own schedule. Our pinning strategy
+# (full SHA / exact version) is preserved; only the notification signal
+# is added.
+#
+# Complements the existing supply-chain-audit.yml workflow (which scans
+# for malicious code patterns in PR diffs) by covering the orthogonal
+# "currently-pinned dep became known-vulnerable" case.
+#
+# Uses Google's officially-recommended reusable workflow, pinned by SHA.
+# Findings land in the repo's Security tab (Code Scanning > OSV-Scanner).
+# fail-on-vuln is disabled so the job does not block merges on pre-existing
+# vulnerabilities in pinned deps that we may need to patch deliberately.
+
+on:
+  pull_request:
+    branches: [main]
+    paths:
+      - 'uv.lock'
+      - 'pyproject.toml'
+      - 'package.json'
+      - 'package-lock.json'
+      - 'ui-tui/package.json'
+      - 'ui-tui/package-lock.json'
+      - 'website/package.json'
+      - 'website/package-lock.json'
+      - '.github/workflows/osv-scanner.yml'
+  push:
+    branches: [main]
+    paths:
+      - 'uv.lock'
+      - 'pyproject.toml'
+      - 'package.json'
+      - 'package-lock.json'
+      - 'ui-tui/package-lock.json'
+      - 'website/package-lock.json'
+  schedule:
+    # Weekly scan against main — catches CVEs published after merge for
+    # deps that haven't changed since.
+    - cron: '0 9 * * 1'
+  workflow_dispatch:
+
+permissions:
+  # Required by the reusable workflow to upload SARIF to the Security tab.
+  actions: read
+  contents: read
+  security-events: write
+
+jobs:
+  scan:
+    name: Scan lockfiles
+    uses: google/osv-scanner-action/.github/workflows/osv-scanner-reusable.yml@c51854704019a247608d928f370c98740469d4b5  # v2.3.5
+    with:
+      # Scan explicit lockfiles rather than recursing, so we only look at
+      # the three sources of truth and skip vendored / test / worktree dirs.
+      scan-args: |-
+        --lockfile=uv.lock
+        --lockfile=ui-tui/package-lock.json
+        --lockfile=website/package-lock.json
+      fail-on-vuln: false

.gitignore

@@ -54,6 +54,10 @@ environments/benchmarks/evals/
 
 # Web UI build output
 hermes_cli/web_dist/
+apps/desktop/build/
+apps/desktop/dist/
+apps/desktop/release/
+apps/desktop/*.tsbuildinfo
 
 # Web UI assets — synced from @nous-research/ui at build time via
 # `npm run sync-assets` (see web/package.json).
@@ -69,3 +73,13 @@ mini-swe-agent/
 .nix-stamps/
 result
 website/static/api/skills-index.json
+models-dev-upstream/
+
+# Local editor / agent tooling (machine-specific; keep in global config, not the repo)
+.codex/
+.cursor/
+.gemini/
+.zed/
+.mcp.json
+opencode.json
+config/mcporter.json

AGENTS.md

@@ -2,81 +2,94 @@
 
 Instructions for AI coding assistants and developers working on the hermes-agent codebase.
 
+**Never give up on the right solution.**
+
 ## Development Environment
 
 ```bash
-source venv/bin/activate  # ALWAYS activate before running Python
+# Prefer .venv; fall back to venv if that's what your checkout has.
+source .venv/bin/activate   # or: source venv/bin/activate
 ```
 
+`scripts/run_tests.sh` probes `.venv` first, then `venv`, then
+`$HOME/.hermes/hermes-agent/venv` (for worktrees that share a venv with the
+main checkout).
+
 ## Project Structure
 
+File counts shift constantly — don't treat the tree below as exhaustive.
+The canonical source is the filesystem. The notes call out the load-bearing
+entry points you'll actually edit.
+
 ```
 hermes-agent/
-├── run_agent.py          # AIAgent class — core conversation loop
+├── run_agent.py          # AIAgent class — core conversation loop (~12k LOC)
 ├── model_tools.py        # Tool orchestration, discover_builtin_tools(), handle_function_call()
 ├── toolsets.py           # Toolset definitions, _HERMES_CORE_TOOLS list
-├── cli.py                # HermesCLI class — interactive CLI orchestrator
+├── cli.py                # HermesCLI class — interactive CLI orchestrator (~11k LOC)
 ├── hermes_state.py       # SessionDB — SQLite session store (FTS5 search)
-├── agent/                # Agent internals
-│   ├── prompt_builder.py     # System prompt assembly
-│   ├── context_compressor.py # Auto context compression
-│   ├── prompt_caching.py     # Anthropic prompt caching
-│   ├── auxiliary_client.py   # Auxiliary LLM client (vision, summarization)
-│   ├── model_metadata.py     # Model context lengths, token estimation
-│   ├── models_dev.py         # models.dev registry integration (provider-aware context)
-│   ├── display.py            # KawaiiSpinner, tool preview formatting
-│   ├── skill_commands.py     # Skill slash commands (shared CLI/gateway)
-│   └── trajectory.py         # Trajectory saving helpers
-├── hermes_cli/           # CLI subcommands and setup
-│   ├── main.py           # Entry point — all `hermes` subcommands
-│   ├── config.py         # DEFAULT_CONFIG, OPTIONAL_ENV_VARS, migration
-│   ├── commands.py       # Slash command definitions + SlashCommandCompleter
-│   ├── callbacks.py      # Terminal callbacks (clarify, sudo, approval)
-│   ├── setup.py          # Interactive setup wizard
-│   ├── skin_engine.py    # Skin/theme engine — CLI visual customization
-│   ├── skills_config.py  # `hermes skills` — enable/disable skills per platform
-│   ├── tools_config.py   # `hermes tools` — enable/disable tools per platform
-│   ├── skills_hub.py     # `/skills` slash command (search, browse, install)
-│   ├── models.py         # Model catalog, provider model lists
-│   ├── model_switch.py   # Shared /model switch pipeline (CLI + gateway)
-│   └── auth.py           # Provider credential resolution
-├── tools/                # Tool implementations (one file per tool)
-│   ├── registry.py       # Central tool registry (schemas, handlers, dispatch)
-│   ├── approval.py       # Dangerous command detection
-│   ├── terminal_tool.py  # Terminal orchestration
-│   ├── process_registry.py # Background process management
-│   ├── file_tools.py     # File read/write/search/patch
-│   ├── web_tools.py      # Web search/extract (Parallel + Firecrawl)
-│   ├── browser_tool.py   # Browserbase browser automation
-│   ├── code_execution_tool.py # execute_code sandbox
-│   ├── delegate_tool.py  # Subagent delegation
-│   ├── mcp_tool.py       # MCP client (~1050 lines)
+├── hermes_constants.py   # get_hermes_home(), display_hermes_home() — profile-aware paths
+├── hermes_logging.py     # setup_logging() — agent.log / errors.log / gateway.log (profile-aware)
+├── batch_runner.py       # Parallel batch processing
+├── agent/                # Agent internals (provider adapters, memory, caching, compression, etc.)
+├── hermes_cli/           # CLI subcommands, setup wizard, plugins loader, skin engine
+├── tools/                # Tool implementations — auto-discovered via tools/registry.py
 │   └── environments/     # Terminal backends (local, docker, ssh, modal, daytona, singularity)
-├── gateway/              # Messaging platform gateway
-│   ├── run.py            # Main loop, slash commands, message dispatch
-│   ├── session.py        # SessionStore — conversation persistence
-│   └── platforms/        # Adapters: telegram, discord, slack, whatsapp, homeassistant, signal, qqbot
+├── gateway/              # Messaging gateway — run.py + session.py + platforms/
+│   ├── platforms/        # Adapter per platform (telegram, discord, slack, whatsapp,
+│   │                     #   homeassistant, signal, matrix, mattermost, email, sms,
+│   │                     #   dingtalk, wecom, weixin, feishu, qqbot, bluebubbles,
+│   │                     #   yuanbao, webhook, api_server, ...). See ADDING_A_PLATFORM.md.
+│   └── builtin_hooks/    # Extension point for always-registered gateway hooks (none shipped)
+├── plugins/              # Plugin system (see "Plugins" section below)
+│   ├── memory/           # Memory-provider plugins (honcho, mem0, supermemory, ...)
+│   ├── context_engine/   # Context-engine plugins
+│   ├── kanban/           # Multi-agent board dispatcher + worker plugin
+│   ├── hermes-achievements/  # Gamified achievement tracking
+│   ├── observability/    # Metrics / traces / logs plugin
+│   ├── image_gen/        # Image-generation providers
+│   └── <others>/         # disk-cleanup, example-dashboard, google_meet, platforms,
+│                         #   spotify, strike-freedom-cockpit, ...
+├── optional-skills/      # Heavier/niche skills shipped but NOT active by default
+├── skills/               # Built-in skills bundled with the repo
 ├── ui-tui/               # Ink (React) terminal UI — `hermes --tui`
-│   ├── src/entry.tsx        # TTY gate + render()
-│   ├── src/app.tsx          # Main state machine and UI
-│   ├── src/gatewayClient.ts # Child process + JSON-RPC bridge
-│   ├── src/app/             # Decomposed app logic (event handler, slash handler, stores, hooks)
-│   ├── src/components/      # Ink components (branding, markdown, prompts, pickers, etc.)
-│   ├── src/hooks/           # useCompletion, useInputHistory, useQueue, useVirtualHistory
-│   └── src/lib/             # Pure helpers (history, osc52, text, rpc, messages)
+│   └── src/              # entry.tsx, app.tsx, gatewayClient.ts + app/components/hooks/lib
 ├── tui_gateway/          # Python JSON-RPC backend for the TUI
-│   ├── entry.py             # stdio entrypoint
-│   ├── server.py            # RPC handlers and session logic
-│   ├── render.py            # Optional rich/ANSI bridge
-│   └── slash_worker.py      # Persistent HermesCLI subprocess for slash commands
-├── hermes_agent/acp/     # ACP server (VS Code / Zed / JetBrains integration)
-├── cron/                 # Scheduler (jobs.py, scheduler.py)
+├── acp_adapter/          # ACP server (VS Code / Zed / JetBrains integration)
+├── cron/                 # Scheduler — jobs.py, scheduler.py
 ├── environments/         # RL training environments (Atropos)
-├── tests/                # Pytest suite (~3000 tests)
-└── batch_runner.py       # Parallel batch processing
+├── scripts/              # run_tests.sh, release.py, auxiliary scripts
+├── website/              # Docusaurus docs site
+└── tests/                # Pytest suite (~17k tests across ~900 files as of May 2026)
 ```
 
-**User config:** `~/.hermes/config.yaml` (settings), `~/.hermes/.env` (API keys)
+**User config:** `~/.hermes/config.yaml` (settings), `~/.hermes/.env` (API keys only).
+**Logs:** `~/.hermes/logs/` — `agent.log` (INFO+), `errors.log` (WARNING+),
+`gateway.log` when running the gateway. Profile-aware via `get_hermes_home()`.
+Browse with `hermes logs [--follow] [--level ...] [--session ...]`.
+
+## TypeScript Style
+
+Applies to TypeScript across Hermes: desktop, TUI, website, and future TS packages.
+
+- Prefer small nanostores over component state when state is shared, reused, or read by distant UI.
+- Let each feature own its atoms. Chat state belongs near chat, shell state near shell, shared state in `src/store`.
+- Components that render from an atom should use `useStore`. Non-rendering actions should read with `$atom.get()`.
+- Do not pass state through three components when the leaf can subscribe to the atom.
+- Keep persistence beside the atom that owns it.
+- Keep route roots thin. They compose routes and shell; they should not become controllers.
+- No monolithic hooks. A hook should own one narrow job.
+- Prefer colocated action modules over hidden god hooks.
+- If a callback is pure side effect, use the terse void form:
+  `onState={st => void setGatewayState(st)}`.
+- Async UI handlers should make intent explicit:
+  `onClick={() => void save()}`.
+- Prefer interfaces for public props and shared object shapes. Avoid `type X = { ... }` for object props.
+- Extend React primitives for props: `React.ComponentProps<'button'>`, `React.ComponentProps<typeof Dialog>`, `Omit<...>`, `Pick<...>`.
+- Table-driven beats condition ladders when mapping ids, routes, or views.
+- `src/app` owns routes, pages, and page-specific components.
+- `src/store` owns shared atoms.
+- `src/lib` owns shared pure helpers.
 
 ## File Dependency Chain
 
@@ -94,20 +107,30 @@ run_agent.py, cli.py, batch_runner.py, environments/
 
 ## AIAgent Class (run_agent.py)
 
+The real `AIAgent.__init__` takes ~60 parameters (credentials, routing, callbacks,
+session context, budget, credential pool, etc.). The signature below is the
+minimum subset you'll usually touch — read `run_agent.py` for the full list.
+
 ```python
 class AIAgent:
     def __init__(self,
-        model: str = "anthropic/claude-opus-4.6",
-        max_iterations: int = 90,
+        base_url: str = None,
+        api_key: str = None,
+        provider: str = None,
+        api_mode: str = None,              # "chat_completions" | "codex_responses" | ...
+        model: str = "",                   # empty → resolved from config/provider later
+        max_iterations: int = 90,          # tool-calling iterations (shared with subagents)
         enabled_toolsets: list = None,
         disabled_toolsets: list = None,
         quiet_mode: bool = False,
         save_trajectories: bool = False,
-        platform: str = None,           # "cli", "telegram", etc.
+        platform: str = None,              # "cli", "telegram", etc.
         session_id: str = None,
         skip_context_files: bool = False,
         skip_memory: bool = False,
-        # ... plus provider, api_mode, callbacks, routing params
+        credential_pool=None,
+        # ... plus callbacks, thread/user/chat IDs, iteration_budget, fallback_model,
+        # checkpoints config, prefill_messages, service_tier, reasoning_config, etc.
     ): ...
 
     def chat(self, message: str) -> str:
@@ -120,10 +143,13 @@ class AIAgent:
 
 ### Agent Loop
 
-The core loop is inside `run_conversation()` — entirely synchronous:
+The core loop is inside `run_conversation()` — entirely synchronous, with
+interrupt checks, budget tracking, and a one-turn grace call:
 
 ```python
-while api_call_count < self.max_iterations and self.iteration_budget.remaining > 0:
+while (api_call_count < self.max_iterations and self.iteration_budget.remaining > 0) \
+        or self._budget_grace_call:
+    if self._interrupt_requested: break
     response = client.chat.completions.create(model=model, messages=messages, tools=tool_schemas)
     if response.tool_calls:
         for tool_call in response.tool_calls:
@@ -134,7 +160,8 @@ while api_call_count < self.max_iterations and self.iteration_budget.remaining >
         return response.content
 ```
 
-Messages follow OpenAI format: `{"role": "system/user/assistant/tool", ...}`. Reasoning content is stored in `assistant_msg["reasoning"]`.
+Messages follow OpenAI format: `{"role": "system/user/assistant/tool", ...}`.
+Reasoning content is stored in `assistant_msg["reasoning"]`.
 
 ---
 
@@ -243,11 +270,33 @@ npm run fmt       # prettier
 npm test          # vitest
 ```
 
+### TUI in the Dashboard (`hermes dashboard` → `/chat`)
+
+The dashboard embeds the real `hermes --tui` — **not** a rewrite.  See `hermes_cli/pty_bridge.py` + the `@app.websocket("/api/pty")` endpoint in `hermes_cli/web_server.py`.
+
+- Browser loads `apps/dashboard/src/pages/ChatPage.tsx`, which mounts xterm.js's `Terminal` with the WebGL renderer, `@xterm/addon-fit` for container-driven resize, and `@xterm/addon-unicode11` for modern wide-character widths.
+- `/api/pty?token=…` upgrades to a WebSocket; auth uses the same ephemeral `_SESSION_TOKEN` as REST, via query param (browsers can't set `Authorization` on WS upgrade).
+- The server spawns whatever `hermes --tui` would spawn, through `ptyprocess` (POSIX PTY — WSL works, native Windows does not).
+- Frames: raw PTY bytes each direction; resize via `\x1b[RESIZE:<cols>;<rows>]` intercepted on the server and applied with `TIOCSWINSZ`.
+
+**Do not re-implement the primary chat experience in React.** The main transcript, composer/input flow (including slash-command behavior), and PTY-backed terminal belong to the embedded `hermes --tui` — anything new you add to Ink shows up in the dashboard automatically. If you find yourself rebuilding the transcript or composer for the dashboard, stop and extend Ink instead.
+
+**Structured React UI around the TUI is allowed when it is not a second chat surface.** Sidebar widgets, inspectors, summaries, status panels, and similar supporting views (e.g. `ChatSidebar`, `ModelPickerDialog`, `ToolCall`) are fine when they complement the embedded TUI rather than replacing the transcript / composer / terminal. Keep their state independent of the PTY child's session and surface their failures non-destructively so the terminal pane keeps working unimpaired.
+
 ---
 
 ## Adding New Tools
 
-Requires changes in **2 files**:
+For most custom or local-only tools, do **not** edit Hermes core. Use the plugin
+route instead: create `~/.hermes/plugins/<name>/plugin.yaml` and
+`~/.hermes/plugins/<name>/__init__.py`, then register tools with
+`ctx.register_tool(...)`. Plugin toolsets are discovered automatically and can be
+enabled or disabled without touching `tools/` or `toolsets.py`.
+
+Use the built-in route below only when the user is explicitly contributing a new
+core Hermes tool that should ship in the base system.
+
+Built-in/core tools require changes in **2 files**:
 
 **1. Create `tools/your_tool.py`:**
 ```python
@@ -270,9 +319,9 @@ registry.register(
 )
 ```
 
-**2. Add to `toolsets.py`** — either `_HERMES_CORE_TOOLS` (all platforms) or a new toolset.
+**2. Add to `toolsets.py`** — either `_HERMES_CORE_TOOLS` (all platforms) or a new toolset. **This step is required:** auto-discovery imports the tool and registers its schema, but the tool is only *exposed to an agent* if its name appears in a toolset. `_HERMES_CORE_TOOLS` is not dead code — it's the default bundle every platform's base toolset inherits from.
 
-Auto-discovery: any `tools/*.py` file with a top-level `registry.register()` call is imported automatically — no manual import list to maintain.
+Auto-discovery: any `tools/*.py` file with a top-level `registry.register()` call is imported automatically — no manual import list to maintain. Wiring into a toolset is still a deliberate, manual step.
 
 The registry handles schema collection, dispatch, availability checking, and error wrapping. All handlers MUST return a JSON string.
 
@@ -280,7 +329,7 @@ The registry handles schema collection, dispatch, availability checking, and err
 
 **State files**: If a tool stores persistent state (caches, logs, checkpoints), use `get_hermes_home()` for the base directory — never `Path.home() / ".hermes"`. This ensures each profile gets its own state.
 
-**Agent-level tools** (todo, memory): intercepted by `run_agent.py` before `handle_function_call()`. See `todo_tool.py` for the pattern.
+**Agent-level tools** (todo, memory): intercepted by `run_agent.py` before `handle_function_call()`. See `tools/todo_tool.py` for the pattern.
 
 ---
 
@@ -288,9 +337,29 @@ The registry handles schema collection, dispatch, availability checking, and err
 
 ### config.yaml options:
 1. Add to `DEFAULT_CONFIG` in `hermes_cli/config.py`
-2. Bump `_config_version` (currently 5) to trigger migration for existing users
+2. Bump `_config_version` (check the current value at the top of `DEFAULT_CONFIG`)
+   ONLY if you need to actively migrate/transform existing user config
+   (renaming keys, changing structure). Adding a new key to an existing
+   section is handled automatically by the deep-merge and does NOT require
+   a version bump.
 
-### .env variables:
+### Top-level `config.yaml` sections (non-exhaustive):
+
+`model`, `agent`, `terminal`, `compression`, `display`, `stt`, `tts`,
+`memory`, `security`, `delegation`, `smart_model_routing`, `checkpoints`,
+`auxiliary`, `curator`, `skills`, `gateway`, `logging`, `cron`, `profiles`,
+`plugins`, `honcho`.
+
+`auxiliary` holds per-task overrides for side-LLM work (curator, vision,
+embedding, title generation, session_search, etc.) — each task can pin
+its own provider/model/base_url/max_tokens/reasoning_effort. See
+`agent/auxiliary_client.py::_resolve_auto` for resolution order.
+
+`curator` holds the background skill-maintenance config —
+`enabled`, `interval_hours`, `min_idle_hours`, `stale_after_days`,
+`archive_after_days`, `backup` (nested).
+
+### .env variables (SECRETS ONLY — API keys, tokens, passwords):
 1. Add to `OPTIONAL_ENV_VARS` in `hermes_cli/config.py` with metadata:
 ```python
 "NEW_API_KEY": {
@@ -302,13 +371,29 @@ The registry handles schema collection, dispatch, availability checking, and err
 },
 ```
 
-### Config loaders (two separate systems):
+Non-secret settings (timeouts, thresholds, feature flags, paths, display
+preferences) belong in `config.yaml`, not `.env`. If internal code needs an
+env var mirror for backward compatibility, bridge it from `config.yaml` to
+the env var in code (see `gateway_timeout`, `terminal.cwd` → `TERMINAL_CWD`).
+
+### Config loaders (three paths — know which one you're in):
 
 | Loader | Used by | Location |
 |--------|---------|----------|
-| `load_cli_config()` | CLI mode | `cli.py` |
-| `load_config()` | `hermes tools`, `hermes setup` | `hermes_cli/config.py` |
-| Direct YAML load | Gateway | `gateway/run.py` |
+| `load_cli_config()` | CLI mode | `cli.py` — merges CLI-specific defaults + user YAML |
+| `load_config()` | `hermes tools`, `hermes setup`, most CLI subcommands | `hermes_cli/config.py` — merges `DEFAULT_CONFIG` + user YAML |
+| Direct YAML load | Gateway runtime | `gateway/run.py` + `gateway/config.py` — reads user YAML raw |
+
+If you add a new key and the CLI sees it but the gateway doesn't (or vice
+versa), you're on the wrong loader. Check `DEFAULT_CONFIG` coverage.
+
+### Working directory:
+- **CLI** — uses the process's current directory (`os.getcwd()`).
+- **Messaging** — uses `terminal.cwd` from `config.yaml`. The gateway bridges this
+  to the `TERMINAL_CWD` env var for child tools. **`MESSAGING_CWD` has been
+  removed** — the config loader prints a deprecation warning if it's set in
+  `.env`. Same for `TERMINAL_CWD` in `.env`; the canonical setting is
+  `terminal.cwd` in `config.yaml`.
 
 ---
 
@@ -401,7 +486,260 @@ Activate with `/skin cyberpunk` or `display.skin: cyberpunk` in config.yaml.
 
 ---
 
+## Plugins
+
+Hermes has two plugin surfaces. Both live under `plugins/` in the repo so
+repo-shipped plugins can be discovered alongside user-installed ones in
+`~/.hermes/plugins/` and pip-installed entry points.
+
+### General plugins (`hermes_cli/plugins.py` + `plugins/<name>/`)
+
+`PluginManager` discovers plugins from `~/.hermes/plugins/`, `./.hermes/plugins/`,
+and pip entry points. Each plugin exposes a `register(ctx)` function that
+can:
+
+- Register Python-callback lifecycle hooks:
+  `pre_tool_call`, `post_tool_call`, `pre_llm_call`, `post_llm_call`,
+  `on_session_start`, `on_session_end`
+- Register new tools via `ctx.register_tool(...)`
+- Register CLI subcommands via `ctx.register_cli_command(...)` — the
+  plugin's argparse tree is wired into `hermes` at startup so
+  `hermes <pluginname> <subcmd>` works with no change to `main.py`
+
+Hooks are invoked from `model_tools.py` (pre/post tool) and `run_agent.py`
+(lifecycle). **Discovery timing pitfall:** `discover_plugins()` only runs
+as a side effect of importing `model_tools.py`. Code paths that read plugin
+state without importing `model_tools.py` first must call `discover_plugins()`
+explicitly (it's idempotent).
+
+### Memory-provider plugins (`plugins/memory/<name>/`)
+
+Separate discovery system for pluggable memory backends. Current built-in
+providers include **honcho, mem0, supermemory, byterover, hindsight,
+holographic, openviking, retaindb**.
+
+Each provider implements the `MemoryProvider` ABC (see `agent/memory_provider.py`)
+and is orchestrated by `agent/memory_manager.py`. Lifecycle hooks include
+`sync_turn(turn_messages)`, `prefetch(query)`, `shutdown()`, and optional
+`post_setup(hermes_home, config)` for setup-wizard integration.
+
+**CLI commands via `plugins/memory/<name>/cli.py`:** if a memory plugin
+defines `register_cli(subparser)`, `discover_plugin_cli_commands()` finds
+it at argparse setup time and wires it into `hermes <plugin>`. The
+framework only exposes CLI commands for the **currently active** memory
+provider (read from `memory.provider` in config.yaml), so disabled
+providers don't clutter `hermes --help`.
+
+**Rule (Teknium, May 2026):** plugins MUST NOT modify core files
+(`run_agent.py`, `cli.py`, `gateway/run.py`, `hermes_cli/main.py`, etc.).
+If a plugin needs a capability the framework doesn't expose, expand the
+generic plugin surface (new hook, new ctx method) — never hardcode
+plugin-specific logic into core. PR #5295 removed 95 lines of hardcoded
+honcho argparse from `main.py` for exactly this reason.
+
+### Dashboard / context-engine / image-gen plugin directories
+
+`plugins/context_engine/`, `plugins/image_gen/`, `plugins/example-dashboard/`,
+etc. follow the same pattern (ABC + orchestrator + per-plugin directory).
+Context engines plug into `agent/context_engine.py`; image-gen providers
+into `agent/image_gen_provider.py`.
+
+---
+
+## Skills
+
+Two parallel surfaces:
+
+- **`skills/`** — built-in skills shipped and loadable by default.
+  Organized by category directories (e.g. `skills/github/`, `skills/mlops/`).
+- **`optional-skills/`** — heavier or niche skills shipped with the repo but
+  NOT active by default. Installed explicitly via
+  `hermes skills install official/<category>/<skill>`. Adapter lives in
+  `tools/skills_hub.py` (`OptionalSkillSource`). Categories include
+  `autonomous-ai-agents`, `blockchain`, `communication`, `creative`,
+  `devops`, `email`, `health`, `mcp`, `migration`, `mlops`, `productivity`,
+  `research`, `security`, `web-development`.
+
+When reviewing skill PRs, check which directory they target — heavy-dep or
+niche skills belong in `optional-skills/`.
+
+### SKILL.md frontmatter
+
+Standard fields: `name`, `description`, `version`, `author`, `license`,
+`platforms` (OS-gating list: `[macos]`, `[linux, macos]`, ...),
+`metadata.hermes.tags`, `metadata.hermes.category`,
+`metadata.hermes.related_skills`, `metadata.hermes.config` (config.yaml
+settings the skill needs — stored under `skills.config.<key>`, prompted
+during setup, injected at load time).
+
+Top-level `tags:` and `category:` are also accepted and mirrored from
+`metadata.hermes.*` by the loader.
+
+---
+
+## Toolsets
+
+All toolsets are defined in `toolsets.py` as a single `TOOLSETS` dict.
+Each platform's adapter picks a base toolset (e.g. Telegram uses
+`"messaging"`); `_HERMES_CORE_TOOLS` is the default bundle most
+platforms inherit from.
+
+Current toolset keys: `browser`, `clarify`, `code_execution`, `cronjob`,
+`debugging`, `delegation`, `discord`, `discord_admin`, `feishu_doc`,
+`feishu_drive`, `file`, `homeassistant`, `image_gen`, `kanban`, `memory`,
+`messaging`, `moa`, `rl`, `safe`, `search`, `session_search`, `skills`,
+`spotify`, `terminal`, `todo`, `tts`, `video`, `vision`, `web`, `yuanbao`.
+
+Enable/disable per platform via `hermes tools` (the curses UI) or the
+`tools.<platform>.enabled` / `tools.<platform>.disabled` lists in
+`config.yaml`.
+
+---
+
+## Delegation (`delegate_task`)
+
+`tools/delegate_tool.py` spawns a subagent with an isolated
+context + terminal session. Synchronous: the parent waits for the
+child's summary before continuing its own loop — if the parent is
+interrupted, the child is cancelled.
+
+Two shapes:
+
+- **Single:** pass `goal` (+ optional `context`, `toolsets`).
+- **Batch (parallel):** pass `tasks: [...]` — each gets its own subagent
+  running concurrently. Concurrency is capped by
+  `delegation.max_concurrent_children` (default 3).
+
+Roles:
+
+- `role="leaf"` (default) — focused worker. Cannot call `delegate_task`,
+  `clarify`, `memory`, `send_message`, `execute_code`.
+- `role="orchestrator"` — retains `delegate_task` so it can spawn its
+  own workers. Gated by `delegation.orchestrator_enabled` (default true)
+  and bounded by `delegation.max_spawn_depth` (default 2).
+
+Key config knobs (under `delegation:` in `config.yaml`):
+`max_concurrent_children`, `max_spawn_depth`, `child_timeout_seconds`,
+`orchestrator_enabled`, `subagent_auto_approve`, `inherit_mcp_toolsets`,
+`max_iterations`.
+
+Synchronicity rule: delegate_task is **not** durable. For long-running
+work that must outlive the current turn, use `cronjob` or
+`terminal(background=True, notify_on_complete=True)` instead.
+
+---
+
+## Curator (skill lifecycle)
+
+Background skill-maintenance system that tracks usage on agent-created
+skills and auto-archives stale ones. Users never lose skills; archives
+go to `~/.hermes/skills/.archive/` and are restorable.
+
+- **Core:** `agent/curator.py` (review loop, auto-transitions, LLM review
+  prompt) + `agent/curator_backup.py` (pre-run tar.gz snapshots).
+- **CLI:** `hermes_cli/curator.py` wires `hermes curator <verb>` where
+  verbs are: `status`, `run`, `pause`, `resume`, `pin`, `unpin`,
+  `archive`, `restore`, `prune`, `backup`, `rollback`.
+- **Telemetry:** `tools/skill_usage.py` owns the sidecar
+  `~/.hermes/skills/.usage.json` — per-skill `use_count`, `view_count`,
+  `patch_count`, `last_activity_at`, `state` (active / stale /
+  archived), `pinned`.
+
+Invariants:
+- Curator only touches skills with `created_by: "agent"` provenance —
+  bundled + hub-installed skills are off-limits.
+- Never deletes; max destructive action is archive.
+- Pinned skills are exempt from every auto-transition and from the
+  LLM review pass.
+- `skill_manage(action="delete")` refuses pinned skills; patch/edit/
+  write_file/remove_file go through so the agent can keep improving
+  pinned skills.
+
+Config section (`curator:` in `config.yaml`):
+`enabled`, `interval_hours`, `min_idle_hours`, `stale_after_days`,
+`archive_after_days`, `backup.*`.
+
+Full user-facing docs: `website/docs/user-guide/features/curator.md`.
+
+---
+
+## Cron (scheduled jobs)
+
+`cron/jobs.py` (job store) + `cron/scheduler.py` (tick loop). Agents
+schedule jobs via the `cronjob` tool; users via `hermes cron <verb>`
+(`list`, `add`, `edit`, `pause`, `resume`, `run`, `remove`) or the
+`/cron` slash command.
+
+Supported schedule formats:
+- Duration: `"30m"`, `"2h"`, `"1d"`
+- "every" phrase: `"every 2h"`, `"every monday 9am"`
+- 5-field cron expression: `"0 9 * * *"`
+- ISO timestamp (one-shot): `"2026-06-01T09:00:00Z"`
+
+Per-job fields include `skills` (load specific skills), `model` /
+`provider` overrides, `script` (pre-run data-collection script whose
+stdout is injected into the prompt; `no_agent=True` turns the script
+into the entire job), `context_from` (chain job A's last output into
+job B's prompt), `workdir` (run in a specific directory with its
+`AGENTS.md`/`CLAUDE.md` loaded), and multi-platform delivery.
+
+Hardening invariants:
+- **3-minute hard interrupt** on cron sessions — runaway agent loops
+  cannot monopolize the scheduler.
+- Catchup window: half the job's period, clamped to 120s–2h.
+- Grace window: 120s for one-shot jobs whose fire time was missed.
+- File lock at `~/.hermes/cron/.tick.lock` prevents duplicate ticks
+  across processes.
+- Cron sessions pass `skip_memory=True` by default; memory providers
+  intentionally do not run during cron.
+
+Cron deliveries are **not** mirrored into the target gateway session —
+they land in their own cron session with a header/footer frame so the
+main conversation's message-role alternation stays intact.
+
+---
+
+## Kanban (multi-agent work queue)
+
+Durable SQLite-backed board that lets multiple profiles / workers
+collaborate on shared tasks. Users drive it via `hermes kanban <verb>`;
+workers spawned by the dispatcher drive it via a dedicated `kanban_*`
+toolset so their schema footprint is zero when they're not inside a
+kanban task.
+
+- **CLI:** `hermes_cli/kanban.py` wires `hermes kanban` with verbs
+  `init`, `create`, `list` (alias `ls`), `show`, `assign`, `link`,
+  `unlink`, `comment`, `complete`, `block`, `unblock`, `archive`,
+  `tail`, plus less-commonly-used `watch`, `stats`, `runs`, `log`,
+  `assignees`, `heartbeat`, `notify-*`, `dispatch`, `daemon`, `gc`.
+- **Worker toolset:** `tools/kanban_tools.py` exposes `kanban_show`,
+  `kanban_complete`, `kanban_block`, `kanban_heartbeat`, `kanban_comment`,
+  `kanban_create`, `kanban_link` — gated by `HERMES_KANBAN_TASK` so
+  the schema only appears for processes actually running as a worker.
+- **Dispatcher:** long-lived loop that (default every 60s) reclaims
+  stale claims, promotes ready tasks, atomically claims, and spawns
+  assigned profiles. Runs **inside the gateway** by default via
+  `kanban.dispatch_in_gateway: true`.
+- **Plugin assets:** `plugins/kanban/dashboard/` (web UI) +
+  `plugins/kanban/systemd/` (`hermes-kanban-dispatcher.service` for
+  standalone dispatcher deployment).
+
+Isolation model:
+- **Board** is the hard boundary — workers are spawned with
+  `HERMES_KANBAN_BOARD` pinned in their env so they can't see other
+  boards.
+- **Tenant** is a soft namespace *within* a board — one specialist
+  fleet can serve multiple businesses with workspace-path + memory-key
+  isolation.
+- After ~5 consecutive spawn failures on the same task the dispatcher
+  auto-blocks it to prevent spin loops.
+
+Full user-facing docs: `website/docs/user-guide/features/kanban.md`.
+
+---
+
 ## Important Policies
+
 ### Prompt Caching Must Not Break
 
 Hermes-Agent ensures caching remains valid throughout a conversation. **Do NOT implement changes that would:**
@@ -411,9 +749,10 @@ Hermes-Agent ensures caching remains valid throughout a conversation. **Do NOT i
 
 Cache-breaking forces dramatically higher costs. The ONLY time we alter context is during context compression.
 
-### Working Directory Behavior
-- **CLI**: Uses current directory (`.` → `os.getcwd()`)
-- **Messaging**: Uses `MESSAGING_CWD` env var (default: home directory)
+Slash commands that mutate system-prompt state (skills, tools, memory, etc.)
+must be **cache-aware**: default to deferred invalidation (change takes
+effect next session), with an opt-in `--now` flag for immediate
+invalidation. See `/skills install --now` for the canonical pattern.
 
 ### Background Process Notifications (Gateway)
 
@@ -435,7 +774,7 @@ Hermes supports **profiles** — multiple fully isolated instances, each with it
 `HERMES_HOME` directory (config, API keys, memory, sessions, skills, gateway, etc.).
 
 The core mechanism: `_apply_profile_override()` in `hermes_cli/main.py` sets
-`HERMES_HOME` before any module imports. All 119+ references to `get_hermes_home()`
+`HERMES_HOME` before any module imports. All `get_hermes_home()` references
 automatically scope to the active profile.
 
 ### Rules for profile-safe code
@@ -492,8 +831,12 @@ Use `get_hermes_home()` from `hermes_constants` for code paths. Use `display_her
 for user-facing print/log messages. Hardcoding `~/.hermes` breaks profiles — each profile
 has its own `HERMES_HOME` directory. This was the source of 5 bugs fixed in PR #3575.
 
-### DO NOT use `simple_term_menu` for interactive menus
-Rendering bugs in tmux/iTerm2 — ghosting on scroll. Use `curses` (stdlib) instead. See `hermes_cli/tools_config.py` for the pattern.
+### DO NOT introduce new `simple_term_menu` usage
+Existing call sites in `hermes_cli/main.py` remain for legacy fallback only;
+the preferred UI is curses (stdlib) because `simple_term_menu` has
+ghost-duplication rendering bugs in tmux/iTerm2 with arrow keys. New
+interactive menus must use `hermes_cli/curses_ui.py` — see
+`hermes_cli/tools_config.py` for the canonical pattern.
 
 ### DO NOT use `\033[K` (ANSI erase-to-EOL) in spinner/display code
 Leaks as literal `?[K` text under `prompt_toolkit`'s `patch_stdout`. Use space-padding: `f"\r{line}{' ' * pad}"`.
@@ -504,6 +847,30 @@ Leaks as literal `?[K` text under `prompt_toolkit`'s `patch_stdout`. Use space-p
 ### DO NOT hardcode cross-tool references in schema descriptions
 Tool schema descriptions must not mention tools from other toolsets by name (e.g., `browser_navigate` saying "prefer web_search"). Those tools may be unavailable (missing API keys, disabled toolset), causing the model to hallucinate calls to non-existent tools. If a cross-reference is needed, add it dynamically in `get_tool_definitions()` in `model_tools.py` — see the `browser_navigate` / `execute_code` post-processing blocks for the pattern.
 
+### The gateway has TWO message guards — both must bypass approval/control commands
+When an agent is running, messages pass through two sequential guards:
+(1) **base adapter** (`gateway/platforms/base.py`) queues messages in
+`_pending_messages` when `session_key in self._active_sessions`, and
+(2) **gateway runner** (`gateway/run.py`) intercepts `/stop`, `/new`,
+`/queue`, `/status`, `/approve`, `/deny` before they reach
+`running_agent.interrupt()`. Any new command that must reach the runner
+while the agent is blocked (e.g. approval prompts) MUST bypass BOTH
+guards and be dispatched inline, not via `_process_message_background()`
+(which races session lifecycle).
+
+### Squash merges from stale branches silently revert recent fixes
+Before squash-merging a PR, ensure the branch is up to date with `main`
+(`git fetch origin main && git reset --hard origin/main` in the worktree,
+then re-apply the PR's commits). A stale branch's version of an unrelated
+file will silently overwrite recent fixes on main when squashed. Verify
+with `git diff HEAD~1..HEAD` after merging — unexpected deletions are a
+red flag.
+
+### Don't wire in dead code without E2E validation
+Unused code that was never shipped was dead for a reason. Before wiring an
+unused module into a live code path, E2E test the real resolution chain
+with actual imports (not mocks) against a temp `HERMES_HOME`.
+
 ### Tests must not write to `~/.hermes/`
 The `_isolate_hermes_home` autouse fixture in `tests/conftest.py` redirects `HERMES_HOME` to a temp dir. Never hardcode `~/.hermes/` paths in tests.
 
@@ -559,7 +926,7 @@ If you can't use the wrapper (e.g. on Windows or inside an IDE that shells
 pytest directly), at minimum activate the venv and pass `-n 4`:
 
 ```bash
-source venv/bin/activate
+source .venv/bin/activate   # or: source venv/bin/activate
 python -m pytest tests/ -q -n 4
 ```
 

CONTRIBUTING.md

@@ -9,7 +9,7 @@ Thank you for contributing to Hermes Agent! This guide covers everything you nee
 We value contributions in this order:
 
 1. **Bug fixes** — crashes, incorrect behavior, data loss. Always top priority.
-2. **Cross-platform compatibility** — Windows, macOS, different Linux distros, different terminal emulators. We want Hermes to work everywhere.
+2. **Cross-platform compatibility** — macOS, different Linux distros, and WSL2 on Windows. We want Hermes to work everywhere.
 3. **Security hardening** — shell injection, prompt injection, path traversal, privilege escalation. See [Security](#security-considerations).
 4. **Performance and robustness** — retry logic, error handling, graceful degradation.
 5. **New skills** — but only broadly useful ones. See [Should it be a Skill or a Tool?](#should-it-be-a-skill-or-a-tool)
@@ -494,7 +494,7 @@ branding:
   agent_name: "My Agent"
   welcome: "Welcome message"
   response_label: " ⚔ Agent "
-  prompt_symbol: "⚔ ❯ "
+  prompt_symbol: "⚔"
 
 tool_prefix: "╎"             # Tool output line prefix
 ```
@@ -515,7 +515,7 @@ See `hermes_cli/skin_engine.py` for the full schema and existing skins as exampl
 
 ## Cross-Platform Compatibility
 
-Hermes runs on Linux, macOS, and Windows. When writing code that touches the OS:
+Hermes runs on Linux, macOS, and WSL2 on Windows. When writing code that touches the OS:
 
 ### Critical rules
 
@@ -597,7 +597,7 @@ refactor/description   # Code restructuring
 
 1. **Run tests**: `pytest tests/ -v`
 2. **Test manually**: Run `hermes` and exercise the code path you changed
-3. **Check cross-platform impact**: If you touch file I/O, process management, or terminal handling, consider Windows and macOS
+3. **Check cross-platform impact**: If you touch file I/O, process management, or terminal handling, consider macOS, Linux, and WSL2
 4. **Keep PRs focused**: One logical change per PR. Don't mix a bug fix with a refactor with a new feature.
 
 ### PR description

Dockerfile

@@ -10,9 +10,11 @@ ENV PYTHONUNBUFFERED=1
 ENV PLAYWRIGHT_BROWSERS_PATH=/opt/hermes/.playwright
 
 # Install system dependencies in one layer, clear APT cache
+# tini reaps orphaned zombie processes (MCP stdio subprocesses, git, bun, etc.)
+# that would otherwise accumulate when hermes runs as PID 1. See #15012.
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        build-essential nodejs npm python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli && \
+    build-essential curl nodejs npm python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli tini && \
     rm -rf /var/lib/apt/lists/*
 
 # Non-root user for runtime; UID can be overridden via HERMES_UID at runtime
@@ -26,24 +28,50 @@ WORKDIR /opt/hermes
 # ---------- Layer-cached dependency install ----------
 # Copy only package manifests first so npm install + Playwright are cached
 # unless the lockfiles themselves change.
+#
+# ui-tui/packages/hermes-ink/ is copied IN FULL (not just its manifests)
+# because it is referenced as a `file:` workspace dependency from
+# ui-tui/package.json.  Copying the tree up front lets npm resolve the
+# workspace to real content instead of stopping at a bare package.json.
 COPY package.json package-lock.json ./
 COPY web/package.json web/package-lock.json web/
+COPY ui-tui/package.json ui-tui/package-lock.json ui-tui/
+COPY ui-tui/packages/hermes-ink/ ui-tui/packages/hermes-ink/
+
+# `npm_config_install_links=false` forces npm to install `file:` deps as
+# symlinks (the npm 10+ default) even on Debian's older bundled npm 9.x,
+# which defaults to `install-links=true` and installs file deps as *copies*.
+# The host-side package-lock.json is generated with a newer npm that uses
+# symlinks, so an install-as-copy produces a hidden node_modules/.package-lock.json
+# that permanently disagrees with the root lock on the @hermes/ink entry.
+# That disagreement trips the TUI launcher's `_tui_need_npm_install()`
+# check on every startup and triggers a runtime `npm install` that then
+# fails with EACCES (node_modules/ is root-owned from build time).
+ENV npm_config_install_links=false
 
 RUN npm install --prefer-offline --no-audit && \
     npx playwright install --with-deps chromium --only-shell && \
     (cd web && npm install --prefer-offline --no-audit) && \
+    (cd ui-tui && npm install --prefer-offline --no-audit) && \
     npm cache clean --force
 
 # ---------- Source code ----------
 # .dockerignore excludes node_modules, so the installs above survive.
 COPY --chown=hermes:hermes . .
 
-# Build web dashboard (Vite outputs to hermes_cli/web_dist/)
-RUN cd web && npm run build
+# Build browser dashboard and terminal UI assets.
+RUN cd web && npm run build && \
+    cd ../ui-tui && npm run build
+
+# ---------- Permissions ----------
+# Make install dir world-readable so any HERMES_UID can read it at runtime.
+# The venv needs to be traversable too.
+USER root
+RUN chmod -R a+rX /opt/hermes
+# Start as root so the entrypoint can usermod/groupmod + gosu.
+# If HERMES_UID is unset, the entrypoint drops to the default hermes user (10000).
 
 # ---------- Python virtualenv ----------
-RUN chown hermes:hermes /opt/hermes
-USER hermes
 RUN uv venv && \
     uv pip install --no-cache-dir -e ".[all]"
 
@@ -52,4 +80,4 @@ ENV HERMES_WEB_DIST=/opt/hermes/hermes_cli/web_dist
 ENV HERMES_HOME=/opt/data
 ENV PATH="/opt/data/.local/bin:${PATH}"
 VOLUME [ "/opt/data" ]
-ENTRYPOINT [ "/opt/hermes/docker/entrypoint.sh" ]
+ENTRYPOINT [ "/usr/bin/tini", "-g", "--", "/opt/hermes/docker/entrypoint.sh" ]

README.md

@@ -76,7 +76,7 @@ Hermes has two entry points: start the terminal UI with `hermes`, or run the gat
 | Set a personality | `/personality [name]` | `/personality [name]` |
 | Retry or undo the last turn | `/retry`, `/undo` | `/retry`, `/undo` |
 | Compress context / check usage | `/compress`, `/usage`, `/insights [--days N]` | `/compress`, `/usage`, `/insights [days]` |
-| Browse skills | `/skills` or `/<skill-name>` | `/skills` or `/<skill-name>` |
+| Browse skills | `/skills` or `/<skill-name>` | `/<skill-name>` |
 | Interrupt current work | `Ctrl+C` or send a new message | `/stop` or send a new message |
 | Platform-specific status | `/platforms` | `/status`, `/sethome` |
 
@@ -157,14 +157,10 @@ curl -LsSf https://astral.sh/uv/install.sh | sh
 uv venv venv --python 3.11
 source venv/bin/activate
 uv pip install -e ".[all,dev]"
-python -m pytest tests/ -q
+scripts/run_tests.sh
 ```
 
-> **RL Training (optional):** To work on the RL/Tinker-Atropos integration:
-> ```bash
-> git submodule update --init tinker-atropos
-> uv pip install -e "./tinker-atropos"
-> ```
+> **RL Training (optional):** The RL/Atropos integration (`environments/`) ships via the `atroposlib` and `tinker` dependencies pulled in by `.[all,dev]` — no submodule setup required.
 
 ---
 

RELEASE_v0.11.0.md

@@ -0,0 +1,453 @@
+# Hermes Agent v0.11.0 (v2026.4.23)
+
+**Release Date:** April 23, 2026
+**Since v0.9.0:** 1,556 commits · 761 merged PRs · 1,314 files changed · 224,174 insertions · 29 community contributors (290 including co-authors)
+
+> The Interface release — a full React/Ink rewrite of the interactive CLI, a pluggable transport architecture underneath every provider, native AWS Bedrock support, five new inference paths, a 17th messaging platform (QQBot), a dramatically expanded plugin surface, and GPT-5.5 via Codex OAuth.
+
+This release also folds in all the highlights deferred from v0.10.0 (which shipped only the Nous Tool Gateway) — so it covers roughly two weeks of work across the whole stack.
+
+---
+
+## ✨ Highlights
+
+- **New Ink-based TUI** — `hermes --tui` is now a full React/Ink rewrite of the interactive CLI, with a Python JSON-RPC backend (`tui_gateway`). Sticky composer, live streaming with OSC-52 clipboard support, stable picker keys, status bar with per-turn stopwatch and git branch, `/clear` confirm, light-theme preset, and a subagent spawn observability overlay. ~310 commits to `ui-tui/` + `tui_gateway/`. (@OutThisLife + Teknium)
+
+- **Transport ABC + Native AWS Bedrock** — Format conversion and HTTP transport were extracted from `run_agent.py` into a pluggable `agent/transports/` layer. `AnthropicTransport`, `ChatCompletionsTransport`, `ResponsesApiTransport`, and `BedrockTransport` each own their own format conversion and API shape. Native AWS Bedrock support via the Converse API ships on top of the new abstraction. ([#10549](https://github.com/NousResearch/hermes-agent/pull/10549), [#13347](https://github.com/NousResearch/hermes-agent/pull/13347), [#13366](https://github.com/NousResearch/hermes-agent/pull/13366), [#13430](https://github.com/NousResearch/hermes-agent/pull/13430), [#13805](https://github.com/NousResearch/hermes-agent/pull/13805), [#13814](https://github.com/NousResearch/hermes-agent/pull/13814) — @kshitijk4poor + Teknium)
+
+- **Five new inference paths** — Native NVIDIA NIM ([#11774](https://github.com/NousResearch/hermes-agent/pull/11774)), Arcee AI ([#9276](https://github.com/NousResearch/hermes-agent/pull/9276)), Step Plan ([#13893](https://github.com/NousResearch/hermes-agent/pull/13893)), Google Gemini CLI OAuth ([#11270](https://github.com/NousResearch/hermes-agent/pull/11270)), and Vercel ai-gateway with pricing + dynamic discovery ([#13223](https://github.com/NousResearch/hermes-agent/pull/13223) — @jerilynzheng). Plus Gemini routed through the native AI Studio API for better performance ([#12674](https://github.com/NousResearch/hermes-agent/pull/12674)).
+
+- **GPT-5.5 over Codex OAuth** — OpenAI's new GPT-5.5 reasoning model is now available through your ChatGPT Codex OAuth, with live model discovery wired into the model picker so new OpenAI releases show up without catalog updates. ([#14720](https://github.com/NousResearch/hermes-agent/pull/14720))
+
+- **QQBot — 17th supported platform** — Native QQBot adapter via QQ Official API v2, with QR scan-to-configure setup wizard, streaming cursor, emoji reactions, and DM/group policy gating that matches WeCom/Weixin parity. ([#9364](https://github.com/NousResearch/hermes-agent/pull/9364), [#11831](https://github.com/NousResearch/hermes-agent/pull/11831))
+
+- **Plugin surface expanded** — Plugins can now register slash commands (`register_command`), dispatch tools directly (`dispatch_tool`), block tool execution from hooks (`pre_tool_call` can veto), rewrite tool results (`transform_tool_result`), transform terminal output (`transform_terminal_output`), ship image_gen backends, and add custom dashboard tabs. The bundled disk-cleanup plugin is opt-in by default as a reference implementation. ([#9377](https://github.com/NousResearch/hermes-agent/pull/9377), [#10626](https://github.com/NousResearch/hermes-agent/pull/10626), [#10763](https://github.com/NousResearch/hermes-agent/pull/10763), [#10951](https://github.com/NousResearch/hermes-agent/pull/10951), [#12929](https://github.com/NousResearch/hermes-agent/pull/12929), [#12944](https://github.com/NousResearch/hermes-agent/pull/12944), [#12972](https://github.com/NousResearch/hermes-agent/pull/12972), [#13799](https://github.com/NousResearch/hermes-agent/pull/13799), [#14175](https://github.com/NousResearch/hermes-agent/pull/14175))
+
+- **`/steer` — mid-run agent nudges** — `/steer <prompt>` injects a note that the running agent sees after its next tool call, without interrupting the turn or breaking prompt cache. For when you want to course-correct an agent in-flight. ([#12116](https://github.com/NousResearch/hermes-agent/pull/12116))
+
+- **Shell hooks** — Wire any shell script as a Hermes lifecycle hook (pre_tool_call, post_tool_call, on_session_start, etc.) without writing a Python plugin. ([#13296](https://github.com/NousResearch/hermes-agent/pull/13296))
+
+- **Webhook direct-delivery mode** — Webhook subscriptions can now forward payloads straight to a platform chat without going through the agent — zero-LLM push notifications for alerting, uptime checks, and event streams. ([#12473](https://github.com/NousResearch/hermes-agent/pull/12473))
+
+- **Smarter delegation** — Subagents now have an explicit `orchestrator` role that can spawn their own workers, with configurable `max_spawn_depth` (default flat). Concurrent sibling subagents share filesystem state through a file-coordination layer so they don't clobber each other's edits. ([#13691](https://github.com/NousResearch/hermes-agent/pull/13691), [#13718](https://github.com/NousResearch/hermes-agent/pull/13718))
+
+- **Auxiliary models — configurable UI + main-model-first** — `hermes model` has a dedicated "Configure auxiliary models" screen for per-task overrides (compression, vision, session_search, title_generation). `auto` routing now defaults to the main model for side tasks across all users (previously aggregator users were silently routed to a cheap provider-side default). ([#11891](https://github.com/NousResearch/hermes-agent/pull/11891), [#11900](https://github.com/NousResearch/hermes-agent/pull/11900))
+
+- **Dashboard plugin system + live theme switching** — The web dashboard is now extensible. Third-party plugins can add custom tabs, widgets, and views without forking. Paired with a live-switching theme system — themes now control colors, fonts, layout, and density — so users can hot-swap the dashboard look without a reload. Same theming discipline the CLI has, now on the web. ([#10951](https://github.com/NousResearch/hermes-agent/pull/10951), [#10687](https://github.com/NousResearch/hermes-agent/pull/10687), [#14725](https://github.com/NousResearch/hermes-agent/pull/14725))
+
+- **Dashboard polish** — i18n (English + Chinese), react-router sidebar layout, mobile-responsive, Vercel deployment, real per-session API call tracking, and one-click update + gateway restart buttons. ([#9228](https://github.com/NousResearch/hermes-agent/pull/9228), [#9370](https://github.com/NousResearch/hermes-agent/pull/9370), [#9453](https://github.com/NousResearch/hermes-agent/pull/9453), [#10686](https://github.com/NousResearch/hermes-agent/pull/10686), [#13526](https://github.com/NousResearch/hermes-agent/pull/13526), [#14004](https://github.com/NousResearch/hermes-agent/pull/14004) — @austinpickett + @DeployFaith + Teknium)
+
+---
+
+## 🏗️ Core Agent & Architecture
+
+### Transport Layer (NEW)
+- **Transport ABC** abstracts format conversion and HTTP transport from `run_agent.py` into `agent/transports/` ([#13347](https://github.com/NousResearch/hermes-agent/pull/13347))
+- **AnthropicTransport** — Anthropic Messages API path ([#13366](https://github.com/NousResearch/hermes-agent/pull/13366), @kshitijk4poor)
+- **ChatCompletionsTransport** — default path for OpenAI-compatible providers ([#13805](https://github.com/NousResearch/hermes-agent/pull/13805))
+- **ResponsesApiTransport** — OpenAI Responses API + Codex build_kwargs wiring ([#13430](https://github.com/NousResearch/hermes-agent/pull/13430), @kshitijk4poor)
+- **BedrockTransport** — AWS Bedrock Converse API transport ([#13814](https://github.com/NousResearch/hermes-agent/pull/13814))
+
+### Provider & Model Support
+- **Native AWS Bedrock provider** via Converse API ([#10549](https://github.com/NousResearch/hermes-agent/pull/10549))
+- **NVIDIA NIM native provider** (salvage of #11703) ([#11774](https://github.com/NousResearch/hermes-agent/pull/11774))
+- **Arcee AI direct provider** ([#9276](https://github.com/NousResearch/hermes-agent/pull/9276))
+- **Step Plan provider** (salvage #6005) ([#13893](https://github.com/NousResearch/hermes-agent/pull/13893), @kshitijk4poor)
+- **Google Gemini CLI OAuth** inference provider ([#11270](https://github.com/NousResearch/hermes-agent/pull/11270))
+- **Vercel ai-gateway** with pricing, attribution, and dynamic discovery ([#13223](https://github.com/NousResearch/hermes-agent/pull/13223), @jerilynzheng)
+- **GPT-5.5 over Codex OAuth** with live model discovery in the picker ([#14720](https://github.com/NousResearch/hermes-agent/pull/14720))
+- **Gemini routed through native AI Studio API** ([#12674](https://github.com/NousResearch/hermes-agent/pull/12674))
+- **xAI Grok upgraded to Responses API** ([#10783](https://github.com/NousResearch/hermes-agent/pull/10783))
+- **Ollama improvements** — Cloud provider support, GLM continuation, `think=false` control, surrogate sanitization, `/v1` hint ([#10782](https://github.com/NousResearch/hermes-agent/pull/10782))
+- **Kimi K2.6** across OpenRouter, Nous Portal, native Kimi, and HuggingFace ([#13148](https://github.com/NousResearch/hermes-agent/pull/13148), [#13152](https://github.com/NousResearch/hermes-agent/pull/13152), [#13169](https://github.com/NousResearch/hermes-agent/pull/13169))
+- **Kimi K2.5** promoted to first position in all model suggestion lists ([#11745](https://github.com/NousResearch/hermes-agent/pull/11745), @kshitijk4poor)
+- **Xiaomi MiMo v2.5-pro + v2.5** on OpenRouter, Nous Portal, and native ([#14184](https://github.com/NousResearch/hermes-agent/pull/14184), [#14635](https://github.com/NousResearch/hermes-agent/pull/14635), @kshitijk4poor)
+- **GLM-5V-Turbo** for coding plan ([#9907](https://github.com/NousResearch/hermes-agent/pull/9907))
+- **Claude Opus 4.7** in Nous Portal catalog ([#11398](https://github.com/NousResearch/hermes-agent/pull/11398))
+- **OpenRouter elephant-alpha** in curated lists ([#9378](https://github.com/NousResearch/hermes-agent/pull/9378))
+- **OpenCode-Go** — Kimi K2.6 and Qwen3.5/3.6 Plus in curated catalog ([#13429](https://github.com/NousResearch/hermes-agent/pull/13429))
+- **minimax/minimax-m2.5:free** in OpenRouter catalog ([#13836](https://github.com/NousResearch/hermes-agent/pull/13836))
+- **`/model` merges models.dev entries** for lesser-loved providers ([#14221](https://github.com/NousResearch/hermes-agent/pull/14221))
+- **Per-provider + per-model `request_timeout_seconds`** config ([#12652](https://github.com/NousResearch/hermes-agent/pull/12652))
+- **Configurable API retry count** via `agent.api_max_retries` ([#14730](https://github.com/NousResearch/hermes-agent/pull/14730))
+- **ctx_size context length key** for Lemonade server (salvage #8536) ([#14215](https://github.com/NousResearch/hermes-agent/pull/14215))
+- **Custom provider display name prompt** ([#9420](https://github.com/NousResearch/hermes-agent/pull/9420))
+- **Recommendation badges** on tool provider selection ([#9929](https://github.com/NousResearch/hermes-agent/pull/9929))
+- Fix: correct GPT-5 family context lengths in fallback defaults ([#9309](https://github.com/NousResearch/hermes-agent/pull/9309))
+- Fix: clamp `minimal` reasoning effort to `low` on Responses API ([#9429](https://github.com/NousResearch/hermes-agent/pull/9429))
+- Fix: strip reasoning item IDs from Responses API input when `store=False` ([#10217](https://github.com/NousResearch/hermes-agent/pull/10217))
+- Fix: OpenViking correct account default + commit session on `/new` and compress ([#10463](https://github.com/NousResearch/hermes-agent/pull/10463))
+- Fix: Kimi `/coding` thinking block survival + empty reasoning_content + block ordering (multiple PRs)
+- Fix: don't send Anthropic thinking to api.kimi.com/coding ([#13826](https://github.com/NousResearch/hermes-agent/pull/13826))
+- Fix: send `max_tokens`, `reasoning_effort`, and `thinking` for Kimi/Moonshot
+- Fix: stream reasoning content through OpenAI-compatible providers that emit it
+
+### Agent Loop & Conversation
+- **`/steer <prompt>`** — mid-run agent nudges after next tool call ([#12116](https://github.com/NousResearch/hermes-agent/pull/12116))
+- **Orchestrator role + configurable spawn depth** for `delegate_task` (default flat) ([#13691](https://github.com/NousResearch/hermes-agent/pull/13691))
+- **Cross-agent file state coordination** for concurrent subagents ([#13718](https://github.com/NousResearch/hermes-agent/pull/13718))
+- **Compressor smart collapse, dedup, anti-thrashing**, template upgrade, hardening ([#10088](https://github.com/NousResearch/hermes-agent/pull/10088))
+- **Compression summaries respect the conversation's language** ([#12556](https://github.com/NousResearch/hermes-agent/pull/12556))
+- **Compression model falls back to main model** on permanent 503/404 ([#10093](https://github.com/NousResearch/hermes-agent/pull/10093))
+- **Auto-continue interrupted agent work** after gateway restart ([#9934](https://github.com/NousResearch/hermes-agent/pull/9934))
+- **Activity heartbeats** prevent false gateway inactivity timeouts ([#10501](https://github.com/NousResearch/hermes-agent/pull/10501))
+- **Auxiliary models UI** — dedicated screen for per-task overrides ([#11891](https://github.com/NousResearch/hermes-agent/pull/11891))
+- **Auxiliary auto routing defaults to main model** for all users ([#11900](https://github.com/NousResearch/hermes-agent/pull/11900))
+- **PLATFORM_HINTS for Matrix, Mattermost, Feishu** ([#14428](https://github.com/NousResearch/hermes-agent/pull/14428), @alt-glitch)
+- Fix: reset retry counters after compression; stop poisoning conversation history ([#10055](https://github.com/NousResearch/hermes-agent/pull/10055))
+- Fix: break compression-exhaustion infinite loop and auto-reset session ([#10063](https://github.com/NousResearch/hermes-agent/pull/10063))
+- Fix: stale agent timeout, uv venv detection, empty response after tools ([#10065](https://github.com/NousResearch/hermes-agent/pull/10065))
+- Fix: prevent premature loop exit when weak models return empty after substantive tool calls ([#10472](https://github.com/NousResearch/hermes-agent/pull/10472))
+- Fix: preserve pre-start terminal interrupts ([#10504](https://github.com/NousResearch/hermes-agent/pull/10504))
+- Fix: improve interrupt responsiveness during concurrent tool execution ([#10935](https://github.com/NousResearch/hermes-agent/pull/10935))
+- Fix: word-wrap spinner, interruptable agent join, and delegate_task interrupt ([#10940](https://github.com/NousResearch/hermes-agent/pull/10940))
+- Fix: `/stop` no longer resets the session ([#9224](https://github.com/NousResearch/hermes-agent/pull/9224))
+- Fix: honor interrupts during MCP tool waits ([#9382](https://github.com/NousResearch/hermes-agent/pull/9382), @helix4u)
+- Fix: break stuck session resume loops after repeated restarts ([#9941](https://github.com/NousResearch/hermes-agent/pull/9941))
+- Fix: empty response nudge crash + placeholder leak to cron targets ([#11021](https://github.com/NousResearch/hermes-agent/pull/11021))
+- Fix: streaming cursor sanitization to prevent message truncation (multiple PRs)
+- Fix: resolve `context_length` for plugin context engines ([#9238](https://github.com/NousResearch/hermes-agent/pull/9238))
+
+### Session & Memory
+- **Auto-prune old sessions + VACUUM state.db** at startup ([#13861](https://github.com/NousResearch/hermes-agent/pull/13861))
+- **Honcho overhaul** — context injection, 5-tool surface, cost safety, session isolation ([#10619](https://github.com/NousResearch/hermes-agent/pull/10619))
+- **Hindsight richer session-scoped retain metadata** (salvage of #6290) ([#13987](https://github.com/NousResearch/hermes-agent/pull/13987))
+- Fix: deduplicate memory provider tools to prevent 400 on strict providers ([#10511](https://github.com/NousResearch/hermes-agent/pull/10511))
+- Fix: discover user-installed memory providers from `$HERMES_HOME/plugins/` ([#10529](https://github.com/NousResearch/hermes-agent/pull/10529))
+- Fix: add `on_memory_write` bridge to sequential tool execution path ([#10507](https://github.com/NousResearch/hermes-agent/pull/10507))
+- Fix: preserve `session_id` across `previous_response_id` chains in `/v1/responses` ([#10059](https://github.com/NousResearch/hermes-agent/pull/10059))
+
+---
+
+## 🖥️ New Ink-based TUI
+
+A full React/Ink rewrite of the interactive CLI — invoked via `hermes --tui` or `HERMES_TUI=1`. Shipped across ~310 commits to `ui-tui/` and `tui_gateway/`.
+
+### TUI Foundations
+- New TUI based on Ink + Python JSON-RPC backend
+- Prettier + ESLint + vitest tooling for `ui-tui/`
+- Entry split between `src/entry.tsx` (TTY gate) and `src/app.tsx` (state machine)
+- Persistent `_SlashWorker` subprocess for slash command dispatch
+
+### UX & Features
+- **Stable picker keys, /clear confirm, light-theme preset** ([#12312](https://github.com/NousResearch/hermes-agent/pull/12312), @OutThisLife)
+- **Git branch in status bar** cwd label ([#12305](https://github.com/NousResearch/hermes-agent/pull/12305), @OutThisLife)
+- **Per-turn elapsed stopwatch in FaceTicker + done-in sys line** ([#13105](https://github.com/NousResearch/hermes-agent/pull/13105), @OutThisLife)
+- **Subagent spawn observability overlay** ([#14045](https://github.com/NousResearch/hermes-agent/pull/14045), @OutThisLife)
+- **Per-prompt elapsed stopwatch in status bar** ([#12948](https://github.com/NousResearch/hermes-agent/pull/12948))
+- Sticky composer that freezes during scroll
+- OSC-52 clipboard support for copy across SSH sessions
+- Virtualized history rendering for performance
+- Slash command autocomplete via `complete.slash` RPC
+- Path autocomplete via `complete.path` RPC
+- Dozens of resize/ghosting/sticky-prompt fixes landed through the week
+
+### Structural Refactors
+- Decomposed `app.tsx` into `app/event-handler`, `app/slash-handler`, `app/stores`, `app/hooks` ([#14640](https://github.com/NousResearch/hermes-agent/pull/14640) and surrounding)
+- Component split: `branding.tsx`, `markdown.tsx`, `prompts.tsx`, `sessionPicker.tsx`, `messageLine.tsx`, `thinking.tsx`, `maskedPrompt.tsx`
+- Hook split: `useCompletion`, `useInputHistory`, `useQueue`, `useVirtualHistory`
+
+---
+
+## 📱 Messaging Platforms (Gateway)
+
+### New Platforms
+- **QQBot (17th platform)** — QQ Official API v2 adapter with QR setup, streaming, package split ([#9364](https://github.com/NousResearch/hermes-agent/pull/9364), [#11831](https://github.com/NousResearch/hermes-agent/pull/11831))
+
+### Telegram
+- **Dedicated `TELEGRAM_PROXY` env var + config.yaml proxy support** (closes #9414, #6530, #9074, #7786) ([#10681](https://github.com/NousResearch/hermes-agent/pull/10681))
+- **`ignored_threads` config** for Telegram groups ([#9530](https://github.com/NousResearch/hermes-agent/pull/9530))
+- **Config option to disable link previews** (closes #8728) ([#10610](https://github.com/NousResearch/hermes-agent/pull/10610))
+- **Auto-wrap markdown tables** in code blocks ([#11794](https://github.com/NousResearch/hermes-agent/pull/11794))
+- Fix: prevent duplicate replies when stream task is cancelled ([#9319](https://github.com/NousResearch/hermes-agent/pull/9319))
+- Fix: prevent streaming cursor (▉) from appearing as standalone messages ([#9538](https://github.com/NousResearch/hermes-agent/pull/9538))
+- Fix: retry transient tool sends + cold-boot budget ([#10947](https://github.com/NousResearch/hermes-agent/pull/10947))
+- Fix: Markdown special char escaping in `send_exec_approval`
+- Fix: parentheses in URLs during MarkdownV2 link conversion
+- Fix: Unicode dash normalization in model switch (closes iOS smart-punctuation issue)
+- Many platform hint / streaming / session-key fixes
+
+### Discord
+- **Forum channel support** (salvage of #10145 + media + polish) ([#11920](https://github.com/NousResearch/hermes-agent/pull/11920))
+- **`DISCORD_ALLOWED_ROLES`** for role-based access control ([#11608](https://github.com/NousResearch/hermes-agent/pull/11608))
+- **Config option to disable slash commands** (salvage #13130) ([#14315](https://github.com/NousResearch/hermes-agent/pull/14315))
+- **Native `send_animation`** for inline GIF playback ([#10283](https://github.com/NousResearch/hermes-agent/pull/10283))
+- **`send_message` Discord media attachments** ([#10246](https://github.com/NousResearch/hermes-agent/pull/10246))
+- **`/skill` command group** with category subcommands ([#9909](https://github.com/NousResearch/hermes-agent/pull/9909))
+- **Extract reply text from message references** ([#9781](https://github.com/NousResearch/hermes-agent/pull/9781))
+
+### Feishu
+- **Intelligent reply on document comments** with 3-tier access control ([#11898](https://github.com/NousResearch/hermes-agent/pull/11898))
+- **Show processing state via reactions** on user messages ([#12927](https://github.com/NousResearch/hermes-agent/pull/12927))
+- **Preserve @mention context for agent consumption** (salvage #13874) ([#14167](https://github.com/NousResearch/hermes-agent/pull/14167))
+
+### DingTalk
+- **`require_mention` + `allowed_users` gating** (parity with Slack/Telegram/Discord) ([#11564](https://github.com/NousResearch/hermes-agent/pull/11564))
+- **QR-code device-flow authorization** for setup wizard ([#11574](https://github.com/NousResearch/hermes-agent/pull/11574))
+- **AI Cards streaming, emoji reactions, and media handling** (salvage of #10985) ([#11910](https://github.com/NousResearch/hermes-agent/pull/11910))
+
+### WhatsApp
+- **`send_voice`** — native audio message delivery ([#13002](https://github.com/NousResearch/hermes-agent/pull/13002))
+- **`dm_policy` and `group_policy`** parity with WeCom/Weixin/QQ adapters ([#13151](https://github.com/NousResearch/hermes-agent/pull/13151))
+
+### WeCom / Weixin
+- **WeCom QR-scan bot creation + interactive setup wizard** (salvage #13923) ([#13961](https://github.com/NousResearch/hermes-agent/pull/13961))
+
+### Signal
+- **Media delivery support** via `send_message` ([#13178](https://github.com/NousResearch/hermes-agent/pull/13178))
+
+### Slack
+- **Per-thread sessions for DMs by default** ([#10987](https://github.com/NousResearch/hermes-agent/pull/10987))
+
+### BlueBubbles (iMessage)
+- Group chat session separation, webhook registration & auth fixes ([#9806](https://github.com/NousResearch/hermes-agent/pull/9806))
+
+### Gateway Core
+- **Gateway proxy mode** — forward messages to a remote API server ([#9787](https://github.com/NousResearch/hermes-agent/pull/9787))
+- **Per-channel ephemeral prompts** (Discord, Telegram, Slack, Mattermost) ([#10564](https://github.com/NousResearch/hermes-agent/pull/10564))
+- **Surface plugin slash commands** natively on all platforms + decision-capable command hook ([#14175](https://github.com/NousResearch/hermes-agent/pull/14175))
+- **Support document/archive extensions in MEDIA: tag extraction** (salvage #8255) ([#14307](https://github.com/NousResearch/hermes-agent/pull/14307))
+- **Recognize `.pdf` in MEDIA: tag extraction** ([#13683](https://github.com/NousResearch/hermes-agent/pull/13683))
+- **`--all` flag for `gateway start` and `restart`** ([#10043](https://github.com/NousResearch/hermes-agent/pull/10043))
+- **Notify active sessions on gateway shutdown** + update health check ([#9850](https://github.com/NousResearch/hermes-agent/pull/9850))
+- **Block agent from self-destructing the gateway** via terminal (closes #6666) ([#9895](https://github.com/NousResearch/hermes-agent/pull/9895))
+- Fix: suppress duplicate replies on interrupt and streaming flood control ([#10235](https://github.com/NousResearch/hermes-agent/pull/10235))
+- Fix: close temporary agents after one-off tasks ([#11028](https://github.com/NousResearch/hermes-agent/pull/11028), @kshitijk4poor)
+- Fix: busy-session ack when user messages during active agent run ([#10068](https://github.com/NousResearch/hermes-agent/pull/10068))
+- Fix: route watch-pattern notifications to the originating session ([#10460](https://github.com/NousResearch/hermes-agent/pull/10460))
+- Fix: preserve notify context in executor threads ([#10921](https://github.com/NousResearch/hermes-agent/pull/10921), @kshitijk4poor)
+- Fix: avoid duplicate replies after interrupted long tasks ([#11018](https://github.com/NousResearch/hermes-agent/pull/11018))
+- Fix: unlink stale PID + lock files on cleanup
+- Fix: force-unlink stale PID file after `--replace` takeover
+
+---
+
+## 🔧 Tool System
+
+### Plugin Surface (major expansion)
+- **`register_command()`** — plugins can now add slash commands ([#10626](https://github.com/NousResearch/hermes-agent/pull/10626))
+- **`dispatch_tool()`** — plugins can invoke tools from their code ([#10763](https://github.com/NousResearch/hermes-agent/pull/10763))
+- **`pre_tool_call` blocking** — plugins can veto tool execution ([#9377](https://github.com/NousResearch/hermes-agent/pull/9377))
+- **`transform_tool_result`** — plugins rewrite tool results generically ([#12972](https://github.com/NousResearch/hermes-agent/pull/12972))
+- **`transform_terminal_output`** — plugins rewrite terminal tool output ([#12929](https://github.com/NousResearch/hermes-agent/pull/12929))
+- **Namespaced skill registration** for plugin skill bundles ([#9786](https://github.com/NousResearch/hermes-agent/pull/9786))
+- **Opt-in-by-default + bundled disk-cleanup plugin** (salvage #12212) ([#12944](https://github.com/NousResearch/hermes-agent/pull/12944))
+- **Pluggable `image_gen` backends + OpenAI provider** ([#13799](https://github.com/NousResearch/hermes-agent/pull/13799))
+- **`openai-codex` image_gen plugin** (gpt-image-2 via Codex OAuth) ([#14317](https://github.com/NousResearch/hermes-agent/pull/14317))
+- **Shell hooks** — wire shell scripts as hook callbacks ([#13296](https://github.com/NousResearch/hermes-agent/pull/13296))
+
+### Browser
+- **`browser_cdp` raw DevTools Protocol passthrough** ([#12369](https://github.com/NousResearch/hermes-agent/pull/12369))
+- Camofox hardening + connection stability across the window
+
+### Execute Code
+- **Project/strict execution modes** (default: project) ([#11971](https://github.com/NousResearch/hermes-agent/pull/11971))
+
+### Image Generation
+- **Multi-model FAL support** with picker in `hermes tools` ([#11265](https://github.com/NousResearch/hermes-agent/pull/11265))
+- **Recraft V3 → V4 Pro, Nano Banana → Pro upgrades** ([#11406](https://github.com/NousResearch/hermes-agent/pull/11406))
+- **GPT Image 2** in FAL catalog ([#13677](https://github.com/NousResearch/hermes-agent/pull/13677))
+- **xAI image generation provider** (grok-imagine-image) ([#14765](https://github.com/NousResearch/hermes-agent/pull/14765))
+
+### TTS / STT / Voice
+- **Google Gemini TTS provider** ([#11229](https://github.com/NousResearch/hermes-agent/pull/11229))
+- **xAI Grok STT provider** ([#14473](https://github.com/NousResearch/hermes-agent/pull/14473))
+- **xAI TTS** (shipped with Responses API upgrade) ([#10783](https://github.com/NousResearch/hermes-agent/pull/10783))
+- **KittenTTS local provider** (salvage of #2109) ([#13395](https://github.com/NousResearch/hermes-agent/pull/13395))
+- **CLI record beep toggle** ([#13247](https://github.com/NousResearch/hermes-agent/pull/13247), @helix4u)
+
+### Webhook / Cron
+- **Webhook direct-delivery mode** — zero-LLM push notifications ([#12473](https://github.com/NousResearch/hermes-agent/pull/12473))
+- **Cron `wakeAgent` gate** — scripts can skip the agent entirely ([#12373](https://github.com/NousResearch/hermes-agent/pull/12373))
+- **Cron per-job `enabled_toolsets`** — cap token overhead + cost per job ([#14767](https://github.com/NousResearch/hermes-agent/pull/14767))
+
+### Delegate
+- **Orchestrator role** + configurable spawn depth (default flat) ([#13691](https://github.com/NousResearch/hermes-agent/pull/13691))
+- **Cross-agent file state coordination** ([#13718](https://github.com/NousResearch/hermes-agent/pull/13718))
+
+### File / Patch
+- **`patch` — "did you mean?" feedback** when patch fails to match ([#13435](https://github.com/NousResearch/hermes-agent/pull/13435))
+
+### API Server
+- **Stream `/v1/responses` SSE tool events** (salvage #9779) ([#10049](https://github.com/NousResearch/hermes-agent/pull/10049))
+- **Inline image inputs** on `/v1/chat/completions` and `/v1/responses` ([#12969](https://github.com/NousResearch/hermes-agent/pull/12969))
+
+### Docker / Podman
+- **Entry-level Podman support** — `find_docker()` + rootless entrypoint ([#10066](https://github.com/NousResearch/hermes-agent/pull/10066))
+- **Add docker-cli to Docker image** (salvage #10096) ([#14232](https://github.com/NousResearch/hermes-agent/pull/14232))
+- **File-sync back to host on teardown** (salvage of #8189 + hardening) ([#11291](https://github.com/NousResearch/hermes-agent/pull/11291))
+
+### MCP
+- 12 MCP improvements across the window (status, timeout handling, tool-call forwarding, etc.)
+
+---
+
+## 🧩 Skills Ecosystem
+
+### Skill System
+- **Namespaced skill registration** for plugin bundles ([#9786](https://github.com/NousResearch/hermes-agent/pull/9786))
+- **`hermes skills reset`** to un-stick bundled skills ([#11468](https://github.com/NousResearch/hermes-agent/pull/11468))
+- **Skills guard opt-in** — `config.skills.guard_agent_created` (default off) ([#14557](https://github.com/NousResearch/hermes-agent/pull/14557))
+- **Bundled skill scripts runnable out of the box** ([#13384](https://github.com/NousResearch/hermes-agent/pull/13384))
+- **`xitter` replaced with `xurl`** — the official X API CLI ([#12303](https://github.com/NousResearch/hermes-agent/pull/12303))
+- **MiniMax-AI/cli as default skill tap** (salvage #7501) ([#14493](https://github.com/NousResearch/hermes-agent/pull/14493))
+- **Fuzzy `@` file completions + mtime sorting** ([#9467](https://github.com/NousResearch/hermes-agent/pull/9467))
+
+### New Skills
+- **concept-diagrams** (salvage of #11045, @v1k22) ([#11363](https://github.com/NousResearch/hermes-agent/pull/11363))
+- **architecture-diagram** (Cocoon AI port) ([#9906](https://github.com/NousResearch/hermes-agent/pull/9906))
+- **pixel-art** with hardware palettes and video animation ([#12663](https://github.com/NousResearch/hermes-agent/pull/12663), [#12725](https://github.com/NousResearch/hermes-agent/pull/12725))
+- **baoyu-comic** ([#13257](https://github.com/NousResearch/hermes-agent/pull/13257), @JimLiu)
+- **baoyu-infographic** — 21 layouts × 21 styles (salvage #9901) ([#12254](https://github.com/NousResearch/hermes-agent/pull/12254))
+- **page-agent** — embed Alibaba's in-page GUI agent in your webapp ([#13976](https://github.com/NousResearch/hermes-agent/pull/13976))
+- **fitness-nutrition** optional skill + optional env var support ([#9355](https://github.com/NousResearch/hermes-agent/pull/9355))
+- **drug-discovery** — ChEMBL, PubChem, OpenFDA, ADMET ([#9443](https://github.com/NousResearch/hermes-agent/pull/9443))
+- **touchdesigner-mcp** (salvage of #10081) ([#12298](https://github.com/NousResearch/hermes-agent/pull/12298))
+- **adversarial-ux-test** optional skill (salvage of #2494, @omnissiah-comelse) ([#13425](https://github.com/NousResearch/hermes-agent/pull/13425))
+- **maps** — added `guest_house`, `camp_site`, and dual-key bakery lookup ([#13398](https://github.com/NousResearch/hermes-agent/pull/13398))
+- **llm-wiki** — port provenance markers, source hashing, and quality signals ([#13700](https://github.com/NousResearch/hermes-agent/pull/13700))
+
+---
+
+## 📊 Web Dashboard
+
+- **i18n (English + Chinese) language switcher** ([#9453](https://github.com/NousResearch/hermes-agent/pull/9453))
+- **Live-switching theme system** ([#10687](https://github.com/NousResearch/hermes-agent/pull/10687))
+- **Dashboard plugin system** — extend the web UI with custom tabs ([#10951](https://github.com/NousResearch/hermes-agent/pull/10951))
+- **react-router, sidebar layout, sticky header, dropdown component** ([#9370](https://github.com/NousResearch/hermes-agent/pull/9370), @austinpickett)
+- **Responsive for mobile** ([#9228](https://github.com/NousResearch/hermes-agent/pull/9228), @DeployFaith)
+- **Vercel deployment** ([#10686](https://github.com/NousResearch/hermes-agent/pull/10686), [#11061](https://github.com/NousResearch/hermes-agent/pull/11061), @austinpickett)
+- **Context window config support** ([#9357](https://github.com/NousResearch/hermes-agent/pull/9357))
+- **HTTP health probe for cross-container gateway detection** ([#9894](https://github.com/NousResearch/hermes-agent/pull/9894))
+- **Update + restart gateway buttons** ([#13526](https://github.com/NousResearch/hermes-agent/pull/13526), @austinpickett)
+- **Real API call count per session** (salvages #10140) ([#14004](https://github.com/NousResearch/hermes-agent/pull/14004))
+
+---
+
+## 🖱️ CLI & User Experience
+
+- **Dynamic shell completion for bash, zsh, and fish** ([#9785](https://github.com/NousResearch/hermes-agent/pull/9785))
+- **Light-mode skins + skin-aware completion menus** ([#9461](https://github.com/NousResearch/hermes-agent/pull/9461))
+- **Numbered keyboard shortcuts** on approval and clarify prompts ([#13416](https://github.com/NousResearch/hermes-agent/pull/13416))
+- **Markdown stripping, compact multiline previews, external editor** ([#12934](https://github.com/NousResearch/hermes-agent/pull/12934))
+- **`--ignore-user-config` and `--ignore-rules` flags** (port codex#18646) ([#14277](https://github.com/NousResearch/hermes-agent/pull/14277))
+- **Account limits section in `/usage`** ([#13428](https://github.com/NousResearch/hermes-agent/pull/13428))
+- **Doctor: Command Installation check** for `hermes` bin symlink ([#10112](https://github.com/NousResearch/hermes-agent/pull/10112))
+- **ESC cancels secret/sudo prompts**, clearer skip messaging ([#9902](https://github.com/NousResearch/hermes-agent/pull/9902))
+- Fix: agent-facing text uses `display_hermes_home()` instead of hardcoded `~/.hermes` ([#10285](https://github.com/NousResearch/hermes-agent/pull/10285))
+- Fix: enforce `config.yaml` as sole CWD source + deprecate `.env` CWD vars + add `hermes memory reset` ([#11029](https://github.com/NousResearch/hermes-agent/pull/11029))
+
+---
+
+## 🔒 Security & Reliability
+
+- **Global toggle to allow private/internal URL resolution** ([#14166](https://github.com/NousResearch/hermes-agent/pull/14166))
+- **Block agent from self-destructing the gateway** via terminal (closes #6666) ([#9895](https://github.com/NousResearch/hermes-agent/pull/9895))
+- **Telegram callback authorization** on update prompts ([#10536](https://github.com/NousResearch/hermes-agent/pull/10536))
+- **SECURITY.md** added ([#10532](https://github.com/NousResearch/hermes-agent/pull/10532), @I3eg1nner)
+- **Warn about legacy hermes.service units** during `hermes update` ([#11918](https://github.com/NousResearch/hermes-agent/pull/11918))
+- **Complete ASCII-locale UnicodeEncodeError recovery** for `api_messages`/`reasoning_content` (closes #6843) ([#10537](https://github.com/NousResearch/hermes-agent/pull/10537))
+- **Prevent stale `os.environ` leak** after `clear_session_vars` ([#10527](https://github.com/NousResearch/hermes-agent/pull/10527))
+- **Prevent agent hang when backgrounding processes** via terminal tool ([#10584](https://github.com/NousResearch/hermes-agent/pull/10584))
+- Many smaller session-resume, interrupt, streaming, and memory-race fixes throughout the window
+
+---
+
+## 🐛 Notable Bug Fixes
+
+The `fix:` category in this window covers 482 PRs. Highlights:
+
+- Streaming cursor artifacts filtered from Matrix, Telegram, WhatsApp, Discord (multiple PRs)
+- `<think>` and `<thought>` blocks filtered from gateway stream consumers ([#9408](https://github.com/NousResearch/hermes-agent/pull/9408))
+- Gateway display.streaming root-config override regression ([#9799](https://github.com/NousResearch/hermes-agent/pull/9799))
+- Context `session_search` coerces limit to int (prevents TypeError) ([#10522](https://github.com/NousResearch/hermes-agent/pull/10522))
+- Memory tool stays available when `fcntl` is unavailable (Windows) ([#9783](https://github.com/NousResearch/hermes-agent/pull/9783))
+- Trajectory compressor credentials load from `HERMES_HOME/.env` ([#9632](https://github.com/NousResearch/hermes-agent/pull/9632), @Dusk1e)
+- `@_context_completions` no longer crashes on `@` mention ([#9683](https://github.com/NousResearch/hermes-agent/pull/9683), @kshitijk4poor)
+- Group session `user_id` no longer treated as `thread_id` in shutdown notifications ([#10546](https://github.com/NousResearch/hermes-agent/pull/10546))
+- Telegram `platform_hint` — markdown is supported (closes #8261) ([#10612](https://github.com/NousResearch/hermes-agent/pull/10612))
+- Doctor checks for Kimi China credentials fixed
+- Streaming: don't suppress final response when commentary message is sent ([#10540](https://github.com/NousResearch/hermes-agent/pull/10540))
+- Rapid Telegram follow-ups no longer get cut off
+
+---
+
+## 🧪 Testing & CI
+
+- **Contributor attribution CI check** on PRs ([#9376](https://github.com/NousResearch/hermes-agent/pull/9376))
+- Hermetic test parity (`scripts/run_tests.sh`) held across this window
+- Test count stabilized post-Transport refactor; CI matrix held green through the transport rollout
+
+---
+
+## 📚 Documentation
+
+- Atropos + wandb links in user guide
+- ACP / VS Code / Zed / JetBrains integration docs refresh
+- Webhook subscription docs updated for direct-delivery mode
+- Plugin author guide expanded for new hooks (`register_command`, `dispatch_tool`, `transform_tool_result`)
+- Transport layer developer guide added
+- Website removed Discussions link from README
+
+---
+
+## 👥 Contributors
+
+### Core
+- **@teknium1** (Teknium)
+
+### Top Community Contributors (by merged PR count)
+- **@kshitijk4poor** — 49 PRs · Transport refactor (AnthropicTransport, ResponsesApiTransport), Step Plan provider, Xiaomi MiMo v2.5 support, numerous gateway fixes, promoted Kimi K2.5, @ mention crash fix
+- **@OutThisLife** (Brooklyn) — 31 PRs · TUI polish, git branch in status bar, per-turn stopwatch, stable picker keys, `/clear` confirm, light-theme preset, subagent spawn observability overlay
+- **@helix4u** — 11 PRs · Voice CLI record beep, MCP tool interrupt handling, assorted stability fixes
+- **@austinpickett** — 8 PRs · Dashboard react-router + sidebar + sticky header + dropdown, Vercel deployment, update + restart buttons
+- **@alt-glitch** — 8 PRs · PLATFORM_HINTS for Matrix/Mattermost/Feishu, Matrix fixes
+- **@ethernet8023** — 3 PRs
+- **@benbarclay** — 3 PRs
+- **@Aslaaen** — 2 PRs
+
+### Also contributing
+@jerilynzheng (ai-gateway pricing), @JimLiu (baoyu-comic skill), @Dusk1e (trajectory compressor credentials), @DeployFaith (mobile-responsive dashboard), @LeonSGP43, @v1k22 (concept-diagrams), @omnissiah-comelse (adversarial-ux-test), @coekfung (Telegram MarkdownV2 expandable blockquotes), @liftaris (TUI provider resolution), @arihantsethia (skill analytics dashboard), @topcheer + @xing8star (QQBot foundation), @kovyrin, @I3eg1nner (SECURITY.md), @PeterBerthelsen, @lengxii, @priveperfumes, @sjz-ks, @cuyua9, @Disaster-Terminator, @leozeli, @LehaoLin, @trevthefoolish, @loongfay, @MrNiceRicee, @WideLee, @bluefishs, @malaiwah, @bobashopcashier, @dsocolobsky, @iamagenius00, @IAvecilla, @aniruddhaadak80, @Es1la, @asheriif, @walli, @jquesnelle (original Tool Gateway work).
+
+### All Contributors (alphabetical)
+
+@0xyg3n, @10ishq, @A-afflatus, @Abnertheforeman, @admin28980, @adybag14-cyber, @akhater, @alexzhu0,
+@AllardQuek, @alt-glitch, @aniruddhaadak80, @anna-oake, @anniesurla, @anthhub, @areu01or00, @arihantsethia,
+@arthurbr11, @asheriif, @Aslaaen, @Asunfly, @austinpickett, @AviArora02-commits, @AxDSan, @azhengbot, @Bartok9,
+@benbarclay, @bennytimz, @bernylinville, @bingo906, @binhnt92, @bkadish, @bluefishs, @bobashopcashier,
+@brantzh6, @BrennerSpear, @brianclemens, @briandevans, @brooklynnicholson, @bugkill3r, @buray, @burtenshaw,
+@cdanis, @cgarwood82, @ChimingLiu, @chongweiliu, @christopherwoodall, @coekfung, @cola-runner, @corazzione,
+@counterposition, @cresslank, @cuyua9, @cypres0099, @danieldoderlein, @davetist, @davidvv, @DeployFaith,
+@Dev-Mriganka, @devorun, @dieutx, @Disaster-Terminator, @dodo-reach, @draix, @DrStrangerUJN, @dsocolobsky,
+@Dusk1e, @dyxushuai, @elkimek, @elmatadorgh, @emozilla, @entropidelic, @Erosika, @erosika, @Es1la, @etcircle,
+@etherman-os, @ethernet8023, @fancydirty, @farion1231, @fatinghenji, @Fatty911, @fengtianyu88, @Feranmi10,
+@flobo3, @francip, @fuleinist, @g-guthrie, @GenKoKo, @gianfrancopiana, @gnanam1990, @GuyCui, @haileymarshall,
+@haimu0x, @handsdiff, @hansnow, @hedgeho9X, @helix4u, @hengm3467, @HenkDz, @heykb, @hharry11, @HiddenPuppy,
+@honghua, @houko, @houziershi, @hsy5571616, @huangke19, @hxp-plus, @Hypn0sis, @I3eg1nner, @iacker,
+@iamagenius00, @IAvecilla, @iborazzi, @Ifkellx, @ifrederico, @imink, @isaachuangGMICLOUD, @ismell0992-afk,
+@j0sephz, @Jaaneek, @jackjin1997, @JackTheGit, @jaffarkeikei, @jerilynzheng, @JiaDe-Wu, @Jiawen-lee, @JimLiu,
+@jinzheng8115, @jneeee, @jplew, @jquesnelle, @Julientalbot, @Junass1, @jvcl, @kagura-agent, @keifergu,
+@kevinskysunny, @keyuyuan, @konsisumer, @kovyrin, @kshitijk4poor, @leeyang1990, @LehaoLin, @lengxii,
+@LeonSGP43, @leozeli, @li0near, @liftaris, @Lind3ey, @Linux2010, @liujinkun2025, @LLQWQ, @Llugaes, @lmoncany,
+@longsizhuo, @lrawnsley, @Lubrsy706, @lumenradley, @luyao618, @lvnilesh, @LVT382009, @m0n5t3r, @Magaav,
+@MagicRay1217, @malaiwah, @manuelschipper, @Marvae, @MassiveMassimo, @mavrickdeveloper, @maxchernin, @memosr,
+@meng93, @mengjian-github, @MestreY0d4-Uninter, @Mibayy, @MikeFac, @mikewaters, @milkoor, @minorgod,
+@MrNiceRicee, @ms-alan, @mvanhorn, @n-WN, @N0nb0at, @Nan93, @NIDNASSER-Abdelmajid, @nish3451, @niyoh120,
+@nocoo, @nosleepcassette, @NousResearch, @ogzerber, @omnissiah-comelse, @Only-Code-A, @opriz, @OwenYWT, @pedh,
+@pefontana, @PeterBerthelsen, @phpoh, @pinion05, @plgonzalezrx8, @pradeep7127, @priveperfumes,
+@projectadmin-dev, @PStarH, @rnijhara, @Roy-oss1, @roytian1217, @RucchiZ, @Ruzzgar, @RyanLee-Dev, @Salt-555,
+@Sanjays2402, @sgaofen, @sharziki, @shenuu, @shin4, @SHL0MS, @shushuzn, @sicnuyudidi, @simon-gtcl,
+@simon-marcus, @sirEven, @Sisyphus, @sjz-ks, @snreynolds, @Societus, @Somme4096, @sontianye, @sprmn24,
+@StefanIsMe, @stephenschoettler, @Swift42, @taeng0204, @taeuk178, @tannerfokkens-maker, @TaroballzChen,
+@ten-ltw, @teyrebaz33, @Tianworld, @topcheer, @Tranquil-Flow, @trevthefoolish, @TroyMitchell911, @UNLINEARITY,
+@v1k22, @vivganes, @vominh1919, @vrinek, @VTRiot, @WadydX, @walli, @wenhao7, @WhiteWorld, @WideLee, @wujhsu,
+@WuTianyi123, @Wysie, @xandersbell, @xiaoqiang243, @xiayh0107, @xinpengdr, @Xowiek, @ycbai, @yeyitech, @ygd58,
+@youngDoo, @yudaiyan, @Yukipukii1, @yule975, @yyq4193, @yzx9, @ZaynJarvis, @zhang9w0v5, @zhanggttry,
+@zhangxicen, @zhongyueming1121, @zhouxiaoya12, @zons-zhaozhy
+
+Also: @maelrx, @Marco Rutsch, @MaxsolcuCrypto, @Mind-Dragon, @Paul Bergeron, @say8hi, @whitehatjr1001.
+
+
+---
+
+**Full Changelog**: [v2026.4.13...v2026.4.23](https://github.com/NousResearch/hermes-agent/compare/v2026.4.13...v2026.4.23)

RELEASE_v0.12.0.md

@@ -0,0 +1,505 @@
+# Hermes Agent v0.12.0 (v2026.4.30)
+
+**Release Date:** April 30, 2026
+**Since v0.11.0:** 1,096 commits · 550 merged PRs · 1,270 files changed · 217,776 insertions · 213 community contributors (including co-authors)
+
+> The Curator release — Hermes Agent now maintains itself. An autonomous background Curator grades, prunes, and consolidates your skill library on its own schedule. The self-improvement loop that reviews what to save got a substantial upgrade. Four new inference providers, a 18th messaging platform, a 19th via Teams plugin, native Spotify + Google Meet integrations, ComfyUI and TouchDesigner-MCP moved from optional to bundled-by-default, and a ~57% cut to visible TUI cold start.
+
+---
+
+## ✨ Highlights
+
+- **Autonomous Curator** — `hermes curator` runs as a background agent on the gateway's cron ticker (7-day cycle default). It grades your skill library, consolidates related skills, prunes dead ones, and writes per-run reports to `logs/curator/run.json` + `REPORT.md`. Archived skills are classified consolidated-vs-pruned via model + heuristic. Defense-in-depth gates protect bundled/hub skills from mutation. Unified under `auxiliary.curator` — pick the curator's model in `hermes model`, manage it from the dashboard. `hermes curator status` ranks skills by usage (most-used / least-used). ([#17277](https://github.com/NousResearch/hermes-agent/pull/17277), [#17307](https://github.com/NousResearch/hermes-agent/pull/17307), [#17941](https://github.com/NousResearch/hermes-agent/pull/17941), [#17868](https://github.com/NousResearch/hermes-agent/pull/17868), [#18033](https://github.com/NousResearch/hermes-agent/pull/18033))
+
+- **Self-improvement loop — substantially upgraded** — The background review fork (the core of Hermes' self-improvement: after each turn it decides what memories/skills to save or update) is now class-first (rubric-based rather than free-form), active-update biased (prefers the skill the agent just loaded), handles `references/`/`templates/` sub-files, and properly inherits the parent's live runtime (provider, model, credentials actually propagate). Restricted to memory + skills toolsets so it can't sprawl. Memory providers shut down cleanly. Prior-turn tool messages excluded from the summary so the fork sees a clean context. ([#16026](https://github.com/NousResearch/hermes-agent/pull/16026), [#17213](https://github.com/NousResearch/hermes-agent/pull/17213), [#16099](https://github.com/NousResearch/hermes-agent/pull/16099), [#16569](https://github.com/NousResearch/hermes-agent/pull/16569), [#16204](https://github.com/NousResearch/hermes-agent/pull/16204), [#15057](https://github.com/NousResearch/hermes-agent/pull/15057))
+
+- **Skill integrations — major expansion** — **ComfyUI v5** with official CLI + REST + hardware-gated local install, moved from optional to **built-in by default** ([#17610](https://github.com/NousResearch/hermes-agent/pull/17610), [#17631](https://github.com/NousResearch/hermes-agent/pull/17631), [#17734](https://github.com/NousResearch/hermes-agent/pull/17734)). **TouchDesigner-MCP** bundled by default, expanded with GLSL, post-FX, audio, geometry, and 9 new reference docs ([#16753](https://github.com/NousResearch/hermes-agent/pull/16753), [#16624](https://github.com/NousResearch/hermes-agent/pull/16624), [#16768](https://github.com/NousResearch/hermes-agent/pull/16768) — @kshitijk4poor + @SHL0MS). **Humanizer** skill ports a text-cleaner that strips AI-isms ([#16787](https://github.com/NousResearch/hermes-agent/pull/16787)). **claude-design** HTML artifact skill + design-md (Google DESIGN.md spec) + airtable salvage + `skill_manage` edits in `external_dirs` + direct-URL skill install + `/reload-skills` slash command. ([#16358](https://github.com/NousResearch/hermes-agent/pull/16358), [#14876](https://github.com/NousResearch/hermes-agent/pull/14876), [#16291](https://github.com/NousResearch/hermes-agent/pull/16291), [#17512](https://github.com/NousResearch/hermes-agent/pull/17512), [#16323](https://github.com/NousResearch/hermes-agent/pull/16323), [#17744](https://github.com/NousResearch/hermes-agent/pull/17744))
+
+- **LM Studio — first-class provider** — upgraded from a custom-endpoint alias to a full-blown native provider: dedicated auth, `hermes doctor` checks, reasoning transport, live `/models` listing. (Salvage of @kshitijk4poor's #17061.) ([#17102](https://github.com/NousResearch/hermes-agent/pull/17102))
+
+- **Four more new inference providers** — **GMI Cloud** (first-class, salvage of #11955 — @isaachuangGMICLOUD), **Azure AI Foundry** with auto-detection, **MiniMax OAuth** with PKCE browser flow (salvage #15203), **Tencent Tokenhub** (salvage of #16860). ([#16663](https://github.com/NousResearch/hermes-agent/pull/16663), [#15845](https://github.com/NousResearch/hermes-agent/pull/15845), [#17524](https://github.com/NousResearch/hermes-agent/pull/17524), [#16960](https://github.com/NousResearch/hermes-agent/pull/16960))
+
+- **Pluggable gateway platforms + Microsoft Teams** — the gateway is now a plugin host. Drop-in messaging adapters live outside the core, and Microsoft Teams is the first plugin-shipped platform. (Salvage of #17664.) ([#17751](https://github.com/NousResearch/hermes-agent/pull/17751), [#17828](https://github.com/NousResearch/hermes-agent/pull/17828))
+
+- **Tencent 元宝 (Yuanbao) — 18th messaging platform** — native gateway adapter with text + media delivery. ([#16298](https://github.com/NousResearch/hermes-agent/pull/16298), [#17424](https://github.com/NousResearch/hermes-agent/pull/17424))
+
+- **Spotify — native tools + bundled skill + wizard** — 7 tools (play, search, queue, playlists, devices) behind PKCE OAuth, interactive setup wizard, bundled skill, surfacing in `hermes tools`, cron usage documented. ([#15121](https://github.com/NousResearch/hermes-agent/pull/15121), [#15130](https://github.com/NousResearch/hermes-agent/pull/15130), [#15154](https://github.com/NousResearch/hermes-agent/pull/15154), [#15180](https://github.com/NousResearch/hermes-agent/pull/15180))
+
+- **Google Meet plugin** — join calls, transcribe, speak, follow up. Realtime OpenAI transport + Node bot server, full pipeline bundled as a plugin. ([#16364](https://github.com/NousResearch/hermes-agent/pull/16364))
+
+- **`hermes -z` one-shot mode + `hermes update --check`** — non-interactive `hermes -z <prompt>` with `--model`/`--provider`/`HERMES_INFERENCE_MODEL`. `hermes update --check` preflight. Opt-in pre-update HERMES_HOME backup. ([#15702](https://github.com/NousResearch/hermes-agent/pull/15702), [#15704](https://github.com/NousResearch/hermes-agent/pull/15704), [#15841](https://github.com/NousResearch/hermes-agent/pull/15841), [#16539](https://github.com/NousResearch/hermes-agent/pull/16539), [#16566](https://github.com/NousResearch/hermes-agent/pull/16566))
+
+- **Models dashboard tab + in-browser model config** — rich per-model analytics, switch main + auxiliary models from the dashboard. ([#17745](https://github.com/NousResearch/hermes-agent/pull/17745), [#17802](https://github.com/NousResearch/hermes-agent/pull/17802))
+
+- **Remote model catalog manifest** — OpenRouter + Nous Portal model catalogs are now pulled from a remote manifest so new models show up without a release. ([#16033](https://github.com/NousResearch/hermes-agent/pull/16033))
+
+- **Native multimodal image routing** — images now route based on the model's actual vision capability rather than provider defaults. ([#16506](https://github.com/NousResearch/hermes-agent/pull/16506))
+
+- **Gateway media parity** — native multi-image sending across Telegram, Discord, Slack, Mattermost, Email, and Signal; centralized audio routing with FLAC support + Telegram document fallback. ([#17909](https://github.com/NousResearch/hermes-agent/pull/17909), [#17833](https://github.com/NousResearch/hermes-agent/pull/17833))
+
+- **TUI catches up to (and past) the classic CLI** — LaTeX rendering (@austinpickett), `/reload` .env hot-reload, pluggable busy-indicator styles (@OutThisLife, #13610), opt-in auto-resume of last session, expanded light-terminal auto-detection, session delete from `/resume` picker with `d`, modified mouse-wheel line scroll, and a `/mouse` toggle that kills ConPTY's phantom mouse injection (@kevin-ho). ([#17175](https://github.com/NousResearch/hermes-agent/pull/17175), [#17286](https://github.com/NousResearch/hermes-agent/pull/17286), [#17150](https://github.com/NousResearch/hermes-agent/pull/17150), [#17130](https://github.com/NousResearch/hermes-agent/pull/17130), [#17113](https://github.com/NousResearch/hermes-agent/pull/17113), [#17668](https://github.com/NousResearch/hermes-agent/pull/17668), [#17669](https://github.com/NousResearch/hermes-agent/pull/17669), [#15488](https://github.com/NousResearch/hermes-agent/pull/15488))
+
+- **Observability + achievements plugins** — bundled Langfuse observability plugin (salvage #16845) + bundled hermes-achievements plugin that scans full session history. ([#16917](https://github.com/NousResearch/hermes-agent/pull/16917), [#17754](https://github.com/NousResearch/hermes-agent/pull/17754))
+
+- **TTS provider registry + Piper local TTS** — pluggable `tts.providers.<name>` registry; Piper ships as a native local TTS provider. (Closes #8508.) ([#17843](https://github.com/NousResearch/hermes-agent/pull/17843), [#17885](https://github.com/NousResearch/hermes-agent/pull/17885))
+
+- **Vercel Sandbox backend** — Vercel sandboxes as an execute_code/terminal backend (@kshitijk4poor). ([#17445](https://github.com/NousResearch/hermes-agent/pull/17445))
+
+- **Secret redaction off by default** — default flipped to off. Prevents the long-standing patch-corruption incidents where fake secret-shaped substrings mangled tool outputs. Opt in via `redaction.enabled: true` when you need it. ([#16794](https://github.com/NousResearch/hermes-agent/pull/16794))
+
+- **Cold-start performance** — visible TUI cold start cut **~57%** via lazy agent init (@OutThisLife), lazy imports of OpenAI / Anthropic / Firecrawl / account_usage, mtime-cached `load_config()`, memoized `get_tool_definitions()` with TTL-cached `check_fn` results, precompiled dangerous-command patterns. ([#17190](https://github.com/NousResearch/hermes-agent/pull/17190), [#17046](https://github.com/NousResearch/hermes-agent/pull/17046), [#17041](https://github.com/NousResearch/hermes-agent/pull/17041), [#17098](https://github.com/NousResearch/hermes-agent/pull/17098), [#17206](https://github.com/NousResearch/hermes-agent/pull/17206))
+
+- **Configurable prompt cache TTL** — `prompt_caching.cache_ttl` (5m default, 1h opt-in — cost savings for bursty sessions that keep cache warm). Salvage of #12659. ([#15065](https://github.com/NousResearch/hermes-agent/pull/15065))
+
+---
+
+## 🧠 Autonomous Curator & Self-Improvement Loop
+
+### Curator — autonomous skill maintenance
+- **`hermes curator` as a background agent** — runs on the gateway's cron ticker, 7-day cycle by default, umbrella-first prompt, inherits parent config, unbounded iterations ([#17277](https://github.com/NousResearch/hermes-agent/pull/17277) — issue #7816)
+- **Per-run reports** — `logs/curator/run.json` + `REPORT.md` per cycle ([#17307](https://github.com/NousResearch/hermes-agent/pull/17307))
+- **Consolidated vs pruned classification** — archived skills split with model + heuristic ([#17941](https://github.com/NousResearch/hermes-agent/pull/17941))
+- **`hermes curator status`** — ranks skills by usage, shows most-used and least-used ([#18033](https://github.com/NousResearch/hermes-agent/pull/18033))
+- **Unified under `auxiliary.curator`** — pick the model in `hermes model`, configure from the dashboard ([#17868](https://github.com/NousResearch/hermes-agent/pull/17868))
+- **Documentation** — dedicated curator feature page on the docs site ([#17563](https://github.com/NousResearch/hermes-agent/pull/17563))
+- Fix: seed defaults on update, create `logs/curator/` directory, defer fire import ([#17927](https://github.com/NousResearch/hermes-agent/pull/17927))
+- Fix: scan nested archive subdirs in `restore_skill` (@0xDevNinja) ([#17951](https://github.com/NousResearch/hermes-agent/pull/17951))
+- Fix: use actual skill activity in curator status (@y0shua1ee) ([#17953](https://github.com/NousResearch/hermes-agent/pull/17953))
+- Fix: `skill_manage` refuses writes on pinned skills; pinning now blocks curator writes ([#17562](https://github.com/NousResearch/hermes-agent/pull/17562), [#17578](https://github.com/NousResearch/hermes-agent/pull/17578))
+- Fix: `bump_use()` wired into skill invocation + preload + skill_view (salvage #17782) ([#17932](https://github.com/NousResearch/hermes-agent/pull/17932))
+
+### Self-improvement loop (background review fork)
+- **Class-first skill-review prompt** — rubric-based grading rather than free-form "should this update" ([#16026](https://github.com/NousResearch/hermes-agent/pull/16026))
+- **Active-update bias** — prefers updating skills the agent just loaded, handles `references/` + `templates/` sub-files ([#17213](https://github.com/NousResearch/hermes-agent/pull/17213))
+- **Fork inherits parent's live runtime** — provider, model, credentials actually propagate now ([#16099](https://github.com/NousResearch/hermes-agent/pull/16099))
+- **Scoped toolsets** — review fork restricted to memory + skills (no shell, no web) ([#16569](https://github.com/NousResearch/hermes-agent/pull/16569))
+- **Clean shutdown** — background review memory providers exit properly (salvage #15289) ([#16204](https://github.com/NousResearch/hermes-agent/pull/16204))
+- **Clean context** — prior-history tool messages excluded from review summary (salvage #14967) ([#15057](https://github.com/NousResearch/hermes-agent/pull/15057))
+
+---
+
+## 🧩 Skills Ecosystem
+
+### Skill integrations — newly bundled or promoted
+- **ComfyUI v5** — official CLI + REST + hardware-gated local install; **moved from optional to built-in** ([#17610](https://github.com/NousResearch/hermes-agent/pull/17610), [#17631](https://github.com/NousResearch/hermes-agent/pull/17631), [#17734](https://github.com/NousResearch/hermes-agent/pull/17734), [#17612](https://github.com/NousResearch/hermes-agent/pull/17612))
+- **TouchDesigner-MCP** — **bundled by default** ([#16753](https://github.com/NousResearch/hermes-agent/pull/16753) — @kshitijk4poor), expanded with GLSL, post-FX, audio, geometry references ([#16624](https://github.com/NousResearch/hermes-agent/pull/16624)), 9 new reference docs ([#16768](https://github.com/NousResearch/hermes-agent/pull/16768) — @SHL0MS)
+- **Humanizer** — strips AI-isms from text ([#16787](https://github.com/NousResearch/hermes-agent/pull/16787))
+- **claude-design** — HTML artifact skill with disambiguation from other design skills ([#16358](https://github.com/NousResearch/hermes-agent/pull/16358))
+- **design-md** — Google's DESIGN.md spec skill ([#14876](https://github.com/NousResearch/hermes-agent/pull/14876))
+- **airtable** — salvaged skill + skill API keys wired into `.env` (#15838) ([#16291](https://github.com/NousResearch/hermes-agent/pull/16291))
+- **pretext** — creative browser demos with @chenglou/pretext ([#17259](https://github.com/NousResearch/hermes-agent/pull/17259))
+- **spike** + **sketch** — throwaway experiments + HTML mockups, adapted from gsd-build ([#17421](https://github.com/NousResearch/hermes-agent/pull/17421))
+
+### Skills UX
+- **Install skills from a direct HTTP(S) URL** — `hermes skills install <url>` ([#16323](https://github.com/NousResearch/hermes-agent/pull/16323))
+- **`/reload-skills`** slash command (salvage #17670) ([#17744](https://github.com/NousResearch/hermes-agent/pull/17744))
+- **`hermes skills list`** shows enabled/disabled status ([#16129](https://github.com/NousResearch/hermes-agent/pull/16129))
+- **`skill_manage` refuses writes on pinned skills** ([#17562](https://github.com/NousResearch/hermes-agent/pull/17562))
+- **`skill_manage` edits external_dirs skills in place** (salvage #9966) ([#17512](https://github.com/NousResearch/hermes-agent/pull/17512), [#17289](https://github.com/NousResearch/hermes-agent/pull/17289))
+- Fix: inline-shell rendering in `skill_view` ([#15376](https://github.com/NousResearch/hermes-agent/pull/15376))
+- Fix: exclude `.archive/` from skill index walk (salvage #17639) ([#17931](https://github.com/NousResearch/hermes-agent/pull/17931))
+- Fix: dedicated docs page per bundled + optional skill ([#14929](https://github.com/NousResearch/hermes-agent/pull/14929))
+- Fix: `google-workspace` shared HERMES_HOME helper + ship deps as optional extra ([#15405](https://github.com/NousResearch/hermes-agent/pull/15405))
+- Fix: auto-wrap ASCII-art code blocks in generated skill pages ([#16497](https://github.com/NousResearch/hermes-agent/pull/16497))
+- Point agent at `hermes-agent` skill + docs site for Hermes questions ([#16535](https://github.com/NousResearch/hermes-agent/pull/16535))
+
+---
+
+## 🏗️ Core Agent & Architecture
+
+### Provider & Model Support
+
+#### New providers
+- **GMI Cloud** — first-class API-key provider on par with Arcee/Kilocode/Xiaomi (salvage of #11955 — @isaachuangGMICLOUD) ([#16663](https://github.com/NousResearch/hermes-agent/pull/16663))
+- **Azure AI Foundry** — auto-detection, full wiring ([#15845](https://github.com/NousResearch/hermes-agent/pull/15845))
+- **LM Studio** — upgraded from custom-endpoint alias to first-class provider: dedicated auth, doctor checks, reasoning transport, live `/models` (salvage of #17061 — @kshitijk4poor) ([#17102](https://github.com/NousResearch/hermes-agent/pull/17102))
+- **MiniMax OAuth** — PKCE browser flow with full OAuth integration (salvage #15203) ([#17524](https://github.com/NousResearch/hermes-agent/pull/17524))
+- **Tencent Tokenhub** — new provider (salvage of #16860) ([#16960](https://github.com/NousResearch/hermes-agent/pull/16960))
+
+#### Model catalog
+- **Remote model catalog manifest** — OpenRouter + Nous Portal catalogs pulled from remote manifest so new models show up without a release ([#16033](https://github.com/NousResearch/hermes-agent/pull/16033))
+- `openai/gpt-5.5` and `gpt-5.5-pro` added to OpenRouter + Nous Portal ([#15343](https://github.com/NousResearch/hermes-agent/pull/15343))
+- `deepseek-v4-pro` and `deepseek-v4-flash` added ([#14934](https://github.com/NousResearch/hermes-agent/pull/14934))
+- `qwen3.6-plus` added to Alibaba-supported models ([#16896](https://github.com/NousResearch/hermes-agent/pull/16896))
+- Gemini free-tier keys blocked at setup with 429 guidance surfacing ([#15100](https://github.com/NousResearch/hermes-agent/pull/15100))
+
+#### Model configuration
+- **Configurable `prompt_caching.cache_ttl`** — 5m default, 1h opt-in (salvage #12659) ([#15065](https://github.com/NousResearch/hermes-agent/pull/15065))
+- `/fast` whitelist broadened to all OpenAI + Anthropic models ([#16883](https://github.com/NousResearch/hermes-agent/pull/16883))
+- `auxiliary.extra_body.reasoning` translates into Codex Responses API ([#17004](https://github.com/NousResearch/hermes-agent/pull/17004))
+- `hermes fallback` command for managing fallback providers ([#16052](https://github.com/NousResearch/hermes-agent/pull/16052))
+
+### Agent Loop & Conversation
+- **Native multimodal image routing** — based on model vision capability, not provider defaults ([#16506](https://github.com/NousResearch/hermes-agent/pull/16506))
+- **Delegate `child_timeout_seconds` default bumped to 600s** ([#14809](https://github.com/NousResearch/hermes-agent/pull/14809))
+- **Diagnostic dump when subagent times out with 0 API calls** ([#15105](https://github.com/NousResearch/hermes-agent/pull/15105))
+- **Gateway busts cached agent on compression/context_length config edits** ([#17008](https://github.com/NousResearch/hermes-agent/pull/17008))
+- **Opt-in runtime-metadata footer on final replies** ([#17026](https://github.com/NousResearch/hermes-agent/pull/17026))
+- `/reload-mcp` awareness — rebuild cached agents + prompt-cache cost confirmation ([#17729](https://github.com/NousResearch/hermes-agent/pull/17729))
+- Fix: repair CamelCase + `_tool` suffix tool-call emissions ([#15124](https://github.com/NousResearch/hermes-agent/pull/15124))
+- Fix: retry on `json.JSONDecodeError` instead of treating as local validation error ([#15107](https://github.com/NousResearch/hermes-agent/pull/15107))
+- Fix: handle unescaped control chars in `tool_call.arguments` ([#15356](https://github.com/NousResearch/hermes-agent/pull/15356))
+- Fix: ordering fix in `_copy_reasoning_content_for_api` — cross-provider reasoning isolation (@Zjianru) ([#15749](https://github.com/NousResearch/hermes-agent/pull/15749))
+- Fix: inject empty `reasoning_content` for DeepSeek/Kimi `tool_calls` unconditionally (@Zjianru) ([#15762](https://github.com/NousResearch/hermes-agent/pull/15762))
+- Fix: persist streamed `reasoning_content` on assistant turns (#16844) ([#16892](https://github.com/NousResearch/hermes-agent/pull/16892))
+- Fix: cancel coroutine on timeout so worker thread exits; full traceback on tool failure ([#17428](https://github.com/NousResearch/hermes-agent/pull/17428))
+- Fix: isolate `get_tool_definitions` quiet_mode cache + dedup LCM injection (#17335) ([#17889](https://github.com/NousResearch/hermes-agent/pull/17889))
+- Fix: serialize concurrent `hermes_tools` RPC calls from `execute_code` (#17770) ([#17894](https://github.com/NousResearch/hermes-agent/pull/17894), [#17902](https://github.com/NousResearch/hermes-agent/pull/17902))
+- Fix: rename `[SYSTEM:` → `[IMPORTANT:` in all user-injected markers (dodges Azure content filter) ([#16114](https://github.com/NousResearch/hermes-agent/pull/16114))
+
+### Compression
+- **Retry summary on main model for unknown errors before giving up** ([#16774](https://github.com/NousResearch/hermes-agent/pull/16774))
+- **Notify users when configured aux model fails even if main-model fallback recovers** ([#16775](https://github.com/NousResearch/hermes-agent/pull/16775))
+- `/compress` wrapped in `_busy_command` to block input during compression ([#15388](https://github.com/NousResearch/hermes-agent/pull/15388))
+- Fix: reserve system + tools headroom when aux binds threshold ([#15631](https://github.com/NousResearch/hermes-agent/pull/15631))
+- Fix: use text-char sum for multimodal token estimation in `_find_tail_cut_by_tokens` ([#16369](https://github.com/NousResearch/hermes-agent/pull/16369))
+
+### Session, Memory & State
+- **Trigram FTS5 index for CJK search, replace LIKE fallback** (@alt-glitch) ([#16651](https://github.com/NousResearch/hermes-agent/pull/16651))
+- **Index `tool_name` + `tool_calls` in FTS5, with repair + migration** (salvages #16866) ([#16914](https://github.com/NousResearch/hermes-agent/pull/16914))
+- **Checkpoints: auto-prune orphan and stale shadow repos at startup** ([#16303](https://github.com/NousResearch/hermes-agent/pull/16303))
+- **Memory providers notified on mid-process session_id rotation** (#6672) ([#17409](https://github.com/NousResearch/hermes-agent/pull/17409))
+- Fix: quote underscored terms in FTS5 query sanitization ([#16915](https://github.com/NousResearch/hermes-agent/pull/16915))
+- Fix: resolve viking_read 500/412 on file URIs + pseudo-summary URIs (salvage #5886) ([#17869](https://github.com/NousResearch/hermes-agent/pull/17869))
+- Fix: skip external-provider sync on interrupted turns ([#15395](https://github.com/NousResearch/hermes-agent/pull/15395))
+- Fix: close embedded Hindsight async client cleanly (salvage #14605) ([#16209](https://github.com/NousResearch/hermes-agent/pull/16209))
+- Fix: pass session transcript to `shutdown_memory_provider` on gateway + CLI (#15165) ([#16571](https://github.com/NousResearch/hermes-agent/pull/16571))
+- Fix: write-origin metadata seam ([#15346](https://github.com/NousResearch/hermes-agent/pull/15346))
+- Fix: preserve symlinks during atomic file writes ([#16980](https://github.com/NousResearch/hermes-agent/pull/16980))
+- Refactor: remove `flush_memories` entirely ([#15696](https://github.com/NousResearch/hermes-agent/pull/15696))
+
+### Auxiliary models
+- Fix: surface auxiliary failures in UI (previously silent) ([#15324](https://github.com/NousResearch/hermes-agent/pull/15324))
+- Fix: surface title-gen auxiliary failures instead of silently dropping ([#16371](https://github.com/NousResearch/hermes-agent/pull/16371))
+- Fix: generalize unsupported-parameter detector and harden `max_tokens` retry ([#15633](https://github.com/NousResearch/hermes-agent/pull/15633))
+
+---
+
+## 📱 Messaging Platforms (Gateway)
+
+### New Platforms
+- **Microsoft Teams (19th platform)** — as a plugin, + xdist collision guard ([#17828](https://github.com/NousResearch/hermes-agent/pull/17828))
+- **Yuanbao (Tencent 元宝, 18th platform)** — native adapter with text + media delivery ([#16298](https://github.com/NousResearch/hermes-agent/pull/16298), [#17424](https://github.com/NousResearch/hermes-agent/pull/17424), [#16880](https://github.com/NousResearch/hermes-agent/pull/16880))
+
+### Pluggable Gateway Platforms
+- **Drop-in messaging adapters** — the gateway is now a plugin host for platforms (salvage of #17664) ([#17751](https://github.com/NousResearch/hermes-agent/pull/17751))
+
+### Telegram
+- **Chat allowlists for groups and forums** (@web3blind) ([#15027](https://github.com/NousResearch/hermes-agent/pull/15027))
+- **Send fresh finals for stale preview streams** (port openclaw#72038) ([#16261](https://github.com/NousResearch/hermes-agent/pull/16261))
+- **Render markdown tables as row-group bullets + prompt hint** ([#16997](https://github.com/NousResearch/hermes-agent/pull/16997))
+- Document fallback in centralized audio routing ([#17833](https://github.com/NousResearch/hermes-agent/pull/17833))
+- Native multi-image sending ([#17909](https://github.com/NousResearch/hermes-agent/pull/17909))
+
+### Discord
+- **Opt-in toolsets + ID injection + tool split + Feishu wiring** (salvage #15457, #15458) ([#15610](https://github.com/NousResearch/hermes-agent/pull/15610), [#15613](https://github.com/NousResearch/hermes-agent/pull/15613))
+- Fix: coerce `limit` parameter to int before `min()` call ([#16319](https://github.com/NousResearch/hermes-agent/pull/16319))
+
+### Slack
+- **Register every gateway command as a native slash (Discord/Telegram parity)** ([#16164](https://github.com/NousResearch/hermes-agent/pull/16164))
+- **`strict_mention` config** — prevents thread auto-engagement ([#16193](https://github.com/NousResearch/hermes-agent/pull/16193))
+- **`channel_skill_bindings`** — bind specific skills to specific Slack channels ([#16283](https://github.com/NousResearch/hermes-agent/pull/16283))
+
+### Signal
+- **Native formatting** — markdown → bodyRanges, reply quotes, reactions ([#17417](https://github.com/NousResearch/hermes-agent/pull/17417))
+- Native multi-image sending ([#17909](https://github.com/NousResearch/hermes-agent/pull/17909))
+
+### Feishu / Mattermost / Email / Signal
+- All participate in **native multi-image sending** ([#17909](https://github.com/NousResearch/hermes-agent/pull/17909))
+
+### Gateway Core
+- **Centralized audio routing + FLAC support + Telegram doc fallback** ([#17833](https://github.com/NousResearch/hermes-agent/pull/17833))
+- **Native multi-image sending** across Telegram, Discord, Slack, Mattermost, Email, Signal ([#17909](https://github.com/NousResearch/hermes-agent/pull/17909))
+- **Make hygiene hard message limit configurable** ([#17000](https://github.com/NousResearch/hermes-agent/pull/17000))
+- **Opt-in runtime-metadata footer on final replies** ([#17026](https://github.com/NousResearch/hermes-agent/pull/17026))
+- **`pre_gateway_dispatch` hook** — plugins can intercept before dispatch ([#15050](https://github.com/NousResearch/hermes-agent/pull/15050))
+- **`pre_approval_request` / `post_approval_response` hooks** ([#16776](https://github.com/NousResearch/hermes-agent/pull/16776))
+- Fix: timeouts — guard `load_config()` call against runtime exceptions ([#16318](https://github.com/NousResearch/hermes-agent/pull/16318))
+- Fix: support passing handler tools via registry ([#15613](https://github.com/NousResearch/hermes-agent/pull/15613))
+
+---
+
+## 🔧 Tool System
+
+### Plugin-first architecture
+- **Pluggable gateway platforms** — platforms can ship as plugins ([#17751](https://github.com/NousResearch/hermes-agent/pull/17751))
+- **Microsoft Teams as first plugin-shipped platform** ([#17828](https://github.com/NousResearch/hermes-agent/pull/17828))
+- **`pre_gateway_dispatch` hook** ([#15050](https://github.com/NousResearch/hermes-agent/pull/15050))
+- **`pre_approval_request` + `post_approval_response` hooks** ([#16776](https://github.com/NousResearch/hermes-agent/pull/16776))
+- **`duration_ms` on `post_tool_call`** (inspired by Claude Code 2.1.119) ([#15429](https://github.com/NousResearch/hermes-agent/pull/15429))
+- **Bundled plugins**: Spotify ([#15174](https://github.com/NousResearch/hermes-agent/pull/15174)), Google Meet ([#16364](https://github.com/NousResearch/hermes-agent/pull/16364)), Langfuse observability ([#16917](https://github.com/NousResearch/hermes-agent/pull/16917)), hermes-achievements ([#17754](https://github.com/NousResearch/hermes-agent/pull/17754))
+- **Page-scoped plugin slots for built-in dashboard pages** ([#15658](https://github.com/NousResearch/hermes-agent/pull/15658))
+- **Declarative plugin installation for NixOS module** (@alt-glitch) ([#15953](https://github.com/NousResearch/hermes-agent/pull/15953))
+
+### Browser
+- **CDP supervisor** — dialog detection + response + cross-origin iframe eval ([#14540](https://github.com/NousResearch/hermes-agent/pull/14540))
+- **Auto-spawn local Chromium for LAN/localhost URLs** when cloud provider is configured ([#16136](https://github.com/NousResearch/hermes-agent/pull/16136))
+
+### Execute code / Terminal
+- **Vercel Sandbox backend** for `execute_code` / terminal (@kshitijk4poor) ([#17445](https://github.com/NousResearch/hermes-agent/pull/17445))
+- **Collapse subagent `task_id`s to shared container** ([#16177](https://github.com/NousResearch/hermes-agent/pull/16177))
+- **Docker: run container as host user** to avoid root-owned bind mounts (@benbarclay) ([#17305](https://github.com/NousResearch/hermes-agent/pull/17305))
+- Fix: safely quote `~/` subpaths in wrapped `cd` commands ([#15394](https://github.com/NousResearch/hermes-agent/pull/15394))
+- Fix: close file descriptor in `LocalEnvironment._update_cwd` ([#17300](https://github.com/NousResearch/hermes-agent/pull/17300))
+- Fix: SSH — prevent tar from overwriting remote home dir permissions ([#17898](https://github.com/NousResearch/hermes-agent/pull/17898), [#17867](https://github.com/NousResearch/hermes-agent/pull/17867))
+
+### Image generation
+- See Provider section for updates; no new image providers this window.
+
+### TTS / Voice
+- **Pluggable TTS provider registry** under `tts.providers.<name>` ([#17843](https://github.com/NousResearch/hermes-agent/pull/17843))
+- **Piper** as native local TTS provider (closes #8508) ([#17885](https://github.com/NousResearch/hermes-agent/pull/17885))
+- **Voice mode CLI parity in the TUI** — VAD loop + TTS + crash forensics ([#14810](https://github.com/NousResearch/hermes-agent/pull/14810))
+- Fix: vision — use HERMES_HOME-based cache dir instead of cwd ([#17719](https://github.com/NousResearch/hermes-agent/pull/17719))
+
+### Cron
+- **Honor `hermes tools` config for the cron platform** ([#14798](https://github.com/NousResearch/hermes-agent/pull/14798))
+- **Per-job `workdir`** — project-aware cron runs ([#15110](https://github.com/NousResearch/hermes-agent/pull/15110))
+- **`context_from` field** — chain cron job outputs ([#15606](https://github.com/NousResearch/hermes-agent/pull/15606))
+- Fix: promote `croniter` to a core dependency ([#17577](https://github.com/NousResearch/hermes-agent/pull/17577))
+
+### Web search
+- **Expose `limit` for `web_search`** ([#16934](https://github.com/NousResearch/hermes-agent/pull/16934))
+
+### Maps
+- Fix: include seconds in timezone UTC offset output ([#16300](https://github.com/NousResearch/hermes-agent/pull/16300))
+
+### Approvals
+- **Hardline blocklist for unrecoverable commands** ([#15878](https://github.com/NousResearch/hermes-agent/pull/15878))
+- Perf: precompile DANGEROUS_PATTERNS and HARDLINE_PATTERNS ([#17206](https://github.com/NousResearch/hermes-agent/pull/17206))
+
+### ACP
+- **Advertise and forward image prompts** ([#18030](https://github.com/NousResearch/hermes-agent/pull/18030))
+
+### API Server
+- **POST `/v1/runs/{run_id}/stop`** (salvage of #15656) ([#15842](https://github.com/NousResearch/hermes-agent/pull/15842))
+- **Expose run status for external UIs** (#17085) ([#17458](https://github.com/NousResearch/hermes-agent/pull/17458))
+
+### Nix
+- **Declarative plugin installation for NixOS module** (@alt-glitch) ([#15953](https://github.com/NousResearch/hermes-agent/pull/15953))
+- Fix: use `--rebuild` in fix-lockfiles to bypass cached FOD store paths ([#15444](https://github.com/NousResearch/hermes-agent/pull/15444))
+- Fix: `extraPackages` now actually works via per-user profile ([#17047](https://github.com/NousResearch/hermes-agent/pull/17047))
+- Fix: refresh web/ npm-deps hash to unblock main builds ([#17174](https://github.com/NousResearch/hermes-agent/pull/17174))
+- Fix: replace magic-nix-cache with Cachix ([#17928](https://github.com/NousResearch/hermes-agent/pull/17928))
+
+---
+
+## 🖥️ TUI
+
+### New features
+- **LaTeX rendering** (@austinpickett) ([#17175](https://github.com/NousResearch/hermes-agent/pull/17175))
+- **`/reload` .env hot-reload** — ported from the classic CLI ([#17286](https://github.com/NousResearch/hermes-agent/pull/17286))
+- **Pluggable busy-indicator styles** (@OutThisLife, #13610) ([#17150](https://github.com/NousResearch/hermes-agent/pull/17150))
+- **Opt-in auto-resume of the most recent session** (@OutThisLife) ([#17130](https://github.com/NousResearch/hermes-agent/pull/17130))
+- **Expanded light-terminal auto-detection** — `HERMES_TUI_THEME` + background hex (@OutThisLife) ([#17113](https://github.com/NousResearch/hermes-agent/pull/17113))
+- **Delete sessions from `/resume` picker with `d`** (@OutThisLife) ([#17668](https://github.com/NousResearch/hermes-agent/pull/17668))
+- **Line-by-line scroll on modified mouse wheel** (@OutThisLife) ([#17669](https://github.com/NousResearch/hermes-agent/pull/17669))
+- **Delete queued message while editing with ctrl-x / cancel with esc** (@OutThisLife) ([#16707](https://github.com/NousResearch/hermes-agent/pull/16707))
+- **Per-section visibility for the details accordion** (@OutThisLife) ([#14968](https://github.com/NousResearch/hermes-agent/pull/14968))
+- **Voice mode CLI parity** — VAD loop + TTS + crash forensics ([#14810](https://github.com/NousResearch/hermes-agent/pull/14810))
+- **Contextual first-touch hints ported to TUI** — `/busy`, `/verbose` ([#16054](https://github.com/NousResearch/hermes-agent/pull/16054))
+- **Mini help menu on `?` in the input field** (@ethernet8023) ([#18043](https://github.com/NousResearch/hermes-agent/pull/18043))
+
+### Fixes
+- Fix: proactive mouse disable on ConPTY + `/mouse` toggle command (@kevin-ho, WSL2 ghost-mouse fix) ([#15488](https://github.com/NousResearch/hermes-agent/pull/15488))
+- Fix: restore skills search RPC ([#15870](https://github.com/NousResearch/hermes-agent/pull/15870))
+- Perf: cache text measurements across yoga flex re-passes ([#14818](https://github.com/NousResearch/hermes-agent/pull/14818))
+- Perf: stabilize long-session scrolling ([#15926](https://github.com/NousResearch/hermes-agent/pull/15926))
+- Perf: lazily seed virtual history heights ([#16523](https://github.com/NousResearch/hermes-agent/pull/16523))
+- Perf: cut visible cold start ~57% with lazy agent init ([#17190](https://github.com/NousResearch/hermes-agent/pull/17190))
+
+---
+
+## 🖱️ CLI & User Experience
+
+### New commands
+- **`hermes -z <prompt>`** — non-interactive one-shot mode ([#15702](https://github.com/NousResearch/hermes-agent/pull/15702))
+- **`hermes -z` with `--model` / `--provider` / `HERMES_INFERENCE_MODEL`** ([#15704](https://github.com/NousResearch/hermes-agent/pull/15704))
+- **`hermes update --check`** preflight flag ([#15841](https://github.com/NousResearch/hermes-agent/pull/15841))
+- **`hermes fallback`** command for managing fallback providers ([#16052](https://github.com/NousResearch/hermes-agent/pull/16052))
+- **`/busy`** slash command for busy input mode ([#15382](https://github.com/NousResearch/hermes-agent/pull/15382))
+- **`/busy` input mode 'steer'** as a third option ([#16279](https://github.com/NousResearch/hermes-agent/pull/16279))
+- **`/btw` as alias for `/background`** ([#16053](https://github.com/NousResearch/hermes-agent/pull/16053))
+- **`/reload-skills`** slash command (salvage #17670) ([#17744](https://github.com/NousResearch/hermes-agent/pull/17744))
+- **Surface `/queue`, `/bg`, `/steer` in agent-running placeholder** ([#16118](https://github.com/NousResearch/hermes-agent/pull/16118))
+
+### Setup / onboarding
+- **Auto-reconfigure on existing installs** ([#15879](https://github.com/NousResearch/hermes-agent/pull/15879))
+- **Contextual first-touch hints for `/busy` and `/verbose`** ([#16046](https://github.com/NousResearch/hermes-agent/pull/16046))
+- **Cost-saving tips from the April 30 tip-of-the-day** ([#17841](https://github.com/NousResearch/hermes-agent/pull/17841))
+- **Hyperlink startup banner title to the latest GitHub Release** ([#14945](https://github.com/NousResearch/hermes-agent/pull/14945))
+
+### Update / backup
+- **Snapshot pairing data before `git pull`** ([#16383](https://github.com/NousResearch/hermes-agent/pull/16383))
+- **Auto-backup HERMES_HOME before `hermes update`** (opt-in, off by default) ([#16539](https://github.com/NousResearch/hermes-agent/pull/16539), [#16566](https://github.com/NousResearch/hermes-agent/pull/16566))
+- **Exclude `checkpoints/` from backups** ([#16572](https://github.com/NousResearch/hermes-agent/pull/16572))
+- **Exclude SQLite WAL/SHM/journal sidecars from backups** ([#16576](https://github.com/NousResearch/hermes-agent/pull/16576))
+- **Installer FHS layout for root installs on Linux** ([#15608](https://github.com/NousResearch/hermes-agent/pull/15608))
+- Fix: kill stale dashboards instead of warning ([#17832](https://github.com/NousResearch/hermes-agent/pull/17832))
+- Fix: show correct update status on nix-built hermes ([#17550](https://github.com/NousResearch/hermes-agent/pull/17550))
+
+### Slash-command housekeeping
+- Refactor: drop `/provider`, `/plan` handler, and clean up slash registry ([#15047](https://github.com/NousResearch/hermes-agent/pull/15047))
+- Refactor: drop `persist_session` plumbing + fix broken `/btw` mid-turn bypass ([#16075](https://github.com/NousResearch/hermes-agent/pull/16075))
+
+### OpenClaw migration (for folks coming from OpenClaw)
+- **Hardened OpenClaw import** — plan-first apply, redaction, pre-migration backup ([#16911](https://github.com/NousResearch/hermes-agent/pull/16911))
+- Fix: case-preserving brand rewrite + one-time `~/.openclaw` residue banner ([#16327](https://github.com/NousResearch/hermes-agent/pull/16327))
+- Fix: resolve `openclaw` workspace files from `agents.defaults.workspace` ([#16879](https://github.com/NousResearch/hermes-agent/pull/16879))
+- Fix: resolve model aliases against real OpenClaw catalog schema (salvage #16778) ([#16977](https://github.com/NousResearch/hermes-agent/pull/16977))
+
+---
+
+## 📊 Web Dashboard
+
+- **Models tab** — rich per-model analytics ([#17745](https://github.com/NousResearch/hermes-agent/pull/17745))
+- **Configure main + auxiliary models from the Models page** ([#17802](https://github.com/NousResearch/hermes-agent/pull/17802))
+- **Dashboard Chat tab — xterm.js + JSON-RPC sidecar** (supersedes #12710 + #13379, @OutThisLife) ([#14890](https://github.com/NousResearch/hermes-agent/pull/14890))
+- **Dashboard layout refresh** (@austinpickett) ([#14899](https://github.com/NousResearch/hermes-agent/pull/14899))
+- **`--stop` and `--status` flags** on the dashboard CLI ([#17840](https://github.com/NousResearch/hermes-agent/pull/17840))
+- **Page-scoped plugin slots for built-in pages** ([#15658](https://github.com/NousResearch/hermes-agent/pull/15658))
+- Fix: replace all buttons for design system buttons ([#17007](https://github.com/NousResearch/hermes-agent/pull/17007))
+
+---
+
+## ⚡ Performance
+
+- **TUI visible cold start cut ~57%** via lazy agent init ([#17190](https://github.com/NousResearch/hermes-agent/pull/17190))
+- **Lazy-import OpenAI, Anthropic, Firecrawl, account_usage** ([#17046](https://github.com/NousResearch/hermes-agent/pull/17046))
+- **mtime-cache `load_config()` and `read_raw_config()`** ([#17041](https://github.com/NousResearch/hermes-agent/pull/17041))
+- **Memoize `get_tool_definitions()` + TTL-cache `check_fn` results** ([#17098](https://github.com/NousResearch/hermes-agent/pull/17098))
+- **Precompile DANGEROUS_PATTERNS and HARDLINE_PATTERNS** ([#17206](https://github.com/NousResearch/hermes-agent/pull/17206))
+- **Cache Ink text measurements across yoga flex re-passes** ([#14818](https://github.com/NousResearch/hermes-agent/pull/14818))
+- **Stabilize long-session scrolling** ([#15926](https://github.com/NousResearch/hermes-agent/pull/15926))
+- **Lazily seed virtual history heights** ([#16523](https://github.com/NousResearch/hermes-agent/pull/16523))
+
+---
+
+## 🔒 Security & Reliability
+
+- **Secret redaction off by default** — stops corrupting patches / API payloads with fake-key substitutions. Opt in via `redaction.enabled: true` ([#16794](https://github.com/NousResearch/hermes-agent/pull/16794))
+- **`[SYSTEM:` → `[IMPORTANT:`** in all user-injected markers (Azure content filter dodge) ([#16114](https://github.com/NousResearch/hermes-agent/pull/16114))
+- **Hardline blocklist for unrecoverable commands** ([#15878](https://github.com/NousResearch/hermes-agent/pull/15878))
+- **Canonical `mask_secret` helper; fix status.py DIM drift** ([#17207](https://github.com/NousResearch/hermes-agent/pull/17207))
+- **Sweep expired paste.rs uploads on a real timer** ([#16431](https://github.com/NousResearch/hermes-agent/pull/16431))
+- **Preserve symlinks during atomic file writes** ([#16980](https://github.com/NousResearch/hermes-agent/pull/16980))
+- **Probe `/dev/tty` by opening it, not bare existence** ([#17024](https://github.com/NousResearch/hermes-agent/pull/17024))
+
+---
+
+## 🐛 Notable Bug Fixes
+
+This window includes 360 `fix:` PRs. Selected highlights from across the stack:
+
+- **Background review fork inherits parent's live runtime** — provider/model/creds now propagate correctly ([#16099](https://github.com/NousResearch/hermes-agent/pull/16099))
+- **Hindsight configurable `HINDSIGHT_TIMEOUT` env var** ([#15077](https://github.com/NousResearch/hermes-agent/pull/15077))
+- **Tools: normalize numeric entries + clear stale `no_mcp` in `_save_platform_tools`** ([#15607](https://github.com/NousResearch/hermes-agent/pull/15607))
+- **MCP: rewrite `definitions` refs to `$defs` in input schemas** — closes provider-side 400s
+- **Azure content filter compatibility** — renamed `[SYSTEM:` markers so Azure's content filter stops flagging them ([#16114](https://github.com/NousResearch/hermes-agent/pull/16114))
+- **Vision cache uses HERMES_HOME instead of cwd** ([#17719](https://github.com/NousResearch/hermes-agent/pull/17719))
+- **FTS5 search** — tool_name + tool_calls indexing with repair + migration ([#16914](https://github.com/NousResearch/hermes-agent/pull/16914))
+- **Streaming reasoning persists on assistant turns** ([#16892](https://github.com/NousResearch/hermes-agent/pull/16892))
+- **execute_code concurrent RPC serialization** (#17770) ([#17894](https://github.com/NousResearch/hermes-agent/pull/17894), [#17902](https://github.com/NousResearch/hermes-agent/pull/17902))
+- **Background reviewer scoped to memory + skills toolsets** — no more accidental web/shell escapes ([#16569](https://github.com/NousResearch/hermes-agent/pull/16569))
+- **Compression recovery** — retry on main before giving up; notify user when aux fails ([#16774](https://github.com/NousResearch/hermes-agent/pull/16774), [#16775](https://github.com/NousResearch/hermes-agent/pull/16775))
+- **`croniter` promoted to a core dependency** ([#17577](https://github.com/NousResearch/hermes-agent/pull/17577))
+- **Discord tool `limit` parameter coerced to int** before `min()` call ([#16319](https://github.com/NousResearch/hermes-agent/pull/16319))
+- **Yuanbao messaging platform entrance fix** ([#16880](https://github.com/NousResearch/hermes-agent/pull/16880))
+- **ACP advertise and forward image prompts** ([#18030](https://github.com/NousResearch/hermes-agent/pull/18030))
+- **DeepSeek / Kimi reasoning content isolation** across cross-provider histories (@Zjianru) ([#15749](https://github.com/NousResearch/hermes-agent/pull/15749), [#15762](https://github.com/NousResearch/hermes-agent/pull/15762))
+- **Preserve reasoning_content replay on DeepSeek v4 + Kimi/Moonshot thinking** ([#18045](https://github.com/NousResearch/hermes-agent/pull/18045))
+
+The vast majority of the 360 fixes landed in the streaming/compression/tool-calling paths across all providers — DeepSeek, Kimi, Moonshot, GLM, Qwen, MiniMax, Gemini, Anthropic, OpenAI — alongside TUI polish (resize, scroll, sticky-prompt) and gateway platform-specific edge cases.
+
+---
+
+## 🧪 Testing & CI
+
+- Hermetic test parity (`scripts/run_tests.sh`) held across this window
+- **Microsoft Teams xdist collision guard** — prevents worker collisions when Teams platform tests run in parallel ([#17828](https://github.com/NousResearch/hermes-agent/pull/17828))
+- Chore: remove unused imports and dead locals (ruff F401, F841) ([#17010](https://github.com/NousResearch/hermes-agent/pull/17010))
+
+---
+
+## 📚 Documentation
+
+- **Curator feature page** added to docs site ([#17563](https://github.com/NousResearch/hermes-agent/pull/17563))
+- **Document pin also blocking `skill_manage` writes** ([#17578](https://github.com/NousResearch/hermes-agent/pull/17578))
+- **Direct-URL skill install documented** across features, reference, guide, and `hermes-agent` skill ([#16355](https://github.com/NousResearch/hermes-agent/pull/16355))
+- **Hooks tutorial — build a BOOT.md startup checklist** (replaces the removed built-in hook) ([#17202](https://github.com/NousResearch/hermes-agent/pull/17202))
+- **ComfyUI docs: ask local vs cloud FIRST before hardware check** ([#17612](https://github.com/NousResearch/hermes-agent/pull/17612))
+- **Obliteratus skill: link YouTube video guide in SKILL.md** ([#15808](https://github.com/NousResearch/hermes-agent/pull/15808))
+- Per-skill docs pages generated for bundled + optional skills; ASCII art code blocks auto-wrapped ([#14929](https://github.com/NousResearch/hermes-agent/pull/14929), [#16497](https://github.com/NousResearch/hermes-agent/pull/16497))
+
+---
+
+## ⚖️ Removed / Reverted
+
+- **Kanban multi-profile collaboration board** — landed in #16081, reverted in ([#16098](https://github.com/NousResearch/hermes-agent/pull/16098)) while the design is reworked
+- **computer-use cua-driver** — 3 preparatory PRs landed then were reverted in ([#16927](https://github.com/NousResearch/hermes-agent/pull/16927))
+- **BOOT.md built-in hook** removed ([#17093](https://github.com/NousResearch/hermes-agent/pull/17093)); the hooks tutorial ([#17202](https://github.com/NousResearch/hermes-agent/pull/17202)) shows how to build the same workflow yourself with a shell hook
+- **`/provider` + `/plan` slash commands dropped** ([#15047](https://github.com/NousResearch/hermes-agent/pull/15047))
+- **`flush_memories` removed entirely** ([#15696](https://github.com/NousResearch/hermes-agent/pull/15696))
+
+---
+
+## 👥 Contributors
+
+### Core
+- **@teknium1** (Teknium)
+
+### Top Community Contributors (by merged PR count since v0.11.0)
+
+- **@OutThisLife** (Brooklyn) — 52 PRs · TUI — light-terminal detection + pluggable busy styles + auto-resume + session-delete from /resume + mouse-wheel scrolling + xterm.js dashboard Chat tab + cold-start cut + accordion polish
+- **@kshitijk4poor** — 12 PRs · LM Studio first-class provider (salvage), Vercel Sandbox backend, GMI Cloud salvage, bundled-by-default touchdesigner-mcp, many tool-call / reasoning fixes
+- **@helix4u** — 10 PRs · MCP schema robustness, assorted stability fixes
+- **@alt-glitch** — 8 PRs · trigram FTS5 CJK search, declarative Nix plugin install, matrix/feishu hints and fixes
+- **@ethernet8023** — 4 PRs
+- **@austinpickett** — 4 PRs · LaTeX rendering in TUI, dashboard layout refresh
+- **@benbarclay** — 3 PRs · Docker run-as-host-user so bind mounts don't get root-owned
+- **@vominh1919** — 2 PRs
+- **@stephenschoettler** — 2 PRs
+- **@kevin-ho** — ConPTY mouse-injection fix (#15488)
+- **@Zjianru** — cross-provider reasoning_content isolation + DeepSeek/Kimi empty-reasoning injection (#15749, #15762)
+- **@web3blind** — Telegram chat allowlists for groups and forums (#15027)
+- **@SHL0MS** — 9 new TouchDesigner-MCP reference docs (#16768)
+- **@0xDevNinja** — curator `restore_skill` nested-archive fix (#17951)
+- **@y0shua1ee** — curator `use` activity fix (#17953)
+
+### Also contributing
+Salvaged or co-authored work from **@isaachuangGMICLOUD** (GMI Cloud), earlier upstream PRs from the original author of each salvage chain, and a long tail of one-shot fixes, documentation nudges, and skill contributions from the community.
+
+### All Contributors (alphabetical, excluding @teknium1)
+
+@0xbyt4, @0xharryriddle, @0xDevNinja, @0z1-ghb, @5park1e, @A-FdL-Prog, @aj-nt, @akhater, @alblez, @alexg0bot,
+@alexzhu0, @AllardQuek, @alt-glitch, @amanning3390, @amanuel2, @AndreKurait, @andrewhosf, @Andy283, @andyylin,
+@angel12, @AntAISecurityLab, @ash, @austinpickett, @badgerbees, @BadTechBandit, @Bartok9, @beenherebefore,
+@beesrsj2500, @BeliefanX, @benbarclay, @benjaminsehl, @BlackishGreen33, @bloodcarter, @BlueBirdBack,
+@briandevans, @brooklynnicholson, @bsgdigital, @buray, @bwjoke, @camaragon, @cdanis, @cgarwood82,
+@charles-brooks, @chen1749144759, @chengoak, @ching-kaching, @Contentment003111, @crayfish-ai, @CruxExperts,
+@cyclingwithelephants, @dandaka, @danklynn, @ddupont808, @dhabibi, @difujia, @dimitrovi, @dlkakbs,
+@dontcallmejames, @EKKOLearnAI, @emozilla, @ericnicolaides, @Erosika, @ethernet8023, @exiao, @Feranmi10,
+@flobo3, @foxion37, @georgeglessner, @georgex8001, @ghostmfr, @H-Ali13381, @HangGlidersRule, @harryplusplus,
+@haru398801, @heathley, @hejuntt1014, @hekaru-agent, @helix4u, @Heltman, @HenkDz, @heyitsaamir, @hharry11,
+@hhhonzik, @hhuang91, @HiddenPuppy, @htsh, @iamagenius00, @in-liberty420, @innocarpe, @irispillars, @iRonin,
+@isaachuangGMICLOUD, @Ito-69, @j3ffffff, @jackjin1997, @jakubkrcmar, @Jason2031, @JayGwod, @jerome-benoit,
+@johnncenae, @Kailigithub, @keiravoss94, @kevin-ho, @knockyai, @konsisumer, @kshitijk4poor, @kunlabs, @l0hde,
+@Leihb, @leoneparise, @LeonSGP43, @liizfq, @liuhao1024, @loongzhao, @lsdsjy, @luyao618, @ma-pony, @Magaav,
+@MagicRay1217, @math0r-be, @MattMaximo, @maxims-oss, @MaxyMoos, @maymuneth, @mcndjxlefnd, @memosr,
+@MestreY0d4-Uninter, @mewwts, @Mirac1eSky, @MorAlekss, @mrhwick, @mrunmayee17, @mssteuer, @Nanako0129,
+@nazirulhafiy, @Nerijusas, @Nicecsh, @nicoloboschi, @nightq, @ningfangbin, @octo-patch, @Octopus,
+@OutThisLife, @Paperclip, @pein892, @perlowja, @prasadus92, @qike-ms, @qiyin-code, @Readon, @ReginaldasR,
+@revaraver, @rfilgueiras, @rmoen, @romanornr, @rugvedS07, @rylena, @samrusani, @Sanjays2402, @sasha-id,
+@Satoshi-agi, @scheidti, @scotttrinh, @season179, @SeeYangZhi, @sgaofen, @shamork, @shannonsands, @SHL0MS,
+@simbam99, @Societus, @socrates1024, @Sonoyunchu, @sprmn24, @stephenschoettler, @tangyuanjc, @TechPrototyper,
+@tekgnosis-net, @ThomassJonax, @tmimmanuel, @tochukwuada, @Tosko4, @Tranquil-Flow, @twozle, @txbxxx,
+@UgwujaGeorge, @Versun, @vlwkaos, @voidborne-d, @vominh1919, @Wang-tianhao, @Wangshengyang2004, @web3blind,
+@westers, @Wysie, @xandersbell, @xiahu88988, @XieNBi, @xinbenlv, @xnbi, @y0shua1ee, @yatesjalex, @yes999zc,
+@yeyitech, @Yoimex, @YueLich, @Yukipukii1, @zhiyanliu, @zicochaos, @Zjianru, @zkl2333, @zons-zhaozhy,
+@ztexydt-cqh.
+
+Also: @Siddharth Balyan, @YuShu.
+
+---
+
+**Full Changelog**: [v2026.4.23...v2026.4.30](https://github.com/NousResearch/hermes-agent/compare/v2026.4.23...v2026.4.30)

acp_adapter/__main__.py

@@ -0,0 +1,5 @@
+"""Allow running the ACP adapter as ``python -m acp_adapter``."""
+
+from .entry import main
+
+main()

acp_adapter/entry.py

@@ -6,7 +6,7 @@ and starts the ACP agent server.
 
 Usage::
 
-    python -m hermes_agent.acp.entry
+    python -m acp_adapter.entry
     # or
     hermes acp
     # or
@@ -16,6 +16,7 @@ Usage::
 import asyncio
 import logging
 import sys
+from pathlib import Path
 from hermes_constants import get_hermes_home
 
 
@@ -103,8 +104,24 @@ def main() -> None:
     logger = logging.getLogger(__name__)
     logger.info("Starting hermes-agent ACP adapter")
 
+    # Ensure the project root is on sys.path so ``from run_agent import AIAgent`` works
+    project_root = str(Path(__file__).resolve().parent.parent)
+    if project_root not in sys.path:
+        sys.path.insert(0, project_root)
+
     import acp
-    from hermes_agent.acp.server import HermesACPAgent
+    from .server import HermesACPAgent
+
+    # MCP tool discovery from config.yaml — run before asyncio.run() so
+    # it's safe to use blocking waits.  (ACP also registers per-session
+    # MCP servers dynamically via asyncio.to_thread inside the event
+    # loop; that path is unaffected.)  Moved from model_tools.py module
+    # scope to avoid freezing the gateway's loop on lazy import (#16856).
+    try:
+        from tools.mcp_tool import discover_mcp_tools
+        discover_mcp_tools()
+    except Exception:
+        logger.debug("MCP tool discovery failed at ACP startup", exc_info=True)
 
     agent = HermesACPAgent()
     try:

acp_adapter/events.py

@@ -15,7 +15,7 @@ from typing import Any, Callable, Deque, Dict
 
 import acp
 
-from hermes_agent.acp.tools import (
+from .tools import (
     build_tool_complete,
     build_tool_start,
     make_tool_call_id,

acp_adapter/server.py

@@ -3,6 +3,8 @@
 from __future__ import annotations
 
 import asyncio
+import contextvars
+import json
 import logging
 import os
 from collections import defaultdict, deque
@@ -12,6 +14,7 @@ from typing import Any, Deque, Optional
 import acp
 from acp.schema import (
     AgentCapabilities,
+    AgentMessageChunk,
     AuthenticateResponse,
     AvailableCommand,
     AvailableCommandsUpdate,
@@ -29,6 +32,7 @@ from acp.schema import (
     McpServerStdio,
     ModelInfo,
     NewSessionResponse,
+    PromptCapabilities,
     PromptResponse,
     ResumeSessionResponse,
     SetSessionConfigOptionResponse,
@@ -44,6 +48,8 @@ from acp.schema import (
     TextContentBlock,
     UnstructuredCommandInput,
     Usage,
+    UsageUpdate,
+    UserMessageChunk,
 )
 
 # AuthMethodAgent was renamed from AuthMethod in agent-client-protocol 0.9.0
@@ -52,15 +58,16 @@ try:
 except ImportError:
     from acp.schema import AuthMethod as AuthMethodAgent  # type: ignore[attr-defined]
 
-from hermes_agent.acp.auth import detect_provider
-from hermes_agent.acp.events import (
+from acp_adapter.auth import detect_provider
+from acp_adapter.events import (
     make_message_cb,
     make_step_cb,
     make_thinking_cb,
     make_tool_progress_cb,
 )
-from hermes_agent.acp.permissions import make_approval_callback
-from hermes_agent.acp.session import SessionManager, SessionState
+from acp_adapter.permissions import make_approval_callback
+from acp_adapter.session import SessionManager, SessionState, _expand_acp_enabled_toolsets
+from acp_adapter.tools import build_tool_complete, build_tool_start
 
 logger = logging.getLogger(__name__)
 
@@ -87,17 +94,69 @@ def _extract_text(
         | EmbeddedResourceContentBlock
     ],
 ) -> str:
-    """Extract plain text from ACP content blocks."""
+    """Extract plain text from ACP content blocks for display/commands."""
     parts: list[str] = []
     for block in prompt:
         if isinstance(block, TextContentBlock):
             parts.append(block.text)
         elif hasattr(block, "text"):
             parts.append(str(block.text))
-        # Non-text blocks are ignored for now.
     return "\n".join(parts)
 
 
+def _image_block_to_openai_part(block: ImageContentBlock) -> dict[str, Any] | None:
+    """Convert an ACP image content block to OpenAI-style multimodal content."""
+    data = str(getattr(block, "data", "") or "").strip()
+    uri = str(getattr(block, "uri", "") or "").strip()
+    mime_type = str(getattr(block, "mime_type", "") or "image/png").strip() or "image/png"
+
+    if data:
+        url = data if data.startswith("data:") else f"data:{mime_type};base64,{data}"
+    elif uri:
+        url = uri
+    else:
+        return None
+
+    return {"type": "image_url", "image_url": {"url": url}}
+
+
+def _content_blocks_to_openai_user_content(
+    prompt: list[
+        TextContentBlock
+        | ImageContentBlock
+        | AudioContentBlock
+        | ResourceContentBlock
+        | EmbeddedResourceContentBlock
+    ],
+) -> str | list[dict[str, Any]]:
+    """Convert ACP prompt blocks into a Hermes/OpenAI-compatible user content payload."""
+    parts: list[dict[str, Any]] = []
+    text_parts: list[str] = []
+
+    for block in prompt:
+        if isinstance(block, TextContentBlock):
+            if block.text:
+                parts.append({"type": "text", "text": block.text})
+                text_parts.append(block.text)
+            continue
+        if isinstance(block, ImageContentBlock):
+            image_part = _image_block_to_openai_part(block)
+            if image_part is not None:
+                parts.append(image_part)
+            continue
+
+    if not parts:
+        return _extract_text(prompt)
+
+    # Keep pure text prompts as strings so slash-command handling and text-only
+    # providers keep the exact legacy path. Switch to structured content only
+    # when an actual non-text block is present.
+    if all(part.get("type") == "text" for part in parts):
+        return "\n".join(text_parts)
+
+    return parts
+
+
 class HermesACPAgent(acp.Agent):
     """ACP Agent implementation wrapping Hermes AIAgent."""
 
@@ -108,6 +167,8 @@ class HermesACPAgent(acp.Agent):
         "context": "Show conversation context info",
         "reset": "Clear conversation history",
         "compact": "Compress conversation context",
+        "steer": "Inject guidance into the currently running agent turn",
+        "queue": "Queue a prompt to run after the current turn finishes",
         "version": "Show Hermes version",
     }
 
@@ -137,6 +198,16 @@ class HermesACPAgent(acp.Agent):
             "name": "compact",
             "description": "Compress conversation context",
         },
+        {
+            "name": "steer",
+            "description": "Inject guidance into the currently running agent turn",
+            "input_hint": "guidance for the active turn",
+        },
+        {
+            "name": "queue",
+            "description": "Queue a prompt to run after the current turn finishes",
+            "input_hint": "prompt to run next",
+        },
         {
             "name": "version",
             "description": "Show Hermes version",
@@ -247,6 +318,66 @@ class HermesACPAgent(acp.Agent):
 
         return target_provider, new_model
 
+    @staticmethod
+    def _build_usage_update(state: SessionState) -> UsageUpdate | None:
+        """Build ACP native context-usage data for clients like Zed.
+
+        Zed's circular context indicator is driven by ACP ``usage_update``
+        session updates: ``size`` is the model context window and ``used`` is
+        the current request pressure.  Hermes estimates ``used`` from the same
+        buckets it sends to providers: system prompt, conversation history, and
+        tool schemas.
+        """
+        agent = state.agent
+        compressor = getattr(agent, "context_compressor", None)
+        size = int(getattr(compressor, "context_length", 0) or 0)
+        if size <= 0:
+            return None
+
+        try:
+            from agent.model_metadata import estimate_request_tokens_rough
+
+            used = estimate_request_tokens_rough(
+                state.history,
+                system_prompt=getattr(agent, "_cached_system_prompt", "") or "",
+                tools=getattr(agent, "tools", None) or None,
+            )
+        except Exception:
+            logger.debug("Could not estimate ACP native context usage", exc_info=True)
+            used = int(getattr(compressor, "last_prompt_tokens", 0) or 0)
+
+        return UsageUpdate(
+            session_update="usage_update",
+            size=max(size, 0),
+            used=max(used, 0),
+        )
+
+    async def _send_usage_update(self, state: SessionState) -> None:
+        """Send ACP native context usage to the connected client."""
+        if not self._conn:
+            return
+        update = self._build_usage_update(state)
+        if update is None:
+            return
+        try:
+            await self._conn.session_update(
+                session_id=state.session_id,
+                update=update,
+            )
+        except Exception:
+            logger.warning(
+                "Failed to send ACP usage update for session %s",
+                state.session_id,
+                exc_info=True,
+            )
+
+    def _schedule_usage_update(self, state: SessionState) -> None:
+        """Schedule native context indicator refresh after ACP responses."""
+        if not self._conn:
+            return
+        loop = asyncio.get_running_loop()
+        loop.call_soon(asyncio.create_task, self._send_usage_update(state))
+
     async def _register_session_mcp_servers(
         self,
         state: SessionState,
@@ -287,7 +418,11 @@ class HermesACPAgent(acp.Agent):
         try:
             from model_tools import get_tool_definitions
 
-            enabled_toolsets = getattr(state.agent, "enabled_toolsets", None) or ["hermes-acp"]
+            enabled_toolsets = _expand_acp_enabled_toolsets(
+                getattr(state.agent, "enabled_toolsets", None) or ["hermes-acp"],
+                mcp_server_names=[server.name for server in mcp_servers],
+            )
+            state.agent.enabled_toolsets = enabled_toolsets
             disabled_toolsets = getattr(state.agent, "disabled_toolsets", None)
             state.agent.tools = get_tool_definitions(
                 enabled_toolsets=enabled_toolsets,
@@ -347,6 +482,7 @@ class HermesACPAgent(acp.Agent):
             agent_info=Implementation(name="hermes-agent", version=HERMES_VERSION),
             agent_capabilities=AgentCapabilities(
                 load_session=True,
+                prompt_capabilities=PromptCapabilities(image=True),
                 session_capabilities=SessionCapabilities(
                     fork=SessionForkCapabilities(),
                     list=SessionListCapabilities(),
@@ -372,6 +508,140 @@ class HermesACPAgent(acp.Agent):
 
     # ---- Session management -------------------------------------------------
 
+    @staticmethod
+    def _history_message_text(message: dict[str, Any]) -> str:
+        """Extract displayable text from a persisted OpenAI-style message."""
+        content = message.get("content")
+        if isinstance(content, str):
+            return content.strip()
+        if isinstance(content, list):
+            parts: list[str] = []
+            for item in content:
+                if isinstance(item, dict):
+                    text = item.get("text")
+                    if isinstance(text, str):
+                        parts.append(text)
+                    elif item.get("type") == "text" and isinstance(item.get("content"), str):
+                        parts.append(item["content"])
+                elif isinstance(item, str):
+                    parts.append(item)
+            return "\n".join(part.strip() for part in parts if part and part.strip()).strip()
+        return ""
+
+    @staticmethod
+    def _history_message_update(
+        *,
+        role: str,
+        text: str,
+    ) -> UserMessageChunk | AgentMessageChunk | None:
+        """Build an ACP history replay update for a user/assistant message."""
+        block = TextContentBlock(type="text", text=text)
+        if role == "user":
+            return UserMessageChunk(
+                session_update="user_message_chunk",
+                content=block,
+            )
+        if role == "assistant":
+            return AgentMessageChunk(
+                session_update="agent_message_chunk",
+                content=block,
+            )
+        return None
+
+    @staticmethod
+    def _history_tool_call_name_args(tool_call: dict[str, Any]) -> tuple[str, dict[str, Any]]:
+        """Extract function name/arguments from an OpenAI-style tool_call."""
+        function = tool_call.get("function") if isinstance(tool_call.get("function"), dict) else {}
+        name = str(function.get("name") or tool_call.get("name") or "unknown_tool")
+        raw_args = function.get("arguments") or tool_call.get("arguments") or tool_call.get("args") or {}
+        if isinstance(raw_args, str):
+            try:
+                parsed = json.loads(raw_args)
+            except Exception:
+                parsed = {"raw": raw_args}
+            raw_args = parsed
+        if not isinstance(raw_args, dict):
+            raw_args = {}
+        return name, raw_args
+
+    @staticmethod
+    def _history_tool_call_id(tool_call: dict[str, Any]) -> str:
+        """Return the stable provider tool call id for ACP history replay."""
+        return str(
+            tool_call.get("id")
+            or tool_call.get("call_id")
+            or tool_call.get("tool_call_id")
+            or ""
+        ).strip()
+
+    async def _replay_session_history(self, state: SessionState) -> None:
+        """Send persisted user/assistant history to clients during session/load.
+
+        Zed's ACP history UI calls ``session/load`` after the user picks an item
+        from the Agents sidebar. The agent must then replay the full conversation
+        as user/assistant chunks plus reconstructed tool-call start/completion
+        notifications; merely restoring server-side state makes Hermes remember
+        context, but leaves the editor looking like a clean thread.
+        """
+        if not self._conn or not state.history:
+            return
+
+        active_tool_calls: dict[str, tuple[str, dict[str, Any]]] = {}
+
+        async def _send(update: Any) -> bool:
+            try:
+                await self._conn.session_update(session_id=state.session_id, update=update)
+                return True
+            except Exception:
+                logger.warning(
+                    "Failed to replay ACP history for session %s",
+                    state.session_id,
+                    exc_info=True,
+                )
+                return False
+
+        for message in state.history:
+            role = str(message.get("role") or "")
+
+            if role in {"user", "assistant"}:
+                text = self._history_message_text(message)
+                if text:
+                    update = self._history_message_update(role=role, text=text)
+                    if update is not None and not await _send(update):
+                        return
+
+            if role == "assistant" and isinstance(message.get("tool_calls"), list):
+                for tool_call in message["tool_calls"]:
+                    if not isinstance(tool_call, dict):
+                        continue
+                    tool_call_id = self._history_tool_call_id(tool_call)
+                    if not tool_call_id:
+                        continue
+                    tool_name, args = self._history_tool_call_name_args(tool_call)
+                    active_tool_calls[tool_call_id] = (tool_name, args)
+                    if not await _send(build_tool_start(tool_call_id, tool_name, args)):
+                        return
+                continue
+
+            if role == "tool":
+                tool_call_id = str(message.get("tool_call_id") or "").strip()
+                tool_name = str(message.get("tool_name") or "").strip()
+                function_args: dict[str, Any] | None = None
+                if tool_call_id in active_tool_calls:
+                    tool_name, function_args = active_tool_calls.pop(tool_call_id)
+                if not tool_call_id or not tool_name:
+                    continue
+                result = message.get("content")
+                if not await _send(
+                    build_tool_complete(
+                        tool_call_id,
+                        tool_name,
+                        result=result if isinstance(result, str) else None,
+                        function_args=function_args,
+                    )
+                ):
+                    return
+
     async def new_session(
         self,
         cwd: str,
@@ -382,11 +652,24 @@ class HermesACPAgent(acp.Agent):
         await self._register_session_mcp_servers(state, mcp_servers)
         logger.info("New session %s (cwd=%s)", state.session_id, cwd)
         self._schedule_available_commands_update(state.session_id)
+        self._schedule_usage_update(state)
         return NewSessionResponse(
             session_id=state.session_id,
             models=self._build_model_state(state),
         )
 
+    def _schedule_history_replay(self, state: SessionState) -> None:
+        """Replay persisted history after session/load or session/resume returns.
+
+        Zed only attaches streamed transcript/tool updates once the load/resume
+        response has completed. Sending replay notifications while the request is
+        still in-flight can make the server look correct in logs while the editor
+        drops or fails to attach the tool-call history.
+        """
+        loop = asyncio.get_running_loop()
+        replay_coro = self._replay_session_history(state)
+        loop.call_soon(asyncio.create_task, replay_coro)
+
     async def load_session(
         self,
         cwd: str,
@@ -400,7 +683,9 @@ class HermesACPAgent(acp.Agent):
             return None
         await self._register_session_mcp_servers(state, mcp_servers)
         logger.info("Loaded session %s", session_id)
+        self._schedule_history_replay(state)
         self._schedule_available_commands_update(session_id)
+        self._schedule_usage_update(state)
         return LoadSessionResponse(models=self._build_model_state(state))
 
     async def resume_session(
@@ -416,12 +701,17 @@ class HermesACPAgent(acp.Agent):
             state = self.session_manager.create_session(cwd=cwd)
         await self._register_session_mcp_servers(state, mcp_servers)
         logger.info("Resumed session %s", state.session_id)
+        self._schedule_history_replay(state)
         self._schedule_available_commands_update(state.session_id)
+        self._schedule_usage_update(state)
         return ResumeSessionResponse(models=self._build_model_state(state))
 
     async def cancel(self, session_id: str, **kwargs: Any) -> None:
         state = self.session_manager.get_session(session_id)
         if state and state.cancel_event:
+            with state.runtime_lock:
+                if state.is_running and state.current_prompt_text:
+                    state.interrupted_prompt_text = state.current_prompt_text
             state.cancel_event.set()
             try:
                 if getattr(state, "agent", None) and hasattr(state.agent, "interrupt"):
@@ -512,18 +802,77 @@ class HermesACPAgent(acp.Agent):
             return PromptResponse(stop_reason="refusal")
 
         user_text = _extract_text(prompt).strip()
-        if not user_text:
+        user_content = _content_blocks_to_openai_user_content(prompt)
+        has_content = bool(user_text) or (
+            isinstance(user_content, list) and bool(user_content)
+        )
+        if not has_content:
             return PromptResponse(stop_reason="end_turn")
 
-        # Intercept slash commands — handle locally without calling the LLM
-        if user_text.startswith("/"):
+        # /steer on an idle session has no in-flight tool call to inject into.
+        # Rewrite it so the payload runs as a normal user prompt, matching the
+        # gateway's behavior (gateway/run.py ~L4898). Two sub-cases:
+        #   1. Zed-interrupt salvage — a prior prompt was cancelled by the
+        #      client right before /steer arrived; replay it with the steer
+        #      text attached as explicit correction/guidance so the user's
+        #      in-flight work isn't lost.
+        #   2. Plain idle — no prior work to salvage; just run the steer
+        #      payload as a regular prompt. Without this, _cmd_steer would
+        #      silently append to state.queued_prompts and respond with
+        #      "No active turn — queued for the next turn", which looks like
+        #      /queue even though the user never typed /queue.
+        if isinstance(user_content, str) and user_text.startswith("/steer"):
+            steer_text = user_text.split(maxsplit=1)[1].strip() if len(user_text.split(maxsplit=1)) > 1 else ""
+            interrupted_prompt = ""
+            rewrite_idle = False
+            with state.runtime_lock:
+                if not state.is_running and steer_text:
+                    if state.interrupted_prompt_text:
+                        interrupted_prompt = state.interrupted_prompt_text
+                        state.interrupted_prompt_text = ""
+                    else:
+                        rewrite_idle = True
+            if interrupted_prompt:
+                user_text = (
+                    f"{interrupted_prompt}\n\n"
+                    f"User correction/guidance after interrupt: {steer_text}"
+                )
+                user_content = user_text
+            elif rewrite_idle:
+                user_text = steer_text
+                user_content = steer_text
+
+        # Intercept slash commands — handle locally without calling the LLM.
+        # Slash commands are text-only; if the client included images/resources,
+        # send the whole multimodal prompt to the agent instead of treating it as
+        # an ACP command.
+        if isinstance(user_content, str) and user_text.startswith("/"):
             response_text = self._handle_slash_command(user_text, state)
             if response_text is not None:
                 if self._conn:
                     update = acp.update_agent_message_text(response_text)
                     await self._conn.session_update(session_id, update)
+                    await self._send_usage_update(state)
                 return PromptResponse(stop_reason="end_turn")
 
+        # If Zed sends another regular prompt while the same ACP session is
+        # still running, queue it instead of racing two AIAgent loops against
+        # the same state.history. /steer and /queue are handled above and can
+        # land immediately.
+        with state.runtime_lock:
+            if state.is_running:
+                queued_text = user_text or "[Image attachment]"
+                state.queued_prompts.append(queued_text)
+                depth = len(state.queued_prompts)
+                if self._conn:
+                    update = acp.update_agent_message_text(
+                        f"Queued for the next turn. ({depth} queued)"
+                    )
+                    await self._conn.session_update(session_id, update)
+                return PromptResponse(stop_reason="end_turn")
+            state.is_running = True
+            state.current_prompt_text = user_text or "[Image attachment]"
+
         logger.info("Prompt on session %s: %s", session_id, user_text[:100])
 
         conn = self._conn
@@ -536,24 +885,37 @@ class HermesACPAgent(acp.Agent):
         tool_call_meta: dict[str, dict[str, Any]] = {}
         previous_approval_cb = None
 
+        streamed_message = False
+
         if conn:
             tool_progress_cb = make_tool_progress_cb(conn, session_id, loop, tool_call_ids, tool_call_meta)
-            thinking_cb = make_thinking_cb(conn, session_id, loop)
+            reasoning_cb = make_thinking_cb(conn, session_id, loop)
             step_cb = make_step_cb(conn, session_id, loop, tool_call_ids, tool_call_meta)
             message_cb = make_message_cb(conn, session_id, loop)
+
+            def stream_delta_cb(text: str) -> None:
+                nonlocal streamed_message
+                if text:
+                    streamed_message = True
+                message_cb(text)
+
             approval_cb = make_approval_callback(conn.request_permission, loop, session_id)
         else:
             tool_progress_cb = None
-            thinking_cb = None
+            reasoning_cb = None
             step_cb = None
-            message_cb = None
+            stream_delta_cb = None
             approval_cb = None
 
         agent = state.agent
         agent.tool_progress_callback = tool_progress_cb
-        agent.thinking_callback = thinking_cb
+        # ACP thought panes should not receive Hermes' local kawaii waiting/status
+        # updates. Route provider/model reasoning deltas instead; if the provider
+        # emits no reasoning, Zed should not get a fake "thinking" accordion.
+        agent.thinking_callback = None
+        agent.reasoning_callback = reasoning_cb
         agent.step_callback = step_cb
-        agent.message_callback = message_cb
+        agent.stream_delta_callback = stream_delta_cb
 
         # Approval callback is per-thread (thread-local, GHSA-qg5c-hvr5-hjgr).
         # Set it INSIDE _run_agent so the TLS write happens in the executor
@@ -570,6 +932,22 @@ class HermesACPAgent(acp.Agent):
 
         def _run_agent() -> dict:
             nonlocal previous_approval_cb, previous_interactive
+            # Bind HERMES_SESSION_KEY for this session so per-session caches
+            # (e.g. the interactive sudo password cache in tools.terminal_tool)
+            # scope to the ACP session rather than leaking across sessions
+            # that land on the same reused executor thread. This call runs
+            # inside a contextvars.copy_context() below, so the ContextVar
+            # write is isolated from other concurrent ACP sessions.
+            try:
+                from gateway.session_context import (
+                    clear_session_vars,
+                    set_session_vars,
+                )
+                session_tokens = set_session_vars(session_key=session_id)
+            except Exception:
+                session_tokens = None
+                clear_session_vars = None  # type: ignore[assignment]
+                logger.debug("Could not set ACP session context", exc_info=True)
             if approval_cb:
                 try:
                     from tools import terminal_tool as _terminal_tool
@@ -583,9 +961,10 @@ class HermesACPAgent(acp.Agent):
             os.environ["HERMES_INTERACTIVE"] = "1"
             try:
                 result = agent.run_conversation(
-                    user_message=user_text,
+                    user_message=user_content,
                     conversation_history=state.history,
                     task_id=session_id,
+                    persist_user_message=user_text or "[Image attachment]",
                 )
                 return result
             except Exception as e:
@@ -603,11 +982,24 @@ class HermesACPAgent(acp.Agent):
                         _terminal_tool.set_approval_callback(previous_approval_cb)
                     except Exception:
                         logger.debug("Could not restore approval callback", exc_info=True)
+                if session_tokens is not None and clear_session_vars is not None:
+                    try:
+                        clear_session_vars(session_tokens)
+                    except Exception:
+                        logger.debug("Could not clear ACP session context", exc_info=True)
 
         try:
-            result = await loop.run_in_executor(_executor, _run_agent)
+            # Wrap the executor call in a fresh copy of the current context so
+            # concurrent ACP sessions on the shared ThreadPoolExecutor don't
+            # stomp on each other's ContextVar writes (HERMES_SESSION_KEY in
+            # particular — used by the interactive sudo password cache scope).
+            ctx = contextvars.copy_context()
+            result = await loop.run_in_executor(_executor, ctx.run, _run_agent)
         except Exception:
             logger.exception("Executor error for session %s", session_id)
+            with state.runtime_lock:
+                state.is_running = False
+                state.current_prompt_text = ""
             return PromptResponse(stop_reason="end_turn")
 
         if result.get("messages"):
@@ -629,10 +1021,32 @@ class HermesACPAgent(acp.Agent):
                 )
             except Exception:
                 logger.debug("Failed to auto-title ACP session %s", session_id, exc_info=True)
-        if final_response and conn:
+        if final_response and conn and not streamed_message:
             update = acp.update_agent_message_text(final_response)
             await conn.session_update(session_id, update)
 
+        # Mark this turn idle before draining queued work so recursive prompt()
+        # calls can acquire the session. Queued turns are intentionally run as
+        # normal follow-up user prompts, preserving role alternation and history.
+        with state.runtime_lock:
+            state.is_running = False
+            state.current_prompt_text = ""
+
+        while True:
+            with state.runtime_lock:
+                if not state.queued_prompts:
+                    break
+                next_prompt = state.queued_prompts.pop(0)
+            if conn:
+                await conn.session_update(
+                    session_id,
+                    acp.update_user_message_text(next_prompt),
+                )
+            await self.prompt(
+                prompt=[TextContentBlock(type="text", text=next_prompt)],
+                session_id=session_id,
+            )
+
         usage = None
         if any(result.get(key) is not None for key in ("prompt_tokens", "completion_tokens", "total_tokens")):
             usage = Usage(
@@ -643,6 +1057,8 @@ class HermesACPAgent(acp.Agent):
                 cached_read_tokens=result.get("cache_read_tokens"),
             )
 
+        await self._send_usage_update(state)
+
         stop_reason = "cancelled" if state.cancel_event and state.cancel_event.is_set() else "end_turn"
         return PromptResponse(stop_reason=stop_reason, usage=usage)
 
@@ -710,6 +1126,8 @@ class HermesACPAgent(acp.Agent):
             "context": self._cmd_context,
             "reset": self._cmd_reset,
             "compact": self._cmd_compact,
+            "steer": self._cmd_steer,
+            "queue": self._cmd_queue,
             "version": self._cmd_version,
         }.get(cmd)
 
@@ -754,7 +1172,9 @@ class HermesACPAgent(acp.Agent):
     def _cmd_tools(self, args: str, state: SessionState) -> str:
         try:
             from model_tools import get_tool_definitions
-            toolsets = getattr(state.agent, "enabled_toolsets", None) or ["hermes-acp"]
+            toolsets = _expand_acp_enabled_toolsets(
+                getattr(state.agent, "enabled_toolsets", None) or ["hermes-acp"]
+            )
             tools = get_tool_definitions(enabled_toolsets=toolsets, quiet_mode=True)
             if not tools:
                 return "No tools available."
@@ -771,22 +1191,84 @@ class HermesACPAgent(acp.Agent):
             return f"Could not list tools: {e}"
 
     def _cmd_context(self, args: str, state: SessionState) -> str:
+        """Show ACP session context pressure and compression guidance."""
         n_messages = len(state.history)
-        if n_messages == 0:
-            return "Conversation is empty (no messages yet)."
-        # Count by role
+
+        # Count by role.
         roles: dict[str, int] = {}
         for msg in state.history:
             role = msg.get("role", "unknown")
             roles[role] = roles.get(role, 0) + 1
+
+        agent = state.agent
+        model = state.model or getattr(agent, "model", "")
+        provider = getattr(agent, "provider", None) or "auto"
+        compressor = getattr(agent, "context_compressor", None)
+        context_length = int(getattr(compressor, "context_length", 0) or 0)
+        threshold_tokens = int(getattr(compressor, "threshold_tokens", 0) or 0)
+
+        try:
+            from agent.model_metadata import estimate_request_tokens_rough
+
+            system_prompt = getattr(agent, "_cached_system_prompt", "") or ""
+            tools = getattr(agent, "tools", None) or None
+            approx_tokens = estimate_request_tokens_rough(
+                state.history,
+                system_prompt=system_prompt,
+                tools=tools,
+            )
+        except Exception:
+            logger.debug("Could not estimate ACP context usage", exc_info=True)
+            approx_tokens = 0
+
+        if threshold_tokens <= 0 and context_length > 0:
+            threshold_tokens = int(context_length * 0.80)
+
         lines = [
-            f"Conversation: {n_messages} messages",
+            f"Conversation: {n_messages} messages"
+            if n_messages
+            else "Conversation is empty (no messages yet).",
             f"  user: {roles.get('user', 0)}, assistant: {roles.get('assistant', 0)}, "
             f"tool: {roles.get('tool', 0)}, system: {roles.get('system', 0)}",
         ]
-        model = state.model or getattr(state.agent, "model", "")
         if model:
             lines.append(f"Model: {model}")
+        lines.append(f"Provider: {provider}")
+
+        if approx_tokens > 0:
+            if context_length > 0:
+                usage_pct = (approx_tokens / context_length) * 100
+                lines.append(
+                    f"Context usage: ~{approx_tokens:,} / {context_length:,} tokens ({usage_pct:.1f}%)"
+                )
+            else:
+                lines.append(f"Context usage: ~{approx_tokens:,} tokens")
+
+        if threshold_tokens > 0:
+            if approx_tokens > 0:
+                threshold_pct = (threshold_tokens / context_length) * 100 if context_length > 0 else 0
+                remaining = max(threshold_tokens - approx_tokens, 0)
+                if approx_tokens >= threshold_tokens:
+                    lines.append(
+                        f"Compression: due now (threshold ~{threshold_tokens:,}"
+                        + (f", {threshold_pct:.0f}%" if threshold_pct else "")
+                        + "). Run /compact."
+                    )
+                else:
+                    lines.append(
+                        f"Compression: ~{remaining:,} tokens until threshold "
+                        f"(~{threshold_tokens:,}"
+                        + (f", {threshold_pct:.0f}%" if threshold_pct else "")
+                        + ")."
+                    )
+            else:
+                lines.append(f"Compression threshold: ~{threshold_tokens:,} tokens")
+
+        if getattr(agent, "compression_enabled", True) is False:
+            lines.append("Compression is disabled for this agent.")
+        else:
+            lines.append("Tip: run /compact to compress manually before the threshold.")
+
         return "\n".join(lines)
 
     def _cmd_reset(self, args: str, state: SessionState) -> str:
@@ -804,10 +1286,16 @@ class HermesACPAgent(acp.Agent):
             if not hasattr(agent, "_compress_context"):
                 return "Context compression not available for this agent."
 
-            from agent.model_metadata import estimate_messages_tokens_rough
+            from agent.model_metadata import estimate_request_tokens_rough
 
             original_count = len(state.history)
-            approx_tokens = estimate_messages_tokens_rough(state.history)
+            # Include system prompt + tool schemas so the figure reflects real
+            # request pressure, not a transcript-only underestimate (#6217).
+            _sys_prompt = getattr(agent, "_cached_system_prompt", "") or ""
+            _tools = getattr(agent, "tools", None) or None
+            approx_tokens = estimate_request_tokens_rough(
+                state.history, system_prompt=_sys_prompt, tools=_tools
+            )
             original_session_db = getattr(agent, "_session_db", None)
 
             try:
@@ -827,7 +1315,13 @@ class HermesACPAgent(acp.Agent):
             self.session_manager.save_session(state.session_id)
 
             new_count = len(state.history)
-            new_tokens = estimate_messages_tokens_rough(state.history)
+            _sys_prompt_after = getattr(agent, "_cached_system_prompt", "") or _sys_prompt
+            _tools_after = getattr(agent, "tools", None) or _tools
+            new_tokens = estimate_request_tokens_rough(
+                state.history,
+                system_prompt=_sys_prompt_after,
+                tools=_tools_after,
+            )
             return (
                 f"Context compressed: {original_count} -> {new_count} messages\n"
                 f"~{approx_tokens:,} -> ~{new_tokens:,} tokens"
@@ -835,6 +1329,34 @@ class HermesACPAgent(acp.Agent):
         except Exception as e:
             return f"Compression failed: {e}"
 
+    def _cmd_steer(self, args: str, state: SessionState) -> str:
+        steer_text = args.strip()
+        if not steer_text:
+            return "Usage: /steer <guidance>"
+
+        if state.is_running and hasattr(state.agent, "steer"):
+            try:
+                if state.agent.steer(steer_text):
+                    preview = steer_text[:80] + ("..." if len(steer_text) > 80 else "")
+                    return f"⏩ Steer queued for the active turn: {preview}"
+            except Exception as exc:
+                logger.warning("ACP steer failed for session %s: %s", state.session_id, exc)
+                return f"⚠️ Steer failed: {exc}"
+
+        with state.runtime_lock:
+            state.queued_prompts.append(steer_text)
+            depth = len(state.queued_prompts)
+        return f"No active turn — queued for the next turn. ({depth} queued)"
+
+    def _cmd_queue(self, args: str, state: SessionState) -> str:
+        queued_text = args.strip()
+        if not queued_text:
+            return "Usage: /queue <prompt>"
+        with state.runtime_lock:
+            state.queued_prompts.append(queued_text)
+            depth = len(state.queued_prompts)
+        return f"Queued for the next turn. ({depth} queued)"
+
     def _cmd_version(self, args: str, state: SessionState) -> str:
         return f"Hermes Agent v{HERMES_VERSION}"
 

acp_adapter/session.py

@@ -26,6 +26,33 @@ from typing import Any, Dict, List, Optional
 logger = logging.getLogger(__name__)
 
 
+def _win_path_to_wsl(path: str) -> str | None:
+    """Convert a Windows drive path to its WSL /mnt/<drive>/... equivalent."""
+    match = re.match(r"^([A-Za-z]):[\\/](.*)$", path)
+    if not match:
+        return None
+    drive = match.group(1).lower()
+    tail = match.group(2).replace("\\", "/")
+    return f"/mnt/{drive}/{tail}"
+
+
+def _translate_acp_cwd(cwd: str) -> str:
+    """Translate Windows ACP cwd values when Hermes itself is running in WSL.
+
+    Windows ACP clients can launch ``hermes acp`` inside WSL while still sending
+    editor workspaces as Windows drive paths such as ``E:\\Projects``. Store
+    and execute against the WSL mount path so agents, tools, and persisted ACP
+    sessions all agree on the usable workspace. Native Linux/macOS keeps the
+    original cwd unchanged.
+    """
+    from hermes_constants import is_wsl
+
+    if not is_wsl():
+        return cwd
+    translated = _win_path_to_wsl(str(cwd))
+    return translated if translated is not None else cwd
+
+
 def _normalize_cwd_for_compare(cwd: str | None) -> str:
     raw = str(cwd or ".").strip()
     if not raw:
@@ -34,11 +61,9 @@ def _normalize_cwd_for_compare(cwd: str | None) -> str:
 
     # Normalize Windows drive paths into the equivalent WSL mount form so
     # ACP history filters match the same workspace across Windows and WSL.
-    match = re.match(r"^([A-Za-z]):[\\/](.*)$", expanded)
-    if match:
-        drive = match.group(1).lower()
-        tail = match.group(2).replace("\\", "/")
-        expanded = f"/mnt/{drive}/{tail}"
+    translated = _win_path_to_wsl(expanded)
+    if translated is not None:
+        expanded = translated
     elif re.match(r"^/mnt/[A-Za-z]/", expanded):
         expanded = f"/mnt/{expanded[5].lower()}/{expanded[7:]}"
 
@@ -96,16 +121,40 @@ def _acp_stderr_print(*args, **kwargs) -> None:
 
 
 def _register_task_cwd(task_id: str, cwd: str) -> None:
-    """Bind a task/session id to the editor's working directory for tools."""
+    """Bind a task/session id to the editor's working directory for tools.
+
+    Zed can launch Hermes from a Windows workspace while the ACP process runs
+    inside WSL. In that case ACP sends cwd as e.g. ``E:\\Projects\\POTI``;
+    local tools need the WSL mount equivalent or subprocess creation fails
+    before the command can run.
+    """
     if not task_id:
         return
     try:
         from tools.terminal_tool import register_task_env_overrides
-        register_task_env_overrides(task_id, {"cwd": cwd})
+        register_task_env_overrides(task_id, {"cwd": _translate_acp_cwd(cwd)})
     except Exception:
         logger.debug("Failed to register ACP task cwd override", exc_info=True)
 
 
+def _expand_acp_enabled_toolsets(
+    toolsets: List[str] | None = None,
+    mcp_server_names: List[str] | None = None,
+) -> List[str]:
+    """Return ACP toolsets plus explicit MCP server toolsets for this session."""
+    expanded: List[str] = []
+    for name in list(toolsets or ["hermes-acp"]):
+        if name and name not in expanded:
+            expanded.append(name)
+
+    for server_name in list(mcp_server_names or []):
+        toolset_name = f"mcp-{server_name}"
+        if server_name and toolset_name not in expanded:
+            expanded.append(toolset_name)
+
+    return expanded
+
+
 def _clear_task_cwd(task_id: str) -> None:
     """Remove task-specific cwd overrides for an ACP session."""
     if not task_id:
@@ -127,6 +176,11 @@ class SessionState:
     model: str = ""
     history: List[Dict[str, Any]] = field(default_factory=list)
     cancel_event: Any = None  # threading.Event
+    is_running: bool = False
+    queued_prompts: List[str] = field(default_factory=list)
+    runtime_lock: Any = field(default_factory=Lock)
+    current_prompt_text: str = ""
+    interrupted_prompt_text: str = ""
 
 
 class SessionManager:
@@ -157,6 +211,7 @@ class SessionManager:
         """Create a new session with a unique ID and a fresh AIAgent."""
         import threading
 
+        cwd = _translate_acp_cwd(cwd)
         session_id = str(uuid.uuid4())
         agent = self._make_agent(session_id=session_id, cwd=cwd)
         state = SessionState(
@@ -199,6 +254,7 @@ class SessionManager:
         """Deep-copy a session's history into a new session."""
         import threading
 
+        cwd = _translate_acp_cwd(cwd)
         original = self.get_session(session_id)  # checks DB too
         if original is None:
             return None
@@ -300,6 +356,7 @@ class SessionManager:
 
     def update_cwd(self, session_id: str, cwd: str) -> Optional[SessionState]:
         """Update the working directory for a session and its tool overrides."""
+        cwd = _translate_acp_cwd(cwd)
         state = self.get_session(session_id)  # checks DB too
         if state is None:
             return None
@@ -409,17 +466,10 @@ class SessionManager:
                 except Exception:
                     logger.debug("Failed to update ACP session metadata", exc_info=True)
 
-            # Replace stored messages with current history.
-            db.clear_messages(state.session_id)
-            for msg in state.history:
-                db.append_message(
-                    session_id=state.session_id,
-                    role=msg.get("role", "user"),
-                    content=msg.get("content"),
-                    tool_name=msg.get("tool_name") or msg.get("name"),
-                    tool_calls=msg.get("tool_calls"),
-                    tool_call_id=msg.get("tool_call_id"),
-                )
+            # Replace stored messages with current history atomically so a
+            # mid-rewrite failure rolls back and the previously persisted
+            # conversation is preserved (salvaged from #13675).
+            db.replace_messages(state.session_id, state.history)
         except Exception:
             logger.warning("Failed to persist ACP session %s", state.session_id, exc_info=True)
 
@@ -537,9 +587,18 @@ class SessionManager:
         elif isinstance(model_cfg, str) and model_cfg.strip():
             default_model = model_cfg.strip()
 
+        configured_mcp_servers = [
+            name
+            for name, cfg in (config.get("mcp_servers") or {}).items()
+            if not isinstance(cfg, dict) or cfg.get("enabled", True) is not False
+        ]
+
         kwargs = {
             "platform": "acp",
-            "enabled_toolsets": ["hermes-acp"],
+            "enabled_toolsets": _expand_acp_enabled_toolsets(
+                ["hermes-acp"],
+                mcp_server_names=configured_mcp_servers,
+            ),
             "quiet_mode": True,
             "session_id": session_id,
             "model": model or default_model,

agent/anthropic_adapter.py

@@ -14,16 +14,33 @@ import copy
 import json
 import logging
 import os
+import platform
+import subprocess
 from pathlib import Path
 
 from hermes_constants import get_hermes_home
 from typing import Any, Dict, List, Optional, Tuple
-from utils import normalize_proxy_env_vars
+from utils import base_url_host_matches, normalize_proxy_env_vars
 
-try:
-    import anthropic as _anthropic_sdk
-except ImportError:
-    _anthropic_sdk = None  # type: ignore[assignment]
+# NOTE: `import anthropic` is deliberately NOT at module top — the SDK pulls
+# ~220 ms of imports (anthropic.types, anthropic.lib.tools._beta_runner, etc.)
+# and the 3 usage sites (build_anthropic_client, build_anthropic_bedrock_client,
+# read_claude_code_credentials_from_keychain) are all on cold user-triggered
+# paths. Access via the `_get_anthropic_sdk()` accessor below, which caches
+# the module after the first call and returns None on ImportError.
+_anthropic_sdk: Any = ...  # sentinel — None means "tried and missing"
+
+
+def _get_anthropic_sdk():
+    """Return the ``anthropic`` SDK module, importing lazily. None if not installed."""
+    global _anthropic_sdk
+    if _anthropic_sdk is ...:
+        try:
+            import anthropic as _sdk
+            _anthropic_sdk = _sdk
+        except ImportError:
+            _anthropic_sdk = None
+    return _anthropic_sdk
 
 logger = logging.getLogger(__name__)
 
@@ -59,6 +76,7 @@ _ADAPTIVE_THINKING_SUBSTRINGS = ("4-6", "4.6", "4-7", "4.7")
 # Models where temperature/top_p/top_k return 400 if set to non-default values.
 # This is the Opus 4.7 contract; future 4.x+ models are expected to follow it.
 _NO_SAMPLING_PARAMS_SUBSTRINGS = ("4-7", "4.7")
+_FAST_MODE_SUPPORTED_SUBSTRINGS = ("opus-4-6", "opus-4.6")
 
 # ── Max output token limits per Anthropic model ───────────────────────
 # Source: Anthropic docs + Cline model catalog.  Anthropic's API requires
@@ -88,6 +106,9 @@ _ANTHROPIC_OUTPUT_LIMITS = {
     "claude-3-haiku":      4_096,
     # Third-party Anthropic-compatible providers
     "minimax":            131_072,
+    # Qwen models via DashScope Anthropic-compatible endpoint
+    # DashScope enforces max_tokens ∈ [1, 65536]
+    "qwen3":               65_536,
 }
 
 # For any model not in the table, assume the highest current limit.
@@ -199,20 +220,45 @@ def _forbids_sampling_params(model: str) -> bool:
     return any(v in model for v in _NO_SAMPLING_PARAMS_SUBSTRINGS)
 
 
+def _supports_fast_mode(model: str) -> bool:
+    """Return True for models that support Anthropic Fast Mode (speed=fast).
+
+    Per Anthropic docs, fast mode is currently supported on Opus 4.6 only.
+    Sending ``speed: "fast"`` to any other Claude model (including Opus 4.7)
+    returns HTTP 400. This guard prevents silently 400'ing when stale config
+    or older callers leave fast mode enabled across a model upgrade.
+    """
+    return any(v in model for v in _FAST_MODE_SUPPORTED_SUBSTRINGS)
+
+
 # Beta headers for enhanced features (sent with ALL auth types).
-# As of Opus 4.7 (2026-04-16), both of these are GA on Claude 4.6+ — the
+# As of Opus 4.7 (2026-04-16), the first two are GA on Claude 4.6+ — the
 # beta headers are still accepted (harmless no-op) but not required. Kept
 # here so older Claude (4.5, 4.1) + third-party Anthropic-compat endpoints
 # that still gate on the headers continue to get the enhanced features.
-# Migration guide: remove these if you no longer support ≤4.5 models.
+#
+# ``context-1m-2025-08-07`` unlocks the 1M context window on Claude Opus 4.6/4.7
+# and Sonnet 4.6 when served via AWS Bedrock or Azure AI Foundry. 1M is GA on
+# native Anthropic (api.anthropic.com) for Opus 4.6+, but Bedrock/Azure still
+# gate it behind this beta header as of 2026-04 — without it Bedrock caps Opus
+# at 200K even though model_metadata.py advertises 1M. The header is a harmless
+# no-op on endpoints where 1M is GA.
+#
+# Migration guide: remove these if you no longer support ≤4.5 models or once
+# Bedrock/Azure promote 1M to GA.
 _COMMON_BETAS = [
     "interleaved-thinking-2025-05-14",
     "fine-grained-tool-streaming-2025-05-14",
+    "context-1m-2025-08-07",
 ]
 # MiniMax's Anthropic-compatible endpoints fail tool-use requests when
 # the fine-grained tool streaming beta is present.  Omit it so tool calls
 # fall back to the provider's default response path.
 _TOOL_STREAMING_BETA = "fine-grained-tool-streaming-2025-05-14"
+# 1M context beta — see comment on _COMMON_BETAS above. Stripped for
+# Bearer-auth (MiniMax) endpoints since they host their own models and
+# unknown Anthropic beta headers risk request rejection.
+_CONTEXT_1M_BETA = "context-1m-2025-08-07"
 
 # Fast mode beta — enables the ``speed: "fast"`` request parameter for
 # significantly higher output token throughput on Opus 4.6 (~2.5x).
@@ -277,8 +323,9 @@ def _is_oauth_token(key: str) -> bool:
     Positively identifies Anthropic OAuth tokens by their key format:
     - ``sk-ant-`` prefix (but NOT ``sk-ant-api``) → setup tokens, managed keys
     - ``eyJ`` prefix → JWTs from the Anthropic OAuth flow
+    - ``cc-`` prefix → Claude Code OAuth access tokens (from CLAUDE_CODE_OAUTH_TOKEN)
 
-    Non-Anthropic keys (MiniMax, Alibaba, etc.) don't match either pattern
+    Non-Anthropic keys (MiniMax, Alibaba, etc.) don't match any pattern
     and correctly return False.
     """
     if not key:
@@ -292,6 +339,9 @@ def _is_oauth_token(key: str) -> bool:
     # JWTs from Anthropic OAuth flow
     if key.startswith("eyJ"):
         return True
+    # Claude Code OAuth access tokens (opaque, from CLAUDE_CODE_OAUTH_TOKEN)
+    if key.startswith("cc-"):
+        return True
     return False
 
 
@@ -330,6 +380,88 @@ def _is_kimi_coding_endpoint(base_url: str | None) -> bool:
     return normalized.rstrip("/").lower().startswith("https://api.kimi.com/coding")
 
 
+# Model-name prefixes that identify the Kimi / Moonshot family.  Covers
+# - official slugs: ``kimi-k2.5``, ``kimi_thinking``, ``moonshot-v1-8k``
+# - common release lines: ``k1.5-...``, ``k2-thinking``, ``k25-...``, ``k2.5-...``
+# Matched case-insensitively against the post-``normalize_model_name`` form,
+# so a caller's ``provider/vendor/model`` slug is handled the same as a
+# bare name.
+_KIMI_FAMILY_MODEL_PREFIXES = (
+    "kimi-", "kimi_",
+    "moonshot-", "moonshot_",
+    "k1.", "k1-",
+    "k2.", "k2-",
+    "k25", "k2.5",
+)
+
+
+def _model_name_is_kimi_family(model: str | None) -> bool:
+    if not isinstance(model, str):
+        return False
+    m = model.strip().lower()
+    if not m:
+        return False
+    # Strip vendor prefix (e.g. ``moonshotai/kimi-k2.5`` → ``kimi-k2.5``)
+    if "/" in m:
+        m = m.rsplit("/", 1)[-1]
+    return m.startswith(_KIMI_FAMILY_MODEL_PREFIXES)
+
+
+def _is_kimi_family_endpoint(base_url: str | None, model: str | None = None) -> bool:
+    """Return True for any Kimi / Moonshot Anthropic-Messages-speaking endpoint.
+
+    Broader than ``_is_kimi_coding_endpoint`` — matches:
+
+    - Kimi's official ``/coding`` URL (legacy check, preserved)
+    - Any ``api.kimi.com`` / ``moonshot.ai`` / ``moonshot.cn`` host
+    - Custom or proxied endpoints whose *model* name is in the Kimi / Moonshot
+      family (``kimi-*``, ``moonshot-*``, ``k1.*``, ``k2.*``, …).  Users with
+      ``api_mode: anthropic_messages`` on a private gateway fronting Kimi
+      fall into this branch — the upstream still enforces Kimi's thinking
+      semantics (reasoning_content required on every replayed tool-call
+      message) regardless of the gateway's hostname.
+
+    Used to decide whether to drop Anthropic's ``thinking`` kwarg and to
+    preserve unsigned reasoning_content-derived thinking blocks on replay.
+    See hermes-agent#13848, #17057.
+    """
+    if _is_kimi_coding_endpoint(base_url):
+        return True
+    for _domain in ("api.kimi.com", "moonshot.ai", "moonshot.cn"):
+        if base_url_host_matches(base_url or "", _domain):
+            return True
+    if _model_name_is_kimi_family(model):
+        return True
+    return False
+
+
+def _is_deepseek_anthropic_endpoint(base_url: str | None) -> bool:
+    """Return True for DeepSeek's Anthropic-compatible endpoint.
+
+    DeepSeek's ``/anthropic`` route speaks the Anthropic Messages protocol
+    but, when thinking mode is enabled, requires the ``thinking`` blocks
+    from prior assistant turns to round-trip on subsequent requests — the
+    generic third-party path strips them and triggers HTTP 400::
+
+        The content[].thinking in the thinking mode must be passed back
+        to the API.
+
+    Per DeepSeek's published compatibility matrix the blocks are unsigned
+    (no Anthropic-proprietary signature, no ``redacted_thinking`` support),
+    so this endpoint is handled with the same strip-signed / keep-unsigned
+    policy used for Kimi's ``/coding`` endpoint.  The match is pinned to
+    the ``/anthropic`` path so the OpenAI-compatible ``api.deepseek.com``
+    base URL (which never reaches this adapter) is not misclassified.
+    See hermes-agent#16748.
+    """
+    if not base_url_host_matches(base_url or "", "api.deepseek.com"):
+        return False
+    normalized = _normalize_base_url_text(base_url)
+    if not normalized:
+        return False
+    return "/anthropic" in normalized.rstrip("/").lower()
+
+
 def _requires_bearer_auth(base_url: str | None) -> bool:
     """Return True for Anthropic-compatible providers that require Bearer auth.
 
@@ -344,20 +476,45 @@ def _requires_bearer_auth(base_url: str | None) -> bool:
     return normalized.startswith(("https://api.minimax.io/anthropic", "https://api.minimaxi.com/anthropic"))
 
 
-def _common_betas_for_base_url(base_url: str | None) -> list[str]:
+def _common_betas_for_base_url(
+    base_url: str | None,
+    *,
+    drop_context_1m_beta: bool = False,
+) -> list[str]:
     """Return the beta headers that are safe for the configured endpoint.
 
     MiniMax's Anthropic-compatible endpoints (Bearer-auth) reject requests
     that include Anthropic's ``fine-grained-tool-streaming`` beta — every
     tool-use message triggers a connection error.  Strip that beta for
     Bearer-auth endpoints while keeping all other betas intact.
+
+    The ``context-1m-2025-08-07`` beta is also stripped for Bearer-auth
+    endpoints — MiniMax hosts its own models, not Claude, so the header is
+    irrelevant at best and risks request rejection at worst.
+
+    ``drop_context_1m_beta=True`` additionally strips the 1M-context beta on
+    otherwise-unrelated endpoints. The OAuth retry path flips this flag after
+    a subscription rejects the beta with
+    "The long context beta is not yet available for this subscription" so
+    subsequent requests in the same session don't repeat the probe. See the
+    reactive recovery loop in ``run_agent.py`` and issue-comment history on
+    PR #17680 for the full rationale.
     """
     if _requires_bearer_auth(base_url):
-        return [b for b in _COMMON_BETAS if b != _TOOL_STREAMING_BETA]
+        _stripped = {_TOOL_STREAMING_BETA, _CONTEXT_1M_BETA}
+        return [b for b in _COMMON_BETAS if b not in _stripped]
+    if drop_context_1m_beta:
+        return [b for b in _COMMON_BETAS if b != _CONTEXT_1M_BETA]
     return _COMMON_BETAS
 
 
-def build_anthropic_client(api_key: str, base_url: str = None, timeout: float = None):
+def build_anthropic_client(
+    api_key: str,
+    base_url: str = None,
+    timeout: float = None,
+    *,
+    drop_context_1m_beta: bool = False,
+):
     """Create an Anthropic client, auto-detecting setup-tokens vs API keys.
 
     If *timeout* is provided it overrides the default 900s read timeout.  The
@@ -366,8 +523,15 @@ def build_anthropic_client(api_key: str, base_url: str = None, timeout: float =
     Anthropic-compatible providers respect the same knob as OpenAI-wire
     providers.
 
+    ``drop_context_1m_beta=True`` strips ``context-1m-2025-08-07`` from the
+    client-level ``anthropic-beta`` header. Used by the reactive OAuth retry
+    path in ``run_agent.py`` when a subscription rejects the beta; leave at
+    its default on fresh clients so 1M-capable subscriptions keep the
+    capability.
+
     Returns an anthropic.Anthropic instance.
     """
+    _anthropic_sdk = _get_anthropic_sdk()
     if _anthropic_sdk is None:
         raise ImportError(
             "The 'anthropic' package is required for the Anthropic provider. "
@@ -384,8 +548,20 @@ def build_anthropic_client(api_key: str, base_url: str = None, timeout: float =
         "timeout": Timeout(timeout=float(_read_timeout), connect=10.0),
     }
     if normalized_base_url:
-        kwargs["base_url"] = normalized_base_url
-    common_betas = _common_betas_for_base_url(normalized_base_url)
+        # Azure Anthropic endpoints require an ``api-version`` query parameter.
+        # Pass it via default_query so the SDK appends it to every request URL
+        # without corrupting the base_url (appending it directly produces
+        # malformed paths like /anthropic?api-version=.../v1/messages).
+        _is_azure_endpoint = "azure.com" in normalized_base_url.lower()
+        if _is_azure_endpoint and "api-version" not in normalized_base_url:
+            kwargs["base_url"] = normalized_base_url.rstrip("/")
+            kwargs["default_query"] = {"api-version": "2025-04-15"}
+        else:
+            kwargs["base_url"] = normalized_base_url
+    common_betas = _common_betas_for_base_url(
+        normalized_base_url,
+        drop_context_1m_beta=drop_context_1m_beta,
+    )
 
     if _is_kimi_coding_endpoint(base_url):
         # Kimi's /coding endpoint requires User-Agent: claude-code/0.1.0
@@ -441,8 +617,16 @@ def build_anthropic_bedrock_client(region: str):
     Claude feature parity: prompt caching, thinking budgets, adaptive
     thinking, fast mode — features not available via the Converse API.
 
+    Attaches the common Anthropic beta headers as client-level defaults so
+    that Bedrock-hosted Claude models get the same enhanced features as
+    native Anthropic. The ``context-1m-2025-08-07`` beta in particular
+    unlocks the 1M context window for Opus 4.6/4.7 on Bedrock — without
+    it, Bedrock caps these models at 200K even though the Anthropic API
+    serves them with 1M natively.
+
     Auth uses the boto3 default credential chain (IAM roles, SSO, env vars).
     """
+    _anthropic_sdk = _get_anthropic_sdk()
     if _anthropic_sdk is None:
         raise ImportError(
             "The 'anthropic' package is required for the Bedrock provider. "
@@ -458,11 +642,73 @@ def build_anthropic_bedrock_client(region: str):
     return _anthropic_sdk.AnthropicBedrock(
         aws_region=region,
         timeout=Timeout(timeout=900.0, connect=10.0),
+        default_headers={"anthropic-beta": ",".join(_COMMON_BETAS)},
     )
 
 
+def _read_claude_code_credentials_from_keychain() -> Optional[Dict[str, Any]]:
+    """Read Claude Code OAuth credentials from the macOS Keychain.
+
+    Claude Code >=2.1.114 stores credentials in the macOS Keychain under the
+    service name "Claude Code-credentials" rather than (or in addition to)
+    the JSON file at ~/.claude/.credentials.json.
+
+    The password field contains a JSON string with the same claudeAiOauth
+    structure as the JSON file.
+
+    Returns dict with {accessToken, refreshToken?, expiresAt?} or None.
+    """
+    if platform.system() != "Darwin":
+        return None
+
+    try:
+        # Read the "Claude Code-credentials" generic password entry
+        result = subprocess.run(
+            ["security", "find-generic-password",
+             "-s", "Claude Code-credentials",
+             "-w"],
+            capture_output=True,
+            text=True,
+            timeout=5,
+        )
+    except (OSError, subprocess.TimeoutExpired):
+        logger.debug("Keychain: security command not available or timed out")
+        return None
+
+    if result.returncode != 0:
+        logger.debug("Keychain: no entry found for 'Claude Code-credentials'")
+        return None
+
+    raw = result.stdout.strip()
+    if not raw:
+        return None
+
+    try:
+        data = json.loads(raw)
+    except json.JSONDecodeError:
+        logger.debug("Keychain: credentials payload is not valid JSON")
+        return None
+
+    oauth_data = data.get("claudeAiOauth")
+    if oauth_data and isinstance(oauth_data, dict):
+        access_token = oauth_data.get("accessToken", "")
+        if access_token:
+            return {
+                "accessToken": access_token,
+                "refreshToken": oauth_data.get("refreshToken", ""),
+                "expiresAt": oauth_data.get("expiresAt", 0),
+                "source": "macos_keychain",
+            }
+
+    return None
+
+
 def read_claude_code_credentials() -> Optional[Dict[str, Any]]:
-    """Read refreshable Claude Code OAuth credentials from ~/.claude/.credentials.json.
+    """Read refreshable Claude Code OAuth credentials.
+
+    Checks two sources in order:
+      1. macOS Keychain (Darwin only) — "Claude Code-credentials" entry
+      2. ~/.claude/.credentials.json file
 
     This intentionally excludes ~/.claude.json primaryApiKey. Opencode's
     subscription flow is OAuth/setup-token based with refreshable credentials,
@@ -471,6 +717,12 @@ def read_claude_code_credentials() -> Optional[Dict[str, Any]]:
 
     Returns dict with {accessToken, refreshToken?, expiresAt?} or None.
     """
+    # Try macOS Keychain first (covers Claude Code >=2.1.114)
+    kc_creds = _read_claude_code_credentials_from_keychain()
+    if kc_creds:
+        return kc_creds
+
+    # Fall back to JSON file
     cred_path = Path.home() / ".claude" / ".credentials.json"
     if cred_path.exists():
         try:
@@ -641,7 +893,9 @@ def _write_claude_code_credentials(
         existing["claudeAiOauth"] = oauth_data
 
         cred_path.parent.mkdir(parents=True, exist_ok=True)
-        cred_path.write_text(json.dumps(existing, indent=2), encoding="utf-8")
+        _tmp_cred = cred_path.with_suffix(".tmp")
+        _tmp_cred.write_text(json.dumps(existing, indent=2), encoding="utf-8")
+        _tmp_cred.replace(cred_path)
         # Restrict permissions (credentials file)
         cred_path.chmod(0o600)
     except (OSError, IOError) as e:
@@ -908,6 +1162,26 @@ def read_hermes_oauth_credentials() -> Optional[Dict[str, Any]]:
 # ---------------------------------------------------------------------------
 
 
+def _is_bedrock_model_id(model: str) -> bool:
+    """Detect AWS Bedrock model IDs that use dots as namespace separators.
+
+    Bedrock model IDs come in two forms:
+    - Bare:    ``anthropic.claude-opus-4-7``
+    - Regional (inference profiles): ``us.anthropic.claude-sonnet-4-5-v1:0``
+
+    In both cases the dots separate namespace components, not version
+    numbers, and must be preserved verbatim for the Bedrock API.
+    """
+    lower = model.lower()
+    # Regional inference-profile prefixes
+    if any(lower.startswith(p) for p in ("global.", "us.", "eu.", "ap.", "jp.")):
+        return True
+    # Bare Bedrock model IDs: provider.model-family
+    if lower.startswith("anthropic."):
+        return True
+    return False
+
+
 def normalize_model_name(model: str, preserve_dots: bool = False) -> str:
     """Normalize a model name for the Anthropic API.
 
@@ -915,14 +1189,25 @@ def normalize_model_name(model: str, preserve_dots: bool = False) -> str:
     - Converts dots to hyphens in version numbers (OpenRouter uses dots,
       Anthropic uses hyphens: claude-opus-4.6 → claude-opus-4-6), unless
       preserve_dots is True (e.g. for Alibaba/DashScope: qwen3.5-plus).
+    - Preserves Bedrock model IDs (``anthropic.claude-opus-4-7``) and
+      regional inference profiles (``us.anthropic.claude-*``) whose dots
+      are namespace separators, not version separators.
     """
     lower = model.lower()
     if lower.startswith("anthropic/"):
         model = model[len("anthropic/"):]
     if not preserve_dots:
-        # OpenRouter uses dots for version separators (claude-opus-4.6),
-        # Anthropic uses hyphens (claude-opus-4-6). Convert dots to hyphens.
-        model = model.replace(".", "-")
+        # Bedrock model IDs use dots as namespace separators
+        # (e.g. "anthropic.claude-opus-4-7", "us.anthropic.claude-*").
+        # These must not be converted to hyphens.  See issue #12295.
+        if _is_bedrock_model_id(model):
+            return model
+        # Only convert dots to hyphens for Anthropic/Claude models.
+        # Non-Anthropic models (gpt-5.4, gemini-2.5, etc.) use dots
+        # as part of their canonical names.  See issue #17171.
+        _lower = model.lower()
+        if _lower.startswith("claude-") or _lower.startswith("anthropic/"):
+            model = model.replace(".", "-")
     return model
 
 
@@ -939,17 +1224,74 @@ def _sanitize_tool_id(tool_id: str) -> str:
     return sanitized or "tool_0"
 
 
+def _normalize_tool_input_schema(schema: Any) -> Dict[str, Any]:
+    """Normalize tool schemas before sending them to Anthropic.
+
+    Anthropic's tool schema validator rejects nullable unions such as
+    ``anyOf: [{"type": "string"}, {"type": "null"}]`` that Pydantic/MCP
+    commonly emits for optional fields. Tool optionality is represented by
+    the parent ``required`` array, so we delegate to the shared
+    ``strip_nullable_unions`` helper to collapse nullable unions to the
+    non-null branch while preserving metadata like description/default.
+
+    ``keep_nullable_hint=False`` because the Anthropic validator does not
+    recognize the OpenAPI-style ``nullable: true`` extension and strict
+    schema-to-grammar converters may reject unknown keywords.
+
+    Top-level ``oneOf``/``allOf``/``anyOf`` are also stripped here: the
+    Anthropic API rejects union keywords at the schema root with a generic
+    HTTP 400. Several upstream and plugin tools ship schemas with one of
+    these keywords at the top level (commonly for Pydantic discriminated
+    unions). If we land here with those keywords still present after
+    nullable-union stripping, drop them and fall back to a plain object
+    schema so the tool still validates at the Anthropic boundary.
+    """
+    if not schema:
+        return {"type": "object", "properties": {}}
+
+    from tools.schema_sanitizer import strip_nullable_unions
+
+    normalized = strip_nullable_unions(schema, keep_nullable_hint=False)
+    if not isinstance(normalized, dict):
+        return {"type": "object", "properties": {}}
+    # Strip top-level union keywords that Anthropic's validator rejects.
+    banned = {"oneOf", "allOf", "anyOf"}
+    if banned & normalized.keys():
+        normalized = {k: v for k, v in normalized.items() if k not in banned}
+        if "type" not in normalized:
+            normalized["type"] = "object"
+    if normalized.get("type") == "object" and not isinstance(normalized.get("properties"), dict):
+        normalized = {**normalized, "properties": {}}
+    return normalized
+
+
 def convert_tools_to_anthropic(tools: List[Dict]) -> List[Dict]:
     """Convert OpenAI tool definitions to Anthropic format."""
     if not tools:
         return []
     result = []
+    seen_names: set = set()
     for t in tools:
         fn = t.get("function", {})
+        name = fn.get("name", "")
+        # Defensive dedup: Anthropic rejects requests with duplicate tool
+        # names.  Upstream injection paths already dedup, but this guard
+        # converts a hard API failure into a warning.  See: #18478
+        if name and name in seen_names:
+            logger.warning(
+                "convert_tools_to_anthropic: duplicate tool name '%s' "
+                "— dropping second occurrence",
+                name,
+            )
+            continue
+        if name:
+            seen_names.add(name)
         result.append({
-            "name": fn.get("name", ""),
+            "name": name,
             "description": fn.get("description", ""),
-            "input_schema": fn.get("parameters", {"type": "object", "properties": {}}),
+            "input_schema": _normalize_tool_input_schema(
+                fn.get("parameters", {"type": "object", "properties": {}})
+            ),
         })
     return result
 
@@ -1080,6 +1422,7 @@ def _convert_content_to_anthropic(content: Any) -> Any:
 def convert_messages_to_anthropic(
     messages: List[Dict],
     base_url: str | None = None,
+    model: str | None = None,
 ) -> Tuple[Optional[Any], List[Dict]]:
     """Convert OpenAI-format messages to Anthropic format.
 
@@ -1091,6 +1434,12 @@ def convert_messages_to_anthropic(
     endpoint, all thinking block signatures are stripped.  Signatures are
     Anthropic-proprietary — third-party endpoints cannot validate them and will
     reject them with HTTP 400 "Invalid signature in thinking block".
+
+    When *model* is provided and matches the Kimi / Moonshot family (or
+    *base_url* is a Kimi / Moonshot host), unsigned thinking blocks
+    synthesised from ``reasoning_content`` are preserved on replayed
+    assistant tool-call messages — Kimi requires the field to exist, even
+    if empty.
     """
     system = None
     result = []
@@ -1319,7 +1668,16 @@ def convert_messages_to_anthropic(
     #    cache markers can interfere with signature validation.
     _THINKING_TYPES = frozenset(("thinking", "redacted_thinking"))
     _is_third_party = _is_third_party_anthropic_endpoint(base_url)
-    _is_kimi = _is_kimi_coding_endpoint(base_url)
+    # Kimi /coding and DeepSeek /anthropic share a contract: both speak the
+    # Anthropic Messages protocol upstream but require that thinking blocks
+    # synthesised from reasoning_content round-trip on subsequent turns when
+    # thinking is enabled.  Signed Anthropic blocks still have to be stripped
+    # (neither endpoint can validate Anthropic's signatures); unsigned blocks
+    # are preserved.  See hermes-agent#13848 (Kimi) and #16748 (DeepSeek).
+    _preserve_unsigned_thinking = (
+        _is_kimi_family_endpoint(base_url, model)
+        or _is_deepseek_anthropic_endpoint(base_url)
+    )
 
     last_assistant_idx = None
     for i in range(len(result) - 1, -1, -1):
@@ -1331,22 +1689,22 @@ def convert_messages_to_anthropic(
         if m.get("role") != "assistant" or not isinstance(m.get("content"), list):
             continue
 
-        if _is_kimi:
-            # Kimi's /coding endpoint enables thinking server-side and
-            # requires unsigned thinking blocks on replayed assistant
-            # tool-call messages.  Strip signed Anthropic blocks (Kimi
-            # can't validate signatures) but preserve the unsigned ones
-            # we synthesised from reasoning_content above.
+        if _preserve_unsigned_thinking:
+            # Kimi's /coding and DeepSeek's /anthropic endpoints both enable
+            # thinking server-side and require unsigned thinking blocks on
+            # replayed assistant tool-call messages.  Strip signed Anthropic
+            # blocks (neither upstream can validate Anthropic signatures) but
+            # preserve the unsigned ones we synthesised from reasoning_content.
             new_content = []
             for b in m["content"]:
                 if not isinstance(b, dict) or b.get("type") not in _THINKING_TYPES:
                     new_content.append(b)
                     continue
                 if b.get("signature") or b.get("data"):
-                    # Anthropic-signed block — Kimi can't validate, strip
+                    # Anthropic-signed block — upstream can't validate, strip
                     continue
                 # Unsigned thinking (synthesised from reasoning_content) —
-                # keep it: Kimi needs it for message-history validation.
+                # keep it: the upstream needs it for message-history validation.
                 new_content.append(b)
             m["content"] = new_content or [{"type": "text", "text": "(empty)"}]
         elif _is_third_party or idx != last_assistant_idx:
@@ -1403,6 +1761,7 @@ def build_anthropic_kwargs(
     context_length: Optional[int] = None,
     base_url: str | None = None,
     fast_mode: bool = False,
+    drop_context_1m_beta: bool = False,
 ) -> Dict[str, Any]:
     """Build kwargs for anthropic.messages.create().
 
@@ -1442,7 +1801,9 @@ def build_anthropic_kwargs(
     Currently only supported on native Anthropic endpoints (not third-party
     compatible ones).
     """
-    system, anthropic_messages = convert_messages_to_anthropic(messages, base_url=base_url)
+    system, anthropic_messages = convert_messages_to_anthropic(
+        messages, base_url=base_url, model=model
+    )
     anthropic_tools = convert_tools_to_anthropic(tools) if tools else []
 
     model = normalize_model_name(model, preserve_dots=preserve_dots)
@@ -1548,7 +1909,7 @@ def build_anthropic_kwargs(
     # silently hides reasoning text that Hermes surfaces in its CLI. We
     # request "summarized" so the reasoning blocks stay populated — matching
     # 4.6 behavior and preserving the activity-feed UX during long tool runs.
-    _is_kimi_coding = _is_kimi_coding_endpoint(base_url)
+    _is_kimi_coding = _is_kimi_family_endpoint(base_url, model)
     if reasoning_config and isinstance(reasoning_config, dict) and not _is_kimi_coding:
         if reasoning_config.get("enabled") is not False and "haiku" not in model.lower():
             effort = str(reasoning_config.get("effort", "medium")).lower()
@@ -1574,22 +1935,31 @@ def build_anthropic_kwargs(
 
     # ── Strip sampling params on 4.7+ ─────────────────────────────────
     # Opus 4.7 rejects any non-default temperature/top_p/top_k with a 400.
-    # Callers (auxiliary_client, flush_memories, etc.) may set these for
-    # older models; drop them here as a safety net so upstream 4.6 → 4.7
-    # migrations don't require coordinated edits everywhere.
+    # Callers (auxiliary_client, etc.) may set these for older models;
+    # drop them here as a safety net so upstream 4.6 → 4.7 migrations
+    # don't require coordinated edits everywhere.
     if _forbids_sampling_params(model):
         for _sampling_key in ("temperature", "top_p", "top_k"):
             kwargs.pop(_sampling_key, None)
 
     # ── Fast mode (Opus 4.6 only) ────────────────────────────────────
     # Adds extra_body.speed="fast" + the fast-mode beta header for ~2.5x
-    # output speed. Only for native Anthropic endpoints — third-party
-    # providers would reject the unknown beta header and speed parameter.
-    if fast_mode and not _is_third_party_anthropic_endpoint(base_url):
+    # output speed. Per Anthropic docs, fast mode is only supported on
+    # Opus 4.6 — Opus 4.7 and other models 400 on the speed parameter.
+    # Only for native Anthropic endpoints — third-party providers would
+    # reject the unknown beta header and speed parameter.
+    if (
+        fast_mode
+        and not _is_third_party_anthropic_endpoint(base_url)
+        and _supports_fast_mode(model)
+    ):
         kwargs.setdefault("extra_body", {})["speed"] = "fast"
         # Build extra_headers with ALL applicable betas (the per-request
         # extra_headers override the client-level anthropic-beta header).
-        betas = list(_common_betas_for_base_url(base_url))
+        betas = list(_common_betas_for_base_url(
+            base_url,
+            drop_context_1m_beta=drop_context_1m_beta,
+        ))
         if is_oauth:
             betas.extend(_OAUTH_ONLY_BETAS)
         betas.append(_FAST_MODE_BETA)
@@ -1598,4 +1968,3 @@ def build_anthropic_kwargs(
     return kwargs
 
 
-

agent/bedrock_adapter.py

@@ -87,6 +87,114 @@ def reset_client_cache():
     _bedrock_control_client_cache.clear()
 
 
+def invalidate_runtime_client(region: str) -> bool:
+    """Evict the cached ``bedrock-runtime`` client for a single region.
+
+    Per-region counterpart to :func:`reset_client_cache`. Used by the converse
+    call wrappers to discard clients whose underlying HTTP connection has
+    gone stale, so the next call allocates a fresh client (with a fresh
+    connection pool) instead of reusing a dead socket.
+
+    Returns True if a cached entry was evicted, False if the region was not
+    cached.
+    """
+    existed = region in _bedrock_runtime_client_cache
+    _bedrock_runtime_client_cache.pop(region, None)
+    return existed
+
+
+# ---------------------------------------------------------------------------
+# Stale-connection detection
+# ---------------------------------------------------------------------------
+#
+# boto3 caches its HTTPS connection pool inside the client object. When a
+# pooled connection is killed out from under us (NAT timeout, VPN flap,
+# server-side TCP RST, proxy idle cull, etc.), the next use surfaces as
+# one of a handful of low-level exceptions — most commonly
+# ``botocore.exceptions.ConnectionClosedError`` or
+# ``urllib3.exceptions.ProtocolError``. urllib3 also trips an internal
+# ``assert`` in a couple of paths (connection pool state checks, chunked
+# response readers) which bubbles up as a bare ``AssertionError`` with an
+# empty ``str(exc)``.
+#
+# In all of these cases the client is the problem, not the request: retrying
+# with the same cached client reproduces the failure until the process
+# restarts. The fix is to evict the region's cached client so the next
+# attempt builds a new one.
+
+_STALE_LIB_MODULE_PREFIXES = (
+    "urllib3.",
+    "botocore.",
+    "boto3.",
+)
+
+
+def _traceback_frames_modules(exc: BaseException):
+    """Yield ``__name__``-style module strings for each frame in exc's traceback."""
+    tb = getattr(exc, "__traceback__", None)
+    while tb is not None:
+        frame = tb.tb_frame
+        module = frame.f_globals.get("__name__", "")
+        yield module or ""
+        tb = tb.tb_next
+
+
+def is_stale_connection_error(exc: BaseException) -> bool:
+    """Return True if ``exc`` indicates a dead/stale Bedrock HTTP connection.
+
+    Matches:
+      * ``botocore.exceptions.ConnectionError`` and subclasses
+        (``ConnectionClosedError``, ``EndpointConnectionError``,
+        ``ReadTimeoutError``, ``ConnectTimeoutError``).
+      * ``urllib3.exceptions.ProtocolError`` / ``NewConnectionError`` /
+        ``ConnectionError`` (best-effort import — urllib3 is a transitive
+        dependency of botocore so it is always available in practice).
+      * Bare ``AssertionError`` raised from a frame inside urllib3, botocore,
+        or boto3. These are internal-invariant failures (typically triggered
+        by corrupted connection-pool state after a dropped socket) and are
+        recoverable by swapping the client.
+
+    Non-library ``AssertionError``s (from application code or tests) are
+    intentionally not matched — only library-internal asserts signal stale
+    connection state.
+    """
+    # botocore: the canonical signal — HTTPClientError is the umbrella for
+    # ConnectionClosedError, ReadTimeoutError, EndpointConnectionError,
+    # ConnectTimeoutError, and ProxyConnectionError. ConnectionError covers
+    # the same family via a different branch of the hierarchy.
+    try:
+        from botocore.exceptions import (
+            ConnectionError as BotoConnectionError,
+            HTTPClientError,
+        )
+        botocore_errors: tuple = (BotoConnectionError, HTTPClientError)
+    except ImportError:  # pragma: no cover — botocore always present with boto3
+        botocore_errors = ()
+    if botocore_errors and isinstance(exc, botocore_errors):
+        return True
+
+    # urllib3: low-level transport failures
+    try:
+        from urllib3.exceptions import (
+            ProtocolError,
+            NewConnectionError,
+            ConnectionError as Urllib3ConnectionError,
+        )
+        urllib3_errors = (ProtocolError, NewConnectionError, Urllib3ConnectionError)
+    except ImportError:  # pragma: no cover
+        urllib3_errors = ()
+    if urllib3_errors and isinstance(exc, urllib3_errors):
+        return True
+
+    # Library-internal AssertionError (urllib3 / botocore / boto3)
+    if isinstance(exc, AssertionError):
+        for module in _traceback_frames_modules(exc):
+            if any(module.startswith(prefix) for prefix in _STALE_LIB_MODULE_PREFIXES):
+                return True
+
+    return False
+
+
 # ---------------------------------------------------------------------------
 # AWS credential detection
 # ---------------------------------------------------------------------------
@@ -183,14 +291,52 @@ def has_aws_credentials(env: Optional[Dict[str, str]] = None) -> bool:
 def resolve_bedrock_region(env: Optional[Dict[str, str]] = None) -> str:
     """Resolve the AWS region for Bedrock API calls.
 
-    Priority: AWS_REGION → AWS_DEFAULT_REGION → us-east-1 (fallback).
+    Priority:
+      1. AWS_REGION env var
+      2. AWS_DEFAULT_REGION env var
+      3. boto3/botocore configured region (from ~/.aws/config or SSO profile)
+      4. us-east-1 (hard fallback)
+
+    The boto3 fallback is critical for EU/AP users who configure their region
+    in ~/.aws/config via a named profile rather than env vars — without it,
+    live model discovery would always return us.* profile IDs regardless of
+    the user's actual region.
     """
     env = env if env is not None else os.environ
-    return (
+    explicit = (
         env.get("AWS_REGION", "").strip()
         or env.get("AWS_DEFAULT_REGION", "").strip()
-        or "us-east-1"
     )
+    if explicit:
+        return explicit
+    try:
+        import botocore.session
+        region = botocore.session.get_session().get_config_variable("region")
+        if region:
+            return region
+    except Exception:
+        pass
+    return "us-east-1"
+
+
+def bedrock_model_ids_or_none() -> Optional[List[str]]:
+    """Live-discover Bedrock model IDs for the active region.
+
+    Returns a list of model ID strings if discovery succeeds and yields
+    at least one model, or ``None`` on failure / empty result.  Callers
+    should fall back to the static curated list when ``None`` is returned.
+
+    This helper consolidates the discover → extract-ids → fallback
+    pattern that was previously duplicated across ``provider_model_ids``,
+    ``list_authenticated_providers`` section 2, and section 3.
+    """
+    try:
+        discovered = discover_bedrock_models(resolve_bedrock_region())
+        if discovered:
+            return [m["id"] for m in discovered]
+    except Exception:
+        pass
+    return None
 
 
 # ---------------------------------------------------------------------------
@@ -787,7 +933,17 @@ def call_converse(
         guardrail_config=guardrail_config,
     )
 
-    response = client.converse(**kwargs)
+    try:
+        response = client.converse(**kwargs)
+    except Exception as exc:
+        if is_stale_connection_error(exc):
+            logger.warning(
+                "bedrock: stale-connection error on converse(region=%s, model=%s): "
+                "%s — evicting cached client so the next call reconnects.",
+                region, model, type(exc).__name__,
+            )
+            invalidate_runtime_client(region)
+        raise
     return normalize_converse_response(response)
 
 
@@ -819,7 +975,17 @@ def call_converse_stream(
         guardrail_config=guardrail_config,
     )
 
-    response = client.converse_stream(**kwargs)
+    try:
+        response = client.converse_stream(**kwargs)
+    except Exception as exc:
+        if is_stale_connection_error(exc):
+            logger.warning(
+                "bedrock: stale-connection error on converse_stream(region=%s, "
+                "model=%s): %s — evicting cached client so the next call reconnects.",
+                region, model, type(exc).__name__,
+            )
+            invalidate_runtime_client(region)
+        raise
     return normalize_converse_stream_events(response)
 
 

agent/codex_responses_adapter.py

@@ -23,26 +23,52 @@ from agent.prompt_builder import DEFAULT_AGENT_IDENTITY
 logger = logging.getLogger(__name__)
 
 
+# Matches Codex/Harmony tool-call serialization that occasionally leaks into
+# assistant-message content when the model fails to emit a structured
+# ``function_call`` item.  Accepts the common forms:
+#
+#   to=functions.exec_command
+#   assistant to=functions.exec_command
+#   <|channel|>commentary to=functions.exec_command
+#
+# ``to=functions.<name>`` is the stable marker — the optional ``assistant`` or
+# Harmony channel prefix varies by degeneration mode.  Case-insensitive to
+# cover lowercase/uppercase ``assistant`` variants.
+_TOOL_CALL_LEAK_PATTERN = re.compile(
+    r"(?:^|[\s>|])to=functions\.[A-Za-z_][\w.]*",
+    re.IGNORECASE,
+)
+
+
 # ---------------------------------------------------------------------------
 # Multimodal content helpers
 # ---------------------------------------------------------------------------
 
-def _chat_content_to_responses_parts(content: Any) -> List[Dict[str, Any]]:
+def _chat_content_to_responses_parts(content: Any, *, role: str = "user") -> List[Dict[str, Any]]:
     """Convert chat-style multimodal content to Responses API input parts.
 
     Input:  ``[{"type":"text"|"image_url", ...}]`` (native OpenAI Chat format)
-    Output: ``[{"type":"input_text"|"input_image", ...}]`` (Responses format)
+    Output: ``[{"type":"input_text"|"output_text"|"input_image", ...}]`` (Responses format)
+
+    The ``role`` parameter controls the text content type:
+    - ``"user"`` (default) → ``"input_text"``
+    - ``"assistant"`` → ``"output_text"``
+
+    The Responses API rejects ``input_text`` inside assistant messages and
+    ``output_text`` inside user messages, so callers MUST pass the correct
+    role for the message being converted.
 
     Returns an empty list when ``content`` is not a list or contains no
     recognized parts — callers fall back to the string path.
     """
+    text_type = "output_text" if role == "assistant" else "input_text"
     if not isinstance(content, list):
         return []
     converted: List[Dict[str, Any]] = []
     for part in content:
         if isinstance(part, str):
             if part:
-                converted.append({"type": "input_text", "text": part})
+                converted.append({"type": text_type, "text": part})
             continue
         if not isinstance(part, dict):
             continue
@@ -50,7 +76,7 @@ def _chat_content_to_responses_parts(content: Any) -> List[Dict[str, Any]]:
         if ptype in {"text", "input_text", "output_text"}:
             text = part.get("text")
             if isinstance(text, str) and text:
-                converted.append({"type": "input_text", "text": text})
+                converted.append({"type": text_type, "text": text})
             continue
         if ptype in {"image_url", "input_image"}:
             image_ref = part.get("image_url")
@@ -201,6 +227,23 @@ def _responses_tools(tools: Optional[List[Dict[str, Any]]] = None) -> Optional[L
 # Message format conversion
 # ---------------------------------------------------------------------------
 
+_RESPONSE_MESSAGE_STATUSES = {"completed", "incomplete", "in_progress"}
+
+
+def _normalize_responses_message_status(value: Any, *, default: str = "completed") -> str:
+    """Normalize a Responses assistant message status for replay.
+
+    The API accepts completed/incomplete/in_progress on replayed assistant
+    output messages.  Preserve those exactly (modulo case/hyphen spelling) so
+    incomplete Codex continuation turns don't get falsely marked completed.
+    """
+    if isinstance(value, str):
+        status = value.strip().lower().replace("-", "_").replace(" ", "_")
+        if status in _RESPONSE_MESSAGE_STATUSES:
+            return status
+    return default
+
+
 def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
     """Convert internal chat-style messages to Responses input items."""
     items: List[Dict[str, Any]] = []
@@ -216,9 +259,10 @@ def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Di
         if role in {"user", "assistant"}:
             content = msg.get("content", "")
             if isinstance(content, list):
-                content_parts = _chat_content_to_responses_parts(content)
+                content_parts = _chat_content_to_responses_parts(content, role=role)
+                text_type = "output_text" if role == "assistant" else "input_text"
                 content_text = "".join(
-                    p.get("text", "") for p in content_parts if p.get("type") == "input_text"
+                    p.get("text", "") for p in content_parts if p.get("type") == text_type
                 )
             else:
                 content_parts = []
@@ -245,7 +289,57 @@ def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Di
                                 seen_item_ids.add(item_id)
                             has_codex_reasoning = True
 
-                if content_parts:
+                # Replay exact assistant message items (with id/phase) from
+                # previous turns so the API can maintain prefix-cache hits.
+                # OpenAI docs: "preserve and resend phase on all assistant
+                # messages — dropping it can degrade performance."
+                codex_message_items = msg.get("codex_message_items")
+                replayed_message_items = 0
+                if isinstance(codex_message_items, list):
+                    for raw_item in codex_message_items:
+                        if not isinstance(raw_item, dict):
+                            continue
+                        if raw_item.get("type") != "message" or raw_item.get("role") != "assistant":
+                            continue
+                        raw_content_parts = raw_item.get("content")
+                        if not isinstance(raw_content_parts, list):
+                            continue
+
+                        normalized_content_parts = []
+                        for part in raw_content_parts:
+                            if not isinstance(part, dict):
+                                continue
+                            part_type = str(part.get("type") or "").strip()
+                            if part_type not in {"output_text", "text"}:
+                                continue
+                            text = part.get("text", "")
+                            if text is None:
+                                text = ""
+                            if not isinstance(text, str):
+                                text = str(text)
+                            normalized_content_parts.append({"type": "output_text", "text": text})
+
+                        if not normalized_content_parts:
+                            continue
+
+                        replay_item = {
+                            "type": "message",
+                            "role": "assistant",
+                            "status": _normalize_responses_message_status(raw_item.get("status")),
+                            "content": normalized_content_parts,
+                        }
+                        item_id = raw_item.get("id")
+                        if isinstance(item_id, str) and item_id.strip():
+                            replay_item["id"] = item_id.strip()
+                        phase = raw_item.get("phase")
+                        if isinstance(phase, str) and phase.strip():
+                            replay_item["phase"] = phase.strip()
+                        items.append(replay_item)
+                        replayed_message_items += 1
+
+                if replayed_message_items > 0:
+                    pass
+                elif content_parts:
                     items.append({"role": "assistant", "content": content_parts})
                 elif content_text.strip():
                     items.append({"role": "assistant", "content": content_text})
@@ -405,6 +499,47 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                 normalized.append(reasoning_item)
             continue
 
+        if item_type == "message":
+            role = item.get("role")
+            if role != "assistant":
+                raise ValueError(f"Codex Responses input[{idx}] message items must have role='assistant'.")
+            content = item.get("content")
+            if not isinstance(content, list):
+                raise ValueError(f"Codex Responses input[{idx}] message item must have content list.")
+            normalized_content = []
+            for part_idx, part in enumerate(content):
+                if not isinstance(part, dict):
+                    raise ValueError(
+                        f"Codex Responses input[{idx}] message content[{part_idx}] must be an object."
+                    )
+                part_type = part.get("type")
+                if part_type not in {"output_text", "text"}:
+                    raise ValueError(
+                        f"Codex Responses input[{idx}] message content[{part_idx}] has unsupported type {part_type!r}."
+                    )
+                text = part.get("text", "")
+                if text is None:
+                    text = ""
+                if not isinstance(text, str):
+                    text = str(text)
+                normalized_content.append({"type": "output_text", "text": text})
+            if not normalized_content:
+                raise ValueError(f"Codex Responses input[{idx}] message item must contain at least one text part.")
+            normalized_item: Dict[str, Any] = {
+                "type": "message",
+                "role": "assistant",
+                "status": _normalize_responses_message_status(item.get("status")),
+                "content": normalized_content,
+            }
+            item_id = item.get("id")
+            if isinstance(item_id, str) and item_id.strip():
+                normalized_item["id"] = item_id.strip()
+            phase = item.get("phase")
+            if isinstance(phase, str) and phase.strip():
+                normalized_item["phase"] = phase.strip()
+            normalized.append(normalized_item)
+            continue
+
         role = item.get("role")
         if role in {"user", "assistant"}:
             content = item.get("content", "")
@@ -412,13 +547,16 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                 content = ""
             if isinstance(content, list):
                 # Multimodal content from ``_chat_messages_to_responses_input``
-                # is already in Responses format (``input_text`` / ``input_image``).
-                # Validate each part and pass through.
+                # is already in Responses format (``input_text`` / ``output_text``
+                # / ``input_image``).  Validate each part and pass through.
+                # Use the correct text type for the role — ``output_text`` for
+                # assistant messages, ``input_text`` for user messages.
+                text_type = "output_text" if role == "assistant" else "input_text"
                 validated: List[Dict[str, Any]] = []
                 for part_idx, part in enumerate(content):
                     if isinstance(part, str):
                         if part:
-                            validated.append({"type": "input_text", "text": part})
+                            validated.append({"type": text_type, "text": part})
                         continue
                     if not isinstance(part, dict):
                         raise ValueError(
@@ -429,7 +567,7 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                         text = part.get("text", "")
                         if not isinstance(text, str):
                             text = str(text or "")
-                        validated.append({"type": "input_text", "text": text})
+                        validated.append({"type": text_type, "text": text})
                     elif ptype in {"input_image", "image_url"}:
                         image_ref = part.get("image_url", "")
                         detail = part.get("detail")
@@ -686,6 +824,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
     content_parts: List[str] = []
     reasoning_parts: List[str] = []
     reasoning_items_raw: List[Dict[str, Any]] = []
+    message_items_raw: List[Dict[str, Any]] = []
     tool_calls: List[Any] = []
     has_incomplete_items = response_status in {"queued", "in_progress", "incomplete"}
     saw_commentary_phase = False
@@ -704,6 +843,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
 
         if item_type == "message":
             item_phase = getattr(item, "phase", None)
+            normalized_phase = None
             if isinstance(item_phase, str):
                 normalized_phase = item_phase.strip().lower()
                 if normalized_phase in {"commentary", "analysis"}:
@@ -713,6 +853,18 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
             message_text = _extract_responses_message_text(item)
             if message_text:
                 content_parts.append(message_text)
+                raw_message_item: Dict[str, Any] = {
+                    "type": "message",
+                    "role": "assistant",
+                    "status": _normalize_responses_message_status(item_status),
+                    "content": [{"type": "output_text", "text": message_text}],
+                }
+                item_id = getattr(item, "id", None)
+                if isinstance(item_id, str) and item_id:
+                    raw_message_item["id"] = item_id
+                if normalized_phase:
+                    raw_message_item["phase"] = normalized_phase
+                message_items_raw.append(raw_message_item)
         elif item_type == "reasoning":
             reasoning_text = _extract_responses_reasoning_text(item)
             if reasoning_text:
@@ -787,6 +939,37 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
         if isinstance(out_text, str):
             final_text = out_text.strip()
 
+    # ── Tool-call leak recovery ──────────────────────────────────
+    # gpt-5.x on the Codex Responses API sometimes degenerates and emits
+    # what should be a structured `function_call` item as plain assistant
+    # text using the Harmony/Codex serialization (``to=functions.foo
+    # {json}`` or ``assistant to=functions.foo {json}``). The model
+    # intended to call a tool, but the intent never made it into
+    # ``response.output`` as a ``function_call`` item, so ``tool_calls``
+    # is empty here. If we pass this through, the parent sees a
+    # confident-looking summary with no audit trail (empty ``tool_trace``)
+    # and no tools actually ran — the Taiwan-embassy-email incident.
+    #
+    # Detection: leaked tokens always contain ``to=functions.<name>`` and
+    # the assistant message has no real tool calls. Treat it as incomplete
+    # so the existing Codex-incomplete continuation path (3 retries,
+    # handled in run_agent.py) gets a chance to re-elicit a proper
+    # ``function_call`` item. The existing loop already handles message
+    # append, dedup, and retry budget.
+    leaked_tool_call_text = False
+    if final_text and not tool_calls and _TOOL_CALL_LEAK_PATTERN.search(final_text):
+        leaked_tool_call_text = True
+        logger.warning(
+            "Codex response contains leaked tool-call text in assistant content "
+            "(no structured function_call items). Treating as incomplete so the "
+            "continuation path can re-elicit a proper tool call. Leaked snippet: %r",
+            final_text[:300],
+        )
+        # Clear the text so downstream code doesn't surface the garbage as
+        # a summary. The encrypted reasoning items (if any) are preserved
+        # so the model keeps its chain-of-thought on the retry.
+        final_text = ""
+
     assistant_message = SimpleNamespace(
         content=final_text,
         tool_calls=tool_calls,
@@ -794,10 +977,13 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
         reasoning_content=None,
         reasoning_details=None,
         codex_reasoning_items=reasoning_items_raw or None,
+        codex_message_items=message_items_raw or None,
     )
 
     if tool_calls:
         finish_reason = "tool_calls"
+    elif leaked_tool_call_text:
+        finish_reason = "incomplete"
     elif has_incomplete_items or (saw_commentary_phase and not saw_final_answer_phase):
         finish_reason = "incomplete"
     elif reasoning_items_raw and not final_text:

agent/context_compressor.py

@@ -61,9 +61,52 @@ _PRUNED_TOOL_PLACEHOLDER = "[Old tool output cleared to save context space]"
 
 # Chars per token rough estimate
 _CHARS_PER_TOKEN = 4
+# Flat token cost per attached image part.  Real cost varies by provider and
+# dimensions (Anthropic ≈ width×height/750, GPT-4o up to ~1700 for
+# high-detail 2048×2048, Gemini 258/tile), but 1600 is a realistic ceiling
+# that keeps compression budgeting honest for multi-image conversations.
+# Matches Claude Code's IMAGE_TOKEN_ESTIMATE constant.
+_IMAGE_TOKEN_ESTIMATE = 1600
+# Same figure expressed in the char-budget currency the rest of the
+# compressor speaks in.  Used when accumulating message "content length"
+# for tail-cut decisions.
+_IMAGE_CHAR_EQUIVALENT = _IMAGE_TOKEN_ESTIMATE * _CHARS_PER_TOKEN
 _SUMMARY_FAILURE_COOLDOWN_SECONDS = 600
 
 
+def _content_length_for_budget(raw_content: Any) -> int:
+    """Return the effective char-length of a message's content for token budgeting.
+
+    Plain strings: ``len(content)``. Multimodal lists: sum of text-part
+    ``len(text)`` plus a flat ``_IMAGE_CHAR_EQUIVALENT`` per image part
+    (``image_url`` / ``input_image`` / Anthropic-style ``image``). This
+    keeps the compressor from treating a turn with 5 attached images as
+    near-zero tokens just because the text part is empty.
+    """
+    if isinstance(raw_content, str):
+        return len(raw_content)
+    if not isinstance(raw_content, list):
+        return len(str(raw_content or ""))
+
+    total = 0
+    for p in raw_content:
+        if isinstance(p, str):
+            total += len(p)
+            continue
+        if not isinstance(p, dict):
+            total += len(str(p))
+            continue
+        ptype = p.get("type")
+        if ptype in {"image_url", "input_image", "image"}:
+            total += _IMAGE_CHAR_EQUIVALENT
+        else:
+            # text / input_text / tool_result-with-text / anything else with
+            # a text field.  Ignore the raw base64 payload inside image_url
+            # dicts — dimensions don't matter, only whether it's an image.
+            total += len(p.get("text", "") or "")
+    return total
+
+
 def _content_text_for_contains(content: Any) -> str:
     """Return a best-effort text view of message content.
 
@@ -294,8 +337,14 @@ class ContextCompressor(ContextEngine):
         self._context_probed = False
         self._context_probe_persistable = False
         self._previous_summary = None
+        self._last_summary_error = None
+        self._last_summary_dropped_count = 0
+        self._last_summary_fallback_used = False
+        self._last_aux_model_failure_error = None
+        self._last_aux_model_failure_model = None
         self._last_compression_savings_pct = 100.0
         self._ineffective_compression_count = 0
+        self._summary_failure_cooldown_until = 0.0  # transient errors must not block a fresh session
 
     def update_model(
         self,
@@ -317,6 +366,13 @@ class ContextCompressor(ContextEngine):
             int(context_length * self.threshold_percent),
             MINIMUM_CONTEXT_LENGTH,
         )
+        # Recalculate token budgets for the new context length so the
+        # compressor stays calibrated after a model switch (e.g. 200K → 32K).
+        target_tokens = int(self.threshold_tokens * self.summary_target_ratio)
+        self.tail_token_budget = target_tokens
+        self.max_summary_tokens = min(
+            int(context_length * 0.05), _SUMMARY_TOKENS_CEILING,
+        )
 
     def __init__(
         self,
@@ -389,6 +445,18 @@ class ContextCompressor(ContextEngine):
         self._last_compression_savings_pct: float = 100.0
         self._ineffective_compression_count: int = 0
         self._summary_failure_cooldown_until: float = 0.0
+        self._last_summary_error: Optional[str] = None
+        # When summary generation fails and a static fallback is inserted,
+        # record how many turns were unrecoverably dropped so callers
+        # (gateway hygiene, /compress) can surface a visible warning.
+        self._last_summary_dropped_count: int = 0
+        self._last_summary_fallback_used: bool = False
+        # When a user-configured summary model fails and we recover by
+        # retrying on the main model, record the failure so gateway /
+        # CLI callers can still warn the user even though compression
+        # succeeded.  Silent recovery would hide the broken config.
+        self._last_aux_model_failure_error: Optional[str] = None
+        self._last_aux_model_failure_model: Optional[str] = None
 
     def update_from_response(self, usage: Dict[str, Any]):
         """Update tracked token usage from API response."""
@@ -471,11 +539,11 @@ class ContextCompressor(ContextEngine):
             # Token-budget approach: walk backward accumulating tokens
             accumulated = 0
             boundary = len(result)
-            min_protect = min(protect_tail_count, len(result) - 1)
+            min_protect = min(protect_tail_count, len(result))
             for i in range(len(result) - 1, -1, -1):
                 msg = result[i]
                 raw_content = msg.get("content") or ""
-                content_len = sum(len(p.get("text", "")) for p in raw_content) if isinstance(raw_content, list) else len(raw_content)
+                content_len = _content_length_for_budget(raw_content)
                 msg_tokens = content_len // _CHARS_PER_TOKEN + 10
                 for tc in msg.get("tool_calls") or []:
                     if isinstance(tc, dict):
@@ -486,7 +554,16 @@ class ContextCompressor(ContextEngine):
                     break
                 accumulated += msg_tokens
                 boundary = i
-            prune_boundary = max(boundary, len(result) - min_protect)
+            # Translate the budget walk into a "protected count", apply the
+            # floor in count-space (where `max` reads naturally: protect at
+            # least `min_protect` messages or whatever the budget reserved,
+            # whichever is more), then convert back to a prune boundary.
+            # Doing this in index-space with `max` would invert the direction
+            # (smaller index = MORE protected), so a generous budget would
+            # silently get truncated back down to `min_protect`.
+            budget_protect_count = len(result) - boundary
+            protected_count = max(budget_protect_count, min_protect)
+            prune_boundary = len(result) - protected_count
         else:
             prune_boundary = len(result) - protect_tail_count
 
@@ -502,6 +579,8 @@ class ContextCompressor(ContextEngine):
             # Skip multimodal content (list of content blocks)
             if isinstance(content, list):
                 continue
+            if not isinstance(content, str):
+                continue
             if len(content) < 200:
                 continue
             h = hashlib.md5(content.encode("utf-8", errors="replace")).hexdigest()[:12]
@@ -521,6 +600,8 @@ class ContextCompressor(ContextEngine):
             # Skip multimodal content (list of content blocks)
             if isinstance(content, list):
                 continue
+            if not isinstance(content, str):
+                continue
             if not content or content == _PRUNED_TOOL_PLACEHOLDER:
                 continue
             # Skip already-deduplicated or previously-summarized results
@@ -812,10 +893,12 @@ The user has requested that this compaction PRIORITISE preserving all informatio
             self._previous_summary = summary
             self._summary_failure_cooldown_until = 0.0
             self._summary_model_fallen_back = False
+            self._last_summary_error = None
             return self._with_summary_prefix(summary)
         except RuntimeError:
             # No provider configured — long cooldown, unlikely to self-resolve
             self._summary_failure_cooldown_until = time.monotonic() + _SUMMARY_FAILURE_COOLDOWN_SECONDS
+            self._last_summary_error = "no auxiliary LLM provider configured"
             logging.warning("Context compression: no provider available for "
                             "summary. Middle turns will be dropped without summary "
                             "for %d seconds.",
@@ -834,25 +917,73 @@ The user has requested that this compaction PRIORITISE preserving all informatio
                 or "does not exist" in _err_str
                 or "no available channel" in _err_str
             )
+            _is_timeout = (
+                _status in (408, 429, 502, 504)
+                or "timeout" in _err_str
+            )
             if (
-                _is_model_not_found
+                (_is_model_not_found or _is_timeout)
                 and self.summary_model
                 and self.summary_model != self.model
                 and not getattr(self, "_summary_model_fallen_back", False)
             ):
                 self._summary_model_fallen_back = True
                 logging.warning(
-                    "Summary model '%s' not available (%s). "
+                    "Summary model '%s' unavailable (%s). "
                     "Falling back to main model '%s' for compression.",
                     self.summary_model, e, self.model,
                 )
+                # Record the aux-model failure so callers can warn the user
+                # even if the retry-on-main succeeds — a misconfigured aux
+                # model is something the user needs to fix.
+                _err_text = str(e).strip() or e.__class__.__name__
+                if len(_err_text) > 220:
+                    _err_text = _err_text[:217].rstrip() + "..."
+                self._last_aux_model_failure_error = _err_text
+                self._last_aux_model_failure_model = self.summary_model
                 self.summary_model = ""  # empty = use main model
                 self._summary_failure_cooldown_until = 0.0  # no cooldown
                 return self._generate_summary(turns_to_summarize, focus_topic=focus_topic)  # retry immediately
 
+            # Unknown-error best-effort retry on main model.  Losing N turns of
+            # context is almost always worse than one extra summary attempt, so
+            # if we haven't already fallen back and the summary model differs
+            # from the main model, try once more on main before entering
+            # cooldown.  Errors that DID match _is_model_not_found above are
+            # already handled by the fast-path retry; this branch catches
+            # everything else (400s, provider-specific "no route" strings,
+            # aggregator rejections, etc.) where auto-retry is still safer
+            # than dropping the turns.
+            if (
+                self.summary_model
+                and self.summary_model != self.model
+                and not getattr(self, "_summary_model_fallen_back", False)
+            ):
+                self._summary_model_fallen_back = True
+                logging.warning(
+                    "Summary model '%s' failed (%s). "
+                    "Retrying on main model '%s' before giving up.",
+                    self.summary_model, e, self.model,
+                )
+                # Record the aux-model failure (see 404 branch above) — user
+                # should know their configured model is broken even if main
+                # recovers the call.
+                _err_text = str(e).strip() or e.__class__.__name__
+                if len(_err_text) > 220:
+                    _err_text = _err_text[:217].rstrip() + "..."
+                self._last_aux_model_failure_error = _err_text
+                self._last_aux_model_failure_model = self.summary_model
+                self.summary_model = ""  # empty = use main model
+                self._summary_failure_cooldown_until = 0.0
+                return self._generate_summary(turns_to_summarize, focus_topic=focus_topic)
+
             # Transient errors (timeout, rate limit, network) — shorter cooldown
             _transient_cooldown = 60
             self._summary_failure_cooldown_until = time.monotonic() + _transient_cooldown
+            err_text = str(e).strip() or e.__class__.__name__
+            if len(err_text) > 220:
+                err_text = err_text[:217].rstrip() + "..."
+            self._last_summary_error = err_text
             logging.warning(
                 "Failed to generate context summary: %s. "
                 "Further summary attempts paused for %d seconds.",
@@ -862,15 +993,39 @@ The user has requested that this compaction PRIORITISE preserving all informatio
             return None
 
     @staticmethod
-    def _with_summary_prefix(summary: str) -> str:
-        """Normalize summary text to the current compaction handoff format."""
+    def _strip_summary_prefix(summary: str) -> str:
+        """Return summary body without the current or legacy handoff prefix."""
         text = (summary or "").strip()
-        for prefix in (LEGACY_SUMMARY_PREFIX, SUMMARY_PREFIX):
+        for prefix in (SUMMARY_PREFIX, LEGACY_SUMMARY_PREFIX):
             if text.startswith(prefix):
-                text = text[len(prefix):].lstrip()
-                break
+                return text[len(prefix):].lstrip()
+        return text
+
+    @classmethod
+    def _with_summary_prefix(cls, summary: str) -> str:
+        """Normalize summary text to the current compaction handoff format."""
+        text = cls._strip_summary_prefix(summary)
         return f"{SUMMARY_PREFIX}\n{text}" if text else SUMMARY_PREFIX
 
+    @staticmethod
+    def _is_context_summary_content(content: Any) -> bool:
+        text = _content_text_for_contains(content).lstrip()
+        return text.startswith(SUMMARY_PREFIX) or text.startswith(LEGACY_SUMMARY_PREFIX)
+
+    @classmethod
+    def _find_latest_context_summary(
+        cls,
+        messages: List[Dict[str, Any]],
+        start: int,
+        end: int,
+    ) -> tuple[Optional[int], str]:
+        """Find the newest handoff summary inside a compression window."""
+        for idx in range(end - 1, start - 1, -1):
+            content = messages[idx].get("content")
+            if cls._is_context_summary_content(content):
+                return idx, cls._strip_summary_prefix(_content_text_for_contains(content))
+        return None, ""
+
     # ------------------------------------------------------------------
     # Tool-call / tool-result pair integrity helpers
     # ------------------------------------------------------------------
@@ -879,8 +1034,8 @@ The user has requested that this compaction PRIORITISE preserving all informatio
     def _get_tool_call_id(tc) -> str:
         """Extract the call ID from a tool_call entry (dict or SimpleNamespace)."""
         if isinstance(tc, dict):
-            return tc.get("id", "")
-        return getattr(tc, "id", "") or ""
+            return tc.get("call_id", "") or tc.get("id", "") or ""
+        return getattr(tc, "call_id", "") or getattr(tc, "id", "") or ""
 
     def _sanitize_tool_pairs(self, messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
         """Fix orphaned tool_call / tool_result pairs after compression.
@@ -1067,8 +1222,9 @@ The user has requested that this compaction PRIORITISE preserving all informatio
 
         for i in range(n - 1, head_end - 1, -1):
             msg = messages[i]
-            content = msg.get("content") or ""
-            msg_tokens = len(content) // _CHARS_PER_TOKEN + 10  # +10 for role/metadata
+            raw_content = msg.get("content") or ""
+            content_len = _content_length_for_budget(raw_content)
+            msg_tokens = content_len // _CHARS_PER_TOKEN + 10  # +10 for role/metadata
             # Include tool call arguments in estimate
             for tc in msg.get("tool_calls") or []:
                 if isinstance(tc, dict):
@@ -1099,6 +1255,21 @@ The user has requested that this compaction PRIORITISE preserving all informatio
 
         return max(cut_idx, head_end + 1)
 
+    # ------------------------------------------------------------------
+    # ContextEngine: manual /compress preflight
+    # ------------------------------------------------------------------
+
+    def has_content_to_compress(self, messages: List[Dict[str, Any]]) -> bool:
+        """Return True if there is a non-empty middle region to compact.
+
+        Overrides the ABC default so the gateway ``/compress`` guard can
+        skip the LLM call when the transcript is still entirely inside
+        the protected head/tail.
+        """
+        compress_start = self._align_boundary_forward(messages, self.protect_first_n)
+        compress_end = self._find_tail_cut_by_tokens(messages, compress_start)
+        return compress_start < compress_end
+
     # ------------------------------------------------------------------
     # Main compression entry point
     # ------------------------------------------------------------------
@@ -1122,6 +1293,13 @@ The user has requested that this compaction PRIORITISE preserving all informatio
                 related to this topic and be more aggressive about compressing
                 everything else.  Inspired by Claude Code's ``/compact``.
         """
+        # Reset per-call summary failure state — callers inspect these fields
+        # after compress() returns to decide whether to surface a warning.
+        self._last_summary_dropped_count = 0
+        self._last_summary_fallback_used = False
+        self._last_summary_error = None
+        self._last_aux_model_failure_error = None
+        self._last_aux_model_failure_model = None
         n_messages = len(messages)
         # Only need head + 3 tail messages minimum (token budget decides the real tail size)
         _min_for_compress = self.protect_first_n + 3 + 1
@@ -1154,6 +1332,15 @@ The user has requested that this compaction PRIORITISE preserving all informatio
             return messages
 
         turns_to_summarize = messages[compress_start:compress_end]
+        summary_idx, summary_body = self._find_latest_context_summary(
+            messages,
+            compress_start,
+            compress_end,
+        )
+        if summary_idx is not None:
+            if summary_body and not self._previous_summary:
+                self._previous_summary = summary_body
+            turns_to_summarize = messages[summary_idx + 1:compress_end]
 
         if not self.quiet_mode:
             logger.info(
@@ -1200,11 +1387,13 @@ The user has requested that this compaction PRIORITISE preserving all informatio
             if not self.quiet_mode:
                 logger.warning("Summary generation failed — inserting static fallback context marker")
             n_dropped = compress_end - compress_start
+            self._last_summary_dropped_count = n_dropped
+            self._last_summary_fallback_used = True
             summary = (
                 f"{SUMMARY_PREFIX}\n"
-                f"Summary generation was unavailable. {n_dropped} conversation turns were "
+                f"Summary generation was unavailable. {n_dropped} message(s) were "
                 f"removed to free context space but could not be summarized. The removed "
-                f"turns contained earlier work in this session. Continue based on the "
+                f"messages contained earlier work in this session. Continue based on the "
                 f"recent messages below and the current state of any files or resources."
             )
 
@@ -1229,6 +1418,19 @@ The user has requested that this compaction PRIORITISE preserving all informatio
                 # Merge the summary into the first tail message instead
                 # of inserting a standalone message that breaks alternation.
                 _merge_summary_into_tail = True
+
+        # When the summary lands as a standalone role="user" message,
+        # weak models read the verbatim "## Active Task" quote of a past
+        # user request as fresh input (#11475, #14521). Append the explicit
+        # end marker — the same one used in the merge-into-tail path — so
+        # the model has a clear "summary above, not new input" signal.
+        if not _merge_summary_into_tail and summary_role == "user":
+            summary = (
+                summary
+                + "\n\n--- END OF CONTEXT SUMMARY — "
+                "respond to the message below, not the summary above ---"
+            )
+
         if not _merge_summary_into_tail:
             compressed.append({"role": summary_role, "content": summary})
 

agent/context_engine.py

@@ -78,6 +78,7 @@ class ContextEngine(ABC):
         self,
         messages: List[Dict[str, Any]],
         current_tokens: int = None,
+        focus_topic: str = None,
     ) -> List[Dict[str, Any]]:
         """Compact the message list and return the new message list.
 
@@ -86,6 +87,12 @@ class ContextEngine(ABC):
         context budget. The implementation is free to summarize, build a
         DAG, or do anything else — as long as the returned list is a valid
         OpenAI-format message sequence.
+
+        Args:
+            focus_topic: Optional topic string from manual ``/compress <focus>``.
+                Engines that support guided compression should prioritise
+                preserving information related to this topic.  Engines that
+                don't support it may simply ignore this argument.
         """
 
     # -- Optional: pre-flight check ----------------------------------------
@@ -98,6 +105,21 @@ class ContextEngine(ABC):
         """
         return False
 
+    # -- Optional: manual /compress preflight ------------------------------
+
+    def has_content_to_compress(self, messages: List[Dict[str, Any]]) -> bool:
+        """Quick check: is there anything in ``messages`` that can be compacted?
+
+        Used by the gateway ``/compress`` command as a preflight guard —
+        returning False lets the gateway report "nothing to compress yet"
+        without making an LLM call.
+
+        Default returns True (always attempt).  Engines with a cheap way
+        to introspect their own head/tail boundaries should override this
+        to return False when the transcript is still entirely protected.
+        """
+        return True
+
     # -- Optional: session lifecycle ---------------------------------------
 
     def on_session_start(self, session_id: str, **kwargs) -> None:

agent/copilot_acp_client.py

@@ -46,6 +46,47 @@ def _resolve_args() -> list[str]:
     return shlex.split(raw)
 
 
+def _resolve_home_dir() -> str:
+    """Return a stable HOME for child ACP processes."""
+
+    try:
+        from hermes_constants import get_subprocess_home
+
+        profile_home = get_subprocess_home()
+        if profile_home:
+            return profile_home
+    except Exception:
+        pass
+
+    home = os.environ.get("HOME", "").strip()
+    if home:
+        return home
+
+    expanded = os.path.expanduser("~")
+    if expanded and expanded != "~":
+        return expanded
+
+    try:
+        import pwd
+
+        resolved = pwd.getpwuid(os.getuid()).pw_dir.strip()
+        if resolved:
+            return resolved
+    except Exception:
+        pass
+
+    # Last resort: /tmp (writable on any POSIX system). Avoids crashing the
+    # subprocess with no HOME; callers can set HERMES_HOME explicitly if they
+    # need a different writable dir.
+    return "/tmp"
+
+
+def _build_subprocess_env() -> dict[str, str]:
+    env = os.environ.copy()
+    env["HOME"] = _resolve_home_dir()
+    return env
+
+
 def _jsonrpc_error(message_id: Any, code: int, message: str) -> dict[str, Any]:
     return {
         "jsonrpc": "2.0",
@@ -382,6 +423,7 @@ class CopilotACPClient:
                 text=True,
                 bufsize=1,
                 cwd=self._acp_cwd,
+                env=_build_subprocess_env(),
             )
         except FileNotFoundError as exc:
             raise RuntimeError(
@@ -566,7 +608,7 @@ class CopilotACPClient:
                     end = start + limit if isinstance(limit, int) and limit > 0 else None
                     content = "".join(lines[start:end])
                 if content:
-                    content = redact_sensitive_text(content)
+                    content = redact_sensitive_text(content, force=True)
                 response = {
                     "jsonrpc": "2.0",
                     "id": message_id,

agent/credential_pool.py

@@ -3,17 +3,18 @@
 from __future__ import annotations
 
 import logging
+import os
 import random
 import threading
 import time
 import uuid
-import os
 import re
 from dataclasses import dataclass, fields, replace
 from datetime import datetime
 from typing import Any, Dict, List, Optional, Set, Tuple
 
 from hermes_constants import OPENROUTER_BASE_URL
+from hermes_cli.config import get_env_value, load_env
 import hermes_cli.auth as auth_mod
 from hermes_cli.auth import (
     CODEX_ACCESS_TOKEN_REFRESH_SKEW_SECONDS,
@@ -455,6 +456,125 @@ class CredentialPool:
             logger.debug("Failed to sync from credentials file: %s", exc)
         return entry
 
+    def _sync_codex_entry_from_auth_store(self, entry: PooledCredential) -> PooledCredential:
+        """Sync a Codex device_code pool entry from auth.json if tokens differ.
+
+        When a Codex OAuth access token expires (or the ChatGPT account hits
+        its 5h/weekly quota), the pool entry gets marked ``STATUS_EXHAUSTED``
+        with a ``last_error_reset_at`` that can be many hours in the future.
+        Meanwhile the user may run ``hermes model`` / ``hermes auth`` which
+        performs a fresh device-code login and writes new tokens to
+        ``auth.json`` under ``_auth_store_lock``.  Without this sync the pool
+        entry stays frozen until ``last_error_reset_at`` elapses — even
+        though fresh credentials are sitting on disk — and every request
+        fails with "no available entries (all exhausted or empty)".
+
+        Mirrors the Nous/Anthropic resync paths above.  Only applies to
+        device_code-sourced entries; env/API-key-sourced entries have no
+        auth.json shadow to sync from.
+        """
+        if self.provider != "openai-codex" or entry.source != "device_code":
+            return entry
+        try:
+            with _auth_store_lock():
+                auth_store = _load_auth_store()
+                state = _load_provider_state(auth_store, "openai-codex")
+            if not isinstance(state, dict):
+                return entry
+            tokens = state.get("tokens")
+            if not isinstance(tokens, dict):
+                return entry
+            store_access = tokens.get("access_token", "")
+            store_refresh = tokens.get("refresh_token", "")
+            # Adopt auth.json tokens when either side differs.  Codex refresh
+            # tokens are single-use too, so a fresh refresh_token from
+            # another process means our entry's pair is consumed/stale.
+            entry_access = entry.access_token or ""
+            entry_refresh = entry.refresh_token or ""
+            if store_access and (
+                store_access != entry_access
+                or (store_refresh and store_refresh != entry_refresh)
+            ):
+                logger.debug(
+                    "Pool entry %s: syncing Codex tokens from auth.json "
+                    "(refreshed by another process)",
+                    entry.id,
+                )
+                field_updates: Dict[str, Any] = {
+                    "access_token": store_access,
+                    "refresh_token": store_refresh or entry.refresh_token,
+                    "last_status": None,
+                    "last_status_at": None,
+                    "last_error_code": None,
+                    "last_error_reason": None,
+                    "last_error_message": None,
+                    "last_error_reset_at": None,
+                }
+                if state.get("last_refresh"):
+                    field_updates["last_refresh"] = state["last_refresh"]
+                updated = replace(entry, **field_updates)
+                self._replace_entry(entry, updated)
+                self._persist()
+                return updated
+        except Exception as exc:
+            logger.debug("Failed to sync Codex entry from auth.json: %s", exc)
+        return entry
+
+    def _sync_nous_entry_from_auth_store(self, entry: PooledCredential) -> PooledCredential:
+        """Sync a Nous pool entry from auth.json if tokens differ.
+
+        Nous OAuth refresh tokens are single-use.  When another process
+        (e.g. a concurrent cron) refreshes the token via
+        ``resolve_nous_runtime_credentials``, it writes fresh tokens to
+        auth.json under ``_auth_store_lock``.  The pool entry's tokens
+        become stale.  This method detects that and adopts the newer pair,
+        avoiding a "refresh token reuse" revocation on the Nous Portal.
+        """
+        if self.provider != "nous" or entry.source != "device_code":
+            return entry
+        try:
+            with _auth_store_lock():
+                auth_store = _load_auth_store()
+                state = _load_provider_state(auth_store, "nous")
+            if not state:
+                return entry
+            store_refresh = state.get("refresh_token", "")
+            store_access = state.get("access_token", "")
+            if store_refresh and store_refresh != entry.refresh_token:
+                logger.debug(
+                    "Pool entry %s: syncing tokens from auth.json (Nous refresh token changed)",
+                    entry.id,
+                )
+                field_updates: Dict[str, Any] = {
+                    "access_token": store_access,
+                    "refresh_token": store_refresh,
+                    "last_status": None,
+                    "last_status_at": None,
+                    "last_error_code": None,
+                }
+                if state.get("expires_at"):
+                    field_updates["expires_at"] = state["expires_at"]
+                if state.get("agent_key"):
+                    field_updates["agent_key"] = state["agent_key"]
+                if state.get("agent_key_expires_at"):
+                    field_updates["agent_key_expires_at"] = state["agent_key_expires_at"]
+                if state.get("inference_base_url"):
+                    field_updates["inference_base_url"] = state["inference_base_url"]
+                extra_updates = dict(entry.extra)
+                for extra_key in ("obtained_at", "expires_in", "agent_key_id",
+                                  "agent_key_expires_in", "agent_key_reused",
+                                  "agent_key_obtained_at"):
+                    val = state.get(extra_key)
+                    if val is not None:
+                        extra_updates[extra_key] = val
+                updated = replace(entry, extra=extra_updates, **field_updates)
+                self._replace_entry(entry, updated)
+                self._persist()
+                return updated
+        except Exception as exc:
+            logger.debug("Failed to sync Nous entry from auth.json: %s", exc)
+        return entry
+
     def _sync_device_code_entry_to_auth_store(self, entry: PooledCredential) -> None:
         """Write refreshed pool entry tokens back to auth.json providers.
 
@@ -561,6 +681,9 @@ class CredentialPool:
                     last_refresh=refreshed.get("last_refresh"),
                 )
             elif self.provider == "nous":
+                synced = self._sync_nous_entry_from_auth_store(entry)
+                if synced is not entry:
+                    entry = synced
                 nous_state = {
                     "access_token": entry.access_token,
                     "refresh_token": entry.refresh_token,
@@ -635,6 +758,26 @@ class CredentialPool:
                     # Credentials file had a valid (non-expired) token — use it directly
                     logger.debug("Credentials file has valid token, using without refresh")
                     return synced
+            # For nous: another process may have consumed the refresh token
+            # between our proactive sync and the HTTP call.  Re-sync from
+            # auth.json and adopt the fresh tokens if available.
+            if self.provider == "nous":
+                synced = self._sync_nous_entry_from_auth_store(entry)
+                if synced.refresh_token != entry.refresh_token:
+                    logger.debug("Nous refresh failed but auth.json has newer tokens — adopting")
+                    updated = replace(
+                        synced,
+                        last_status=STATUS_OK,
+                        last_status_at=None,
+                        last_error_code=None,
+                        last_error_reason=None,
+                        last_error_message=None,
+                        last_error_reset_at=None,
+                    )
+                    self._replace_entry(synced, updated)
+                    self._persist()
+                    self._sync_device_code_entry_to_auth_store(updated)
+                    return updated
             self._mark_exhausted(entry, None)
             return None
 
@@ -698,6 +841,29 @@ class CredentialPool:
                 if synced is not entry:
                     entry = synced
                     cleared_any = True
+            # For nous entries, sync from auth.json before status checks.
+            # Another process may have successfully refreshed via
+            # resolve_nous_runtime_credentials(), making this entry's
+            # exhausted status stale.
+            if (self.provider == "nous"
+                    and entry.source == "device_code"
+                    and entry.last_status == STATUS_EXHAUSTED):
+                synced = self._sync_nous_entry_from_auth_store(entry)
+                if synced is not entry:
+                    entry = synced
+                    cleared_any = True
+            # For openai-codex entries, same pattern: the user may have
+            # re-authed via `hermes model` / `hermes auth` after a 429/401,
+            # leaving fresh tokens on disk while the pool entry is still
+            # frozen behind last_error_reset_at (can be hours in the
+            # future for ChatGPT weekly windows).
+            if (self.provider == "openai-codex"
+                    and entry.source == "device_code"
+                    and entry.last_status == STATUS_EXHAUSTED):
+                synced = self._sync_codex_entry_from_auth_store(entry)
+                if synced is not entry:
+                    entry = synced
+                    cleared_any = True
             if entry.last_status == STATUS_EXHAUSTED:
                 exhausted_until = _exhausted_until(entry)
                 if exhausted_until is not None and now < exhausted_until:
@@ -739,8 +905,11 @@ class CredentialPool:
 
         if self._strategy == STRATEGY_LEAST_USED and len(available) > 1:
             entry = min(available, key=lambda e: e.request_count)
+            # Increment usage counter so subsequent selections distribute load
+            updated = replace(entry, request_count=entry.request_count + 1)
+            self._replace_entry(entry, updated)
             self._current_id = entry.id
-            return entry
+            return updated
 
         if self._strategy == STRATEGY_ROUND_ROBIN and len(available) > 1:
             entry = available[0]
@@ -1056,6 +1225,18 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
                     "inference_base_url": state.get("inference_base_url"),
                     "agent_key": state.get("agent_key"),
                     "agent_key_expires_at": state.get("agent_key_expires_at"),
+                    # Carry the mint/refresh timestamps into the pool so
+                    # freshness-sensitive consumers (self-heal hooks, pool
+                    # pruning by age) can distinguish just-minted credentials
+                    # from stale ones.  Without these, fresh device_code
+                    # entries get obtained_at=None and look older than they
+                    # are (#15099).
+                    "obtained_at": state.get("obtained_at"),
+                    "expires_in": state.get("expires_in"),
+                    "agent_key_id": state.get("agent_key_id"),
+                    "agent_key_expires_in": state.get("agent_key_expires_in"),
+                    "agent_key_reused": state.get("agent_key_reused"),
+                    "agent_key_obtained_at": state.get("agent_key_obtained_at"),
                     "tls": state.get("tls") if isinstance(state.get("tls"), dict) else None,
                     "label": seeded_label,
                 },
@@ -1066,9 +1247,10 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
         # env vars (COPILOT_GITHUB_TOKEN / GH_TOKEN).  They don't live in
         # the auth store or credential pool, so we resolve them here.
         try:
-            from hermes_cli.copilot_auth import resolve_copilot_token
+            from hermes_cli.copilot_auth import resolve_copilot_token, get_copilot_api_token
             token, source = resolve_copilot_token()
             if token:
+                api_token = get_copilot_api_token(token)
                 source_name = "gh_cli" if "gh" in source.lower() else f"env:{source}"
                 if not _is_suppressed(provider, source_name):
                     active_sources.add(source_name)
@@ -1080,7 +1262,7 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
                         {
                             "source": source_name,
                             "auth_type": AUTH_TYPE_API_KEY,
-                            "access_token": token,
+                            "access_token": api_token,
                             "base_url": pconfig.inference_base_url if pconfig else "",
                             "label": source,
                         },
@@ -1118,6 +1300,48 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
         except Exception as exc:
             logger.debug("Qwen OAuth token seed failed: %s", exc)
 
+    elif provider == "minimax-oauth":
+        # MiniMax OAuth tokens live in ~/.hermes/auth.json providers.minimax-oauth.
+        # Seed the pool so `/auth list` reflects the logged-in state and the
+        # standard `hermes auth remove minimax-oauth <N>` flow works.
+        # Use refresh_if_expiring=False equivalent: resolve_minimax_oauth_runtime_credentials
+        # always refreshes on expiry, so instead read raw state here to avoid
+        # surprise network calls during provider discovery.
+        try:
+            from hermes_cli.auth import get_provider_auth_state
+            state = get_provider_auth_state("minimax-oauth")
+            if state and state.get("access_token"):
+                source_name = "oauth"
+                if not _is_suppressed(provider, source_name):
+                    active_sources.add(source_name)
+                    expires_at_ms = None
+                    try:
+                        from datetime import datetime as _dt
+                        raw = state.get("expires_at", "")
+                        if raw:
+                            expires_at_ms = int(_dt.fromisoformat(raw).timestamp() * 1000)
+                    except Exception:
+                        expires_at_ms = None
+                    base_url = str(state.get("inference_base_url", "") or "").rstrip("/")
+                    changed |= _upsert_entry(
+                        entries,
+                        provider,
+                        source_name,
+                        {
+                            "source": source_name,
+                            "auth_type": AUTH_TYPE_OAUTH,
+                            "access_token": state["access_token"],
+                            "refresh_token": state.get("refresh_token"),
+                            "expires_at_ms": expires_at_ms,
+                            "base_url": base_url,
+                            "label": state.get("label", "") or label_from_token(
+                                state.get("access_token", ""), source_name
+                            ),
+                        },
+                    )
+        except Exception as exc:
+            logger.debug("MiniMax OAuth token seed failed: %s", exc)
+
     elif provider == "openai-codex":
         # Respect user suppression — `hermes auth remove openai-codex` marks
         # the device_code source as suppressed so it won't be re-seeded from
@@ -1157,6 +1381,16 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
 def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool, Set[str]]:
     changed = False
     active_sources: Set[str] = set()
+
+    # Prefer ~/.hermes/.env over os.environ — the user's config file is the
+    # authoritative source for Hermes credentials. Stale env vars from parent
+    # processes (Codex CLI, test scripts, etc.) should not override deliberate
+    # changes to the .env file.
+    def _get_env_prefer_dotenv(key: str) -> str:
+        env_file = load_env()
+        val = env_file.get(key) or os.environ.get(key) or ""
+        return val.strip()
+
     # Honour user suppression — `hermes auth remove <provider> <N>` for an
     # env-seeded credential marks the env:<VAR> source as suppressed so it
     # won't be re-seeded from the user's shell environment or ~/.hermes/.env.
@@ -1168,7 +1402,8 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
         def _is_source_suppressed(_p, _s):  # type: ignore[misc]
             return False
     if provider == "openrouter":
-        token = os.getenv("OPENROUTER_API_KEY", "").strip()
+        # Prefer ~/.hermes/.env over os.environ
+        token = _get_env_prefer_dotenv("OPENROUTER_API_KEY")
         if token:
             source = "env:OPENROUTER_API_KEY"
             if _is_source_suppressed(provider, source):
@@ -1194,7 +1429,7 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
 
     env_url = ""
     if pconfig.base_url_env_var:
-        env_url = os.getenv(pconfig.base_url_env_var, "").strip().rstrip("/")
+        env_url = _get_env_prefer_dotenv(pconfig.base_url_env_var).rstrip("/")
 
     env_vars = list(pconfig.api_key_env_vars)
     if provider == "anthropic":
@@ -1205,7 +1440,8 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
         ]
 
     for env_var in env_vars:
-        token = os.getenv(env_var, "").strip()
+        # Prefer ~/.hermes/.env over os.environ
+        token = _get_env_prefer_dotenv(env_var)
         if not token:
             continue
         source = f"env:{env_var}"

agent/credential_sources.py

@@ -47,7 +47,6 @@ from __future__ import annotations
 
 import os
 from dataclasses import dataclass, field
-from pathlib import Path
 from typing import Callable, List, Optional
 
 
@@ -253,6 +252,19 @@ def _remove_nous_device_code(provider: str, removed) -> RemovalResult:
     return result
 
 
+def _remove_minimax_oauth(provider: str, removed) -> RemovalResult:
+    """MiniMax OAuth lives in auth.json providers.minimax-oauth — clear it.
+
+    Same pattern as Nous: single-source OAuth state with refresh tokens.
+    Suppression of the `oauth` source ensures the pool reseed path
+    (_seed_from_singletons) doesn't instantly undo the removal.
+    """
+    result = RemovalResult()
+    if _clear_auth_store_provider(provider):
+        result.cleaned.append(f"Cleared {provider} OAuth tokens from auth store")
+    return result
+
+
 def _remove_codex_device_code(provider: str, removed) -> RemovalResult:
     """Codex tokens live in TWO places: our auth store AND ~/.codex/auth.json.
 
@@ -390,6 +402,11 @@ def _register_all_sources() -> None:
         remove_fn=_remove_qwen_cli,
         description="~/.qwen/oauth_creds.json",
     ))
+    register(RemovalStep(
+        provider="minimax-oauth", source_id="oauth",
+        remove_fn=_remove_minimax_oauth,
+        description="auth.json providers.minimax-oauth",
+    ))
     register(RemovalStep(
         provider="*", source_id="config:",
         match_fn=lambda src: src.startswith("config:") or src == "model_config",

agent/curator_backup.py

@@ -0,0 +1,693 @@
+"""Curator snapshot + rollback.
+
+A pre-run snapshot of ``~/.hermes/skills/`` (excluding ``.curator_backups/``
+itself) is taken before any mutating curator pass. Snapshots are tar.gz
+files under ``~/.hermes/skills/.curator_backups/<utc-iso>/`` with a
+companion ``manifest.json`` describing the snapshot (reason, time, size,
+counted skill files). Rollback picks a snapshot, moves the current
+``skills/`` tree aside into another snapshot so even the rollback itself
+is undoable, then extracts the chosen snapshot into place.
+
+The snapshot does NOT include:
+  - ``.curator_backups/`` (would recurse)
+  - ``.hub/`` (hub-installed skills — managed by the hub, not us)
+
+It DOES include:
+  - all SKILL.md files + their directories (``scripts/``, ``references/``,
+    ``templates/``, ``assets/``)
+  - ``.usage.json`` (usage telemetry — needed to rehydrate state cleanly)
+  - ``.archive/`` (so rollback restores previously-archived skills too)
+  - ``.curator_state`` (so rolling back also restores the last-run-at
+    pointer — otherwise the curator would immediately re-fire on the next
+    tick)
+  - ``.bundled_manifest`` (so protection markers stay consistent)
+
+Alongside the skills tarball, each snapshot also captures a copy of
+``~/.hermes/cron/jobs.json`` as ``cron-jobs.json`` when it exists. Cron
+jobs reference skills by name in their ``skills``/``skill`` fields; the
+curator's consolidation pass rewrites those in place via
+``cron.jobs.rewrite_skill_refs()``. Without capturing the pre-run state,
+rolling back the skills tree would leave cron jobs pointing at the
+umbrella skills even though the narrow skills they were originally
+configured with have been restored. We store the whole jobs.json for
+fidelity but rollback only touches the ``skills``/``skill`` fields — the
+rest (schedule, next_run_at, enabled, prompt, etc.) is live state and
+we leave it alone.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import re
+import shutil
+import tarfile
+import tempfile
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple
+
+from hermes_constants import get_hermes_home
+
+logger = logging.getLogger(__name__)
+
+
+DEFAULT_KEEP = 5
+
+# Entries under skills/ that should NEVER be rolled up into a snapshot.
+# .hub/ is managed by the skills hub; rolling it back would break lockfile
+# invariants. .curator_backups is the backup dir itself — recursion bomb.
+_EXCLUDE_TOP_LEVEL = {".curator_backups", ".hub"}
+
+# Snapshot id regex: UTC ISO with colons replaced by dashes so the filename
+# is portable (Windows-safe). An optional ``-NN`` suffix handles two
+# snapshots landing in the same wallclock second.
+_ID_RE = re.compile(r"^\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}Z(-\d{2})?$")
+
+
+def _backups_dir() -> Path:
+    return get_hermes_home() / "skills" / ".curator_backups"
+
+
+def _skills_dir() -> Path:
+    return get_hermes_home() / "skills"
+
+
+def _cron_jobs_file() -> Path:
+    """Source path for the live cron jobs store (``~/.hermes/cron/jobs.json``)."""
+    return get_hermes_home() / "cron" / "jobs.json"
+
+
+CRON_JOBS_FILENAME = "cron-jobs.json"
+
+
+def _backup_cron_jobs_into(dest: Path) -> Dict[str, Any]:
+    """Copy the live cron jobs.json into ``dest`` as ``cron-jobs.json``.
+
+    Returns a small dict describing what was captured so the caller can
+    fold it into the manifest. Never raises — if the cron file is missing
+    or unreadable, the return dict has ``backed_up=False`` and the reason,
+    and the snapshot proceeds without cron data (the snapshot is still
+    useful for rolling back skills).
+    """
+    src = _cron_jobs_file()
+    info: Dict[str, Any] = {"backed_up": False, "jobs_count": 0}
+    if not src.exists():
+        info["reason"] = "no cron/jobs.json present"
+        return info
+    try:
+        raw = src.read_text(encoding="utf-8")
+    except OSError as e:
+        logger.debug("Failed to read cron/jobs.json for backup: %s", e)
+        info["reason"] = f"read error: {e}"
+        return info
+    # Count jobs as a nice diagnostic — but don't fail the snapshot if the
+    # file is unparseable; just store the raw text and let rollback deal
+    # with it (or not, if it's corrupted). jobs.json wraps the list as
+    # `{"jobs": [...], "updated_at": ...}` — we count via that shape, and
+    # fall back to bare-list shape just in case the format ever changes.
+    try:
+        parsed = json.loads(raw)
+        if isinstance(parsed, dict):
+            inner = parsed.get("jobs")
+            if isinstance(inner, list):
+                info["jobs_count"] = len(inner)
+        elif isinstance(parsed, list):
+            info["jobs_count"] = len(parsed)
+    except (json.JSONDecodeError, TypeError):
+        info["jobs_count"] = 0
+        info["parse_warning"] = "jobs.json was not valid JSON at snapshot time"
+    try:
+        (dest / CRON_JOBS_FILENAME).write_text(raw, encoding="utf-8")
+    except OSError as e:
+        logger.debug("Failed to write cron backup file: %s", e)
+        info["reason"] = f"write error: {e}"
+        return info
+    info["backed_up"] = True
+    return info
+
+
+def _utc_id(now: Optional[datetime] = None) -> str:
+    """UTC ISO-ish filesystem-safe timestamp: ``2026-05-01T13-05-42Z``."""
+    if now is None:
+        now = datetime.now(timezone.utc)
+    # isoformat → "2026-05-01T13:05:42.123456+00:00"; strip subseconds and tz.
+    s = now.replace(microsecond=0).isoformat()
+    if s.endswith("+00:00"):
+        s = s[:-6]
+    return s.replace(":", "-") + "Z"
+
+
+def _load_config() -> Dict[str, Any]:
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config()
+    except Exception as e:
+        logger.debug("Failed to load config for curator backup: %s", e)
+        return {}
+    if not isinstance(cfg, dict):
+        return {}
+    cur = cfg.get("curator") or {}
+    if not isinstance(cur, dict):
+        return {}
+    bk = cur.get("backup") or {}
+    return bk if isinstance(bk, dict) else {}
+
+
+def is_enabled() -> bool:
+    """Default ON — the whole point of the backup is safety by default."""
+    return bool(_load_config().get("enabled", True))
+
+
+def get_keep() -> int:
+    cfg = _load_config()
+    try:
+        n = int(cfg.get("keep", DEFAULT_KEEP))
+    except (TypeError, ValueError):
+        n = DEFAULT_KEEP
+    return max(1, n)
+
+
+# ---------------------------------------------------------------------------
+# Snapshot
+# ---------------------------------------------------------------------------
+
+def _count_skill_files(base: Path) -> int:
+    try:
+        return sum(1 for _ in base.rglob("SKILL.md"))
+    except OSError:
+        return 0
+
+
+def _write_manifest(dest: Path, reason: str, archive_path: Path,
+                    skills_counted: int,
+                    cron_info: Optional[Dict[str, Any]] = None) -> None:
+    manifest = {
+        "id": dest.name,
+        "reason": reason,
+        "created_at": datetime.now(timezone.utc).isoformat(),
+        "archive": archive_path.name,
+        "archive_bytes": archive_path.stat().st_size,
+        "skill_files": skills_counted,
+    }
+    if cron_info is not None:
+        manifest["cron_jobs"] = {
+            "backed_up": bool(cron_info.get("backed_up", False)),
+            "jobs_count": int(cron_info.get("jobs_count", 0)),
+        }
+        if not cron_info.get("backed_up"):
+            manifest["cron_jobs"]["reason"] = cron_info.get("reason", "not captured")
+        if cron_info.get("parse_warning"):
+            manifest["cron_jobs"]["parse_warning"] = cron_info["parse_warning"]
+    (dest / "manifest.json").write_text(
+        json.dumps(manifest, indent=2, sort_keys=True), encoding="utf-8"
+    )
+
+
+def snapshot_skills(reason: str = "manual") -> Optional[Path]:
+    """Create a tar.gz snapshot of ``~/.hermes/skills/`` and prune old ones.
+
+    Returns the snapshot directory path, or ``None`` if the snapshot was
+    skipped (backup disabled, skills dir missing, or an IO error occurred —
+    in which case we log at debug and return None so the curator never
+    aborts a pass because of a backup failure).
+    """
+    if not is_enabled():
+        logger.debug("Curator backup disabled by config; skipping snapshot")
+        return None
+
+    skills = _skills_dir()
+    if not skills.exists():
+        logger.debug("No ~/.hermes/skills/ directory — nothing to back up")
+        return None
+
+    backups = _backups_dir()
+    try:
+        backups.mkdir(parents=True, exist_ok=True)
+    except OSError as e:
+        logger.debug("Failed to create backups dir %s: %s", backups, e)
+        return None
+
+    # Uniquify: if a snapshot with the same second already exists (can
+    # happen if two curator runs fire in the same second), append a short
+    # counter. Avoids clobbering and avoids timestamp collisions.
+    base_id = _utc_id()
+    snap_id = base_id
+    counter = 1
+    while (backups / snap_id).exists():
+        snap_id = f"{base_id}-{counter:02d}"
+        counter += 1
+
+    dest = backups / snap_id
+    try:
+        dest.mkdir(parents=True, exist_ok=False)
+    except OSError as e:
+        logger.debug("Failed to create snapshot dir %s: %s", dest, e)
+        return None
+
+    archive = dest / "skills.tar.gz"
+    try:
+        # Stream into the tarball — no tempdir copy needed.
+        with tarfile.open(archive, "w:gz", compresslevel=6) as tf:
+            for entry in sorted(skills.iterdir()):
+                if entry.name in _EXCLUDE_TOP_LEVEL:
+                    continue
+                # arcname: store paths relative to skills/ so extraction
+                # drops cleanly back into the skills dir.
+                tf.add(str(entry), arcname=entry.name, recursive=True)
+        # Capture cron/jobs.json alongside the tarball. Never fails the
+        # snapshot — the skills side is the core guarantee; cron is
+        # additive. We still record in the manifest whether it was
+        # captured so rollback can surface "no cron data in this snapshot".
+        cron_info = _backup_cron_jobs_into(dest)
+        _write_manifest(dest, reason, archive,
+                        _count_skill_files(skills),
+                        cron_info=cron_info)
+    except (OSError, tarfile.TarError) as e:
+        logger.debug("Curator snapshot failed: %s", e, exc_info=True)
+        # Clean up partial snapshot
+        try:
+            shutil.rmtree(dest, ignore_errors=True)
+        except OSError:
+            pass
+        return None
+
+    _prune_old(keep=get_keep())
+    logger.info("Curator snapshot created: %s (%s)", snap_id, reason)
+    return dest
+
+
+def _prune_old(keep: int) -> List[str]:
+    """Delete regular snapshots beyond the newest *keep*. Returns deleted
+    ids. Staging dirs (``.rollback-staging-*``) are implementation detail
+    and pruned independently on every call."""
+    backups = _backups_dir()
+    if not backups.exists():
+        return []
+    entries: List[Tuple[str, Path]] = []
+    stale_staging: List[Path] = []
+    for child in backups.iterdir():
+        if not child.is_dir():
+            continue
+        if child.name.startswith(".rollback-staging-"):
+            # Staging dirs are only supposed to exist briefly during a
+            # rollback. If we find one here (e.g. from a crashed rollback),
+            # clean it up opportunistically.
+            stale_staging.append(child)
+            continue
+        if _ID_RE.match(child.name):
+            entries.append((child.name, child))
+    # Newest first (lexicographic works because the id is UTC ISO).
+    entries.sort(key=lambda t: t[0], reverse=True)
+    deleted: List[str] = []
+    for _, path in entries[keep:]:
+        try:
+            shutil.rmtree(path)
+            deleted.append(path.name)
+        except OSError as e:
+            logger.debug("Failed to prune %s: %s", path, e)
+    for path in stale_staging:
+        try:
+            shutil.rmtree(path)
+        except OSError as e:
+            logger.debug("Failed to clean stale staging dir %s: %s", path, e)
+    return deleted
+
+
+# ---------------------------------------------------------------------------
+# List + rollback
+# ---------------------------------------------------------------------------
+
+def _read_manifest(snap_dir: Path) -> Dict[str, Any]:
+    mf = snap_dir / "manifest.json"
+    if not mf.exists():
+        return {}
+    try:
+        return json.loads(mf.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError):
+        return {}
+
+
+def list_backups() -> List[Dict[str, Any]]:
+    """Return all restorable snapshots, newest first. Only entries with a
+    real ``skills.tar.gz`` tarball are listed — transient
+    ``.rollback-staging-*`` directories created mid-rollback are
+    implementation detail and not shown."""
+    backups = _backups_dir()
+    if not backups.exists():
+        return []
+    out: List[Dict[str, Any]] = []
+    for child in sorted(backups.iterdir(), reverse=True):
+        if not child.is_dir():
+            continue
+        if not _ID_RE.match(child.name):
+            continue
+        if not (child / "skills.tar.gz").exists():
+            continue
+        mf = _read_manifest(child)
+        mf.setdefault("id", child.name)
+        mf.setdefault("path", str(child))
+        if "archive_bytes" not in mf:
+            arc = child / "skills.tar.gz"
+            try:
+                mf["archive_bytes"] = arc.stat().st_size
+            except OSError:
+                mf["archive_bytes"] = 0
+        out.append(mf)
+    return out
+
+
+def _resolve_backup(backup_id: Optional[str]) -> Optional[Path]:
+    """Return the path of the requested backup, or the newest one if
+    *backup_id* is None. Returns None if no match."""
+    backups = _backups_dir()
+    if not backups.exists():
+        return None
+    if backup_id:
+        target = backups / backup_id
+        if (
+            target.is_dir()
+            and _ID_RE.match(backup_id)
+            and (target / "skills.tar.gz").exists()
+        ):
+            return target
+        return None
+    candidates = [
+        c for c in sorted(backups.iterdir(), reverse=True)
+        if c.is_dir() and _ID_RE.match(c.name) and (c / "skills.tar.gz").exists()
+    ]
+    return candidates[0] if candidates else None
+
+
+def _restore_cron_skill_links(snapshot_dir: Path) -> Dict[str, Any]:
+    """Reconcile backed-up cron skill links into the live ``cron/jobs.json``.
+
+    We do NOT overwrite the whole cron file. Only the ``skills`` and
+    ``skill`` fields are restored, and only on jobs that still exist in the
+    current file (matched by ``id``). Everything else about the job —
+    schedule, next_run_at, last_run_at, enabled, prompt, workdir, hooks —
+    is live state that the user/scheduler has modified since the snapshot;
+    overwriting it would regress unrelated cron activity.
+
+    Rules:
+    - Jobs present in backup AND live, with differing skills → skills restored.
+    - Jobs present in backup AND live, with matching skills → no-op.
+    - Jobs present in backup but gone from live (user deleted the job
+      after the snapshot) → skipped, noted in the return report.
+    - Jobs present in live but not in backup (user created a new cron
+      job after the snapshot) → left untouched.
+
+    Never raises; failures are captured in the return dict. Writes through
+    ``cron.jobs`` to pick up the same lock + atomic-write path that tick()
+    uses, so we don't race the scheduler.
+    """
+    report: Dict[str, Any] = {
+        "attempted": False,
+        "restored": [],
+        "skipped_missing": [],
+        "unchanged": 0,
+        "error": None,
+    }
+    backup_file = snapshot_dir / CRON_JOBS_FILENAME
+    if not backup_file.exists():
+        report["error"] = f"snapshot has no {CRON_JOBS_FILENAME}"
+        return report
+
+    try:
+        backup_text = backup_file.read_text(encoding="utf-8")
+        backup_parsed = json.loads(backup_text)
+    except (OSError, json.JSONDecodeError) as e:
+        report["error"] = f"failed to load backed-up jobs: {e}"
+        return report
+    # jobs.json on disk is `{"jobs": [...], "updated_at": ...}`; accept both
+    # that shape and a bare list for forward compat.
+    if isinstance(backup_parsed, dict):
+        backup_jobs = backup_parsed.get("jobs")
+    elif isinstance(backup_parsed, list):
+        backup_jobs = backup_parsed
+    else:
+        backup_jobs = None
+    if not isinstance(backup_jobs, list):
+        report["error"] = "backed-up cron-jobs.json has no jobs list"
+        return report
+
+    # Build a lookup of the backed-up skill state keyed by job id.
+    # We only need the two skill-ish fields (legacy single and modern list).
+    backup_by_id: Dict[str, Dict[str, Any]] = {}
+    for job in backup_jobs:
+        if not isinstance(job, dict):
+            continue
+        jid = job.get("id")
+        if not isinstance(jid, str) or not jid:
+            continue
+        backup_by_id[jid] = {
+            "skills": job.get("skills"),
+            "skill": job.get("skill"),
+            "name": job.get("name") or jid,
+        }
+
+    if not backup_by_id:
+        report["attempted"] = True  # we tried but there was nothing to do
+        return report
+
+    # Load and rewrite the live jobs under the scheduler's lock.
+    try:
+        from cron.jobs import load_jobs, save_jobs, _jobs_file_lock
+    except ImportError as e:
+        report["error"] = f"cron module unavailable: {e}"
+        return report
+
+    report["attempted"] = True
+    try:
+        with _jobs_file_lock:
+            live_jobs = load_jobs()
+            changed = False
+
+            live_ids = set()
+            for live in live_jobs:
+                if not isinstance(live, dict):
+                    continue
+                jid = live.get("id")
+                if not isinstance(jid, str) or not jid:
+                    continue
+                live_ids.add(jid)
+
+                backup = backup_by_id.get(jid)
+                if backup is None:
+                    continue  # live job didn't exist at snapshot time
+
+                cur_skills = live.get("skills")
+                cur_skill = live.get("skill")
+                bkp_skills = backup.get("skills")
+                bkp_skill = backup.get("skill")
+
+                if cur_skills == bkp_skills and cur_skill == bkp_skill:
+                    report["unchanged"] += 1
+                    continue
+
+                # Restore. Preserve absence (don't force the key to appear
+                # if the backup didn't have it either).
+                if bkp_skills is None:
+                    live.pop("skills", None)
+                else:
+                    live["skills"] = bkp_skills
+                if bkp_skill is None:
+                    live.pop("skill", None)
+                else:
+                    live["skill"] = bkp_skill
+
+                report["restored"].append({
+                    "job_id": jid,
+                    "job_name": backup.get("name") or jid,
+                    "from": {"skills": cur_skills, "skill": cur_skill},
+                    "to": {"skills": bkp_skills, "skill": bkp_skill},
+                })
+                changed = True
+
+            # Jobs in backup but not in live = user deleted them after snapshot
+            for jid, backup in backup_by_id.items():
+                if jid not in live_ids:
+                    report["skipped_missing"].append({
+                        "job_id": jid,
+                        "job_name": backup.get("name") or jid,
+                    })
+
+            if changed:
+                save_jobs(live_jobs)
+    except Exception as e:  # noqa: BLE001 — rollback must not die mid-restore
+        logger.debug("Cron skill-link restore failed: %s", e, exc_info=True)
+        report["error"] = f"restore failed mid-flight: {e}"
+
+    return report
+
+
+
+def rollback(backup_id: Optional[str] = None) -> Tuple[bool, str, Optional[Path]]:
+    """Restore ``~/.hermes/skills/`` from a snapshot.
+
+    Strategy:
+      1. Resolve the target snapshot (explicit id or newest regular).
+      2. Take a safety snapshot of the CURRENT skills tree under
+         ``.curator_backups/pre-rollback-<ts>/`` so the rollback itself is
+         undoable.
+      3. Move all current top-level entries (except ``.curator_backups``
+         and ``.hub``) into a tempdir.
+      4. Extract the chosen snapshot into ``~/.hermes/skills/``.
+      5. On failure during 4, move the tempdir contents back (best-effort)
+         and return failure.
+
+    Returns ``(ok, message, snapshot_path)``.
+    """
+    target = _resolve_backup(backup_id)
+    if target is None:
+        return (
+            False,
+            f"no matching backup found"
+            + (f" for id '{backup_id}'" if backup_id else "")
+            + " (use `hermes curator rollback --list` to see available snapshots)",
+            None,
+        )
+    archive = target / "skills.tar.gz"
+    if not archive.exists():
+        return (False, f"snapshot {target.name} has no skills.tar.gz — corrupted?", None)
+
+    skills = _skills_dir()
+    skills.mkdir(parents=True, exist_ok=True)
+    backups = _backups_dir()
+    backups.mkdir(parents=True, exist_ok=True)
+
+    # Step 2: safety snapshot of current state FIRST. If this fails we bail
+    # out before touching anything — otherwise a failed extract could leave
+    # the user with no skills.
+    try:
+        snapshot_skills(reason=f"pre-rollback to {target.name}")
+    except Exception as e:
+        return (False, f"pre-rollback safety snapshot failed: {e}", None)
+
+    # Additionally move current entries into an internal staging dir so
+    # the extract happens into an empty skills tree (predictable result).
+    # This dir is implementation detail — not listed as a restorable
+    # backup. The safety snapshot above is the user-facing undo handle.
+    staged = backups / f".rollback-staging-{_utc_id()}"
+    try:
+        staged.mkdir(parents=True, exist_ok=False)
+    except OSError as e:
+        return (False, f"failed to create staging dir: {e}", None)
+
+    moved: List[Tuple[Path, Path]] = []
+    try:
+        for entry in list(skills.iterdir()):
+            if entry.name in _EXCLUDE_TOP_LEVEL:
+                continue
+            dest = staged / entry.name
+            shutil.move(str(entry), str(dest))
+            moved.append((entry, dest))
+    except OSError as e:
+        # Best-effort rollback of the move
+        for orig, dest in moved:
+            try:
+                shutil.move(str(dest), str(orig))
+            except OSError:
+                pass
+        try:
+            shutil.rmtree(staged, ignore_errors=True)
+        except OSError:
+            pass
+        return (False, f"failed to stage current skills: {e}", None)
+
+    # Step 4: extract the snapshot into skills/
+    try:
+        with tarfile.open(archive, "r:gz") as tf:
+            # Python 3.12+ supports filter='data' for safer extraction.
+            # Fall back to the unfiltered call for older interpreters but
+            # still reject absolute paths and .. components defensively.
+            for member in tf.getmembers():
+                name = member.name
+                if name.startswith("/") or ".." in Path(name).parts:
+                    raise tarfile.TarError(
+                        f"refusing to extract unsafe path: {name!r}"
+                    )
+            try:
+                tf.extractall(str(skills), filter="data")  # type: ignore[call-arg]
+            except TypeError:
+                # Python < 3.12 — no filter kwarg
+                tf.extractall(str(skills))
+    except (OSError, tarfile.TarError) as e:
+        # Best-effort recover: move staged contents back
+        for orig, dest in moved:
+            try:
+                shutil.move(str(dest), str(orig))
+            except OSError:
+                pass
+        try:
+            shutil.rmtree(staged, ignore_errors=True)
+        except OSError:
+            pass
+        return (False, f"snapshot extract failed (state restored): {e}", None)
+
+    # Extract succeeded — the staging dir has served its purpose. The
+    # user's undo handle is the safety snapshot tarball we took earlier.
+    try:
+        shutil.rmtree(staged, ignore_errors=True)
+    except OSError:
+        pass
+
+    # Reconcile cron skill-links. Surgical: only the skills/skill fields
+    # on jobs matched by id. Everything else in jobs.json is live state
+    # (schedule, next_run_at, enabled, prompt, etc.) and we leave it
+    # alone. Failures here don't fail the overall rollback — the skills
+    # tree is already restored, which is the main guarantee.
+    cron_report = _restore_cron_skill_links(target)
+
+    summary_bits = [f"restored from snapshot {target.name}"]
+    if cron_report.get("attempted"):
+        restored_n = len(cron_report.get("restored") or [])
+        skipped_n = len(cron_report.get("skipped_missing") or [])
+        if cron_report.get("error"):
+            summary_bits.append(f"cron links: error — {cron_report['error']}")
+        elif restored_n == 0 and skipped_n == 0 and cron_report.get("unchanged", 0) == 0:
+            # Attempted but nothing matched — empty snapshot or no overlapping ids.
+            pass
+        else:
+            parts = []
+            if restored_n:
+                parts.append(f"{restored_n} job(s) had skill links restored")
+            if skipped_n:
+                parts.append(f"{skipped_n} backed-up job(s) no longer exist (skipped)")
+            if cron_report.get("unchanged"):
+                parts.append(f"{cron_report['unchanged']} already matched")
+            summary_bits.append("cron links: " + ", ".join(parts))
+
+    logger.info("Curator rollback: restored from %s (cron_report=%s)",
+                target.name, cron_report)
+    return (True, "; ".join(summary_bits), target)
+
+
+# ---------------------------------------------------------------------------
+# Human-readable summary for CLI
+# ---------------------------------------------------------------------------
+
+def format_size(n: int) -> str:
+    for unit in ("B", "KB", "MB", "GB"):
+        if n < 1024 or unit == "GB":
+            return f"{n:.1f} {unit}" if unit != "B" else f"{n} B"
+        n /= 1024
+    return f"{n:.1f} GB"
+
+
+def summarize_backups() -> str:
+    rows = list_backups()
+    if not rows:
+        return "No curator snapshots yet."
+    lines = [f"{'id':<24}  {'reason':<40}  {'skills':>6}  {'size':>8}"]
+    lines.append("─" * len(lines[0]))
+    for r in rows:
+        lines.append(
+            f"{r.get('id','?'):<24}  "
+            f"{(r.get('reason','?') or '?')[:40]:<40}  "
+            f"{r.get('skill_files', 0):>6}  "
+            f"{format_size(int(r.get('archive_bytes', 0))):>8}"
+        )
+    return "\n".join(lines)

agent/error_classifier.py

@@ -42,9 +42,11 @@ class FailoverReason(enum.Enum):
     # Context / payload
     context_overflow = "context_overflow"  # Context too large — compress, not failover
     payload_too_large = "payload_too_large"  # 413 — compress payload
+    image_too_large = "image_too_large"   # Native image part exceeds provider's per-image limit — shrink and retry
 
     # Model
     model_not_found = "model_not_found"  # 404 or invalid model — fallback to different model
+    provider_policy_blocked = "provider_policy_blocked"  # Aggregator (e.g. OpenRouter) blocked the only endpoint due to account data/privacy policy
 
     # Request format
     format_error = "format_error"        # 400 bad request — abort or strip + retry
@@ -52,6 +54,8 @@ class FailoverReason(enum.Enum):
     # Provider-specific
     thinking_signature = "thinking_signature"  # Anthropic thinking block sig invalid
     long_context_tier = "long_context_tier"    # Anthropic "extra usage" tier gate
+    oauth_long_context_beta_forbidden = "oauth_long_context_beta_forbidden"  # Anthropic OAuth subscription rejects 1M context beta — disable beta and retry
+    llama_cpp_grammar_pattern = "llama_cpp_grammar_pattern"  # llama.cpp json-schema-to-grammar rejects regex escapes in `pattern` / `format` — strip from tools and retry
 
     # Catch-all
     unknown = "unknown"                  # Unclassifiable — retry with backoff
@@ -89,6 +93,7 @@ class ClassifiedError:
 _BILLING_PATTERNS = [
     "insufficient credits",
     "insufficient_quota",
+    "insufficient balance",
     "credit balance",
     "credits have been exhausted",
     "top up your credits",
@@ -146,6 +151,20 @@ _PAYLOAD_TOO_LARGE_PATTERNS = [
     "error code: 413",
 ]
 
+# Image-size patterns.  Matched against 400 bodies (not 413) because most
+# providers return a 400 with a specific image-too-big message before the
+# whole request hits the 413 size limit.  Anthropic's wording is the most
+# important here (hard 5 MB per image, returned as
+# "messages.N.content.K.image.source.base64: image exceeds 5 MB maximum").
+_IMAGE_TOO_LARGE_PATTERNS = [
+    "image exceeds",        # Anthropic: "image exceeds 5 MB maximum"
+    "image too large",      # generic
+    "image_too_large",      # error_code variant
+    "image size exceeds",   # variant
+    # "request_too_large" on a request known to contain an image → image is
+    # the likely culprit; we still try the shrink path before giving up.
+]
+
 # Context overflow patterns
 _CONTEXT_OVERFLOW_PATTERNS = [
     "context length",
@@ -194,6 +213,29 @@ _MODEL_NOT_FOUND_PATTERNS = [
     "unsupported model",
 ]
 
+# OpenRouter aggregator policy-block patterns.
+#
+# When a user's OpenRouter account privacy setting (or a per-request
+# `provider.data_collection: deny` preference) excludes the only endpoint
+# serving a model, OpenRouter returns 404 with a *specific* message that is
+# distinct from "model not found":
+#
+#   "No endpoints available matching your guardrail restrictions and
+#    data policy. Configure: https://openrouter.ai/settings/privacy"
+#
+# We classify this as `provider_policy_blocked` rather than
+# `model_not_found` because:
+#   - The model *exists* — model_not_found is misleading in logs
+#   - Provider fallback won't help: the account-level setting applies to
+#     every call on the same OpenRouter account
+#   - The error body already contains the fix URL, so the user gets
+#     actionable guidance without us rewriting the message
+_PROVIDER_POLICY_BLOCKED_PATTERNS = [
+    "no endpoints available matching your guardrail",
+    "no endpoints available matching your data policy",
+    "no endpoints found matching your data policy",
+]
+
 # Auth patterns (non-status-code signals)
 _AUTH_PATTERNS = [
     "invalid api key",
@@ -319,6 +361,11 @@ def classify_api_error(
     """
     status_code = _extract_status_code(error)
     error_type = type(error).__name__
+    # Copilot/GitHub Models RateLimitError may not set .status_code; force 429
+    # so downstream rate-limit handling (classifier reason, pool rotation,
+    # fallback gating) fires correctly instead of misclassifying as generic.
+    if status_code is None and error_type == "RateLimitError":
+        status_code = 429
     body = _extract_error_body(error)
     error_code = _extract_error_code(body)
 
@@ -405,6 +452,50 @@ def classify_api_error(
             should_compress=True,
         )
 
+    # Anthropic OAuth subscription rejects the 1M-context beta header.
+    # Observed error body: "The long context beta is not yet available for
+    # this subscription." Returned as HTTP 400 from native Anthropic when
+    # the subscription doesn't include 1M context, even though the request
+    # carries ``anthropic-beta: context-1m-2025-08-07``. The recovery path
+    # in run_agent.py rebuilds the Anthropic client with the beta stripped
+    # and retries once. Pattern is narrow enough that it won't collide with
+    # the 429 tier-gate pattern above (different status, different phrase).
+    if (
+        status_code == 400
+        and "long context beta" in error_msg
+        and "not yet available" in error_msg
+    ):
+        return _result(
+            FailoverReason.oauth_long_context_beta_forbidden,
+            retryable=True,
+            should_compress=False,
+        )
+
+    # llama.cpp's ``json-schema-to-grammar`` converter (used by its OAI
+    # server to build GBNF tool-call parsers) rejects regex escape classes
+    # like ``\d``/``\w``/``\s`` and most ``format`` values. MCP servers
+    # routinely emit ``"pattern": "\\d{4}-\\d{2}-\\d{2}"`` for date/phone/
+    # email params. llama.cpp surfaces this as HTTP 400 with one of a few
+    # recognizable phrases; on match we strip ``pattern``/``format`` from
+    # ``self.tools`` in the retry loop and retry once. Cloud providers are
+    # unaffected — they accept these keywords and we never hit this branch.
+    if (
+        status_code == 400
+        and (
+            "error parsing grammar" in error_msg
+            or "json-schema-to-grammar" in error_msg
+            or (
+                "unable to generate parser" in error_msg
+                and "template" in error_msg
+            )
+        )
+    ):
+        return _result(
+            FailoverReason.llama_cpp_grammar_pattern,
+            retryable=True,
+            should_compress=False,
+        )
+
     # ── 2. HTTP status code classification ──────────────────────────
 
     if status_code is not None:
@@ -455,7 +546,12 @@ def classify_api_error(
 
     is_disconnect = any(p in error_msg for p in _SERVER_DISCONNECT_PATTERNS)
     if is_disconnect and not status_code:
-        is_large = approx_tokens > context_length * 0.6 or approx_tokens > 120000 or num_messages > 200
+        # Absolute token/message-count thresholds are only a proxy for smaller
+        # context windows.  Large-context sessions can have hundreds of
+        # messages while still being far below their actual token budget.
+        is_large = approx_tokens > context_length * 0.6 or (
+            context_length <= 256000 and (approx_tokens > 120000 or num_messages > 200)
+        )
         if is_large:
             return _result(
                 FailoverReason.context_overflow,
@@ -523,6 +619,17 @@ def _classify_by_status(
         return _classify_402(error_msg, result_fn)
 
     if status_code == 404:
+        # OpenRouter policy-block 404 — distinct from "model not found".
+        # The model exists; the user's account privacy setting excludes the
+        # only endpoint serving it. Falling back to another provider won't
+        # help (same account setting applies).  The error body already
+        # contains the fix URL, so just surface it.
+        if any(p in error_msg for p in _PROVIDER_POLICY_BLOCKED_PATTERNS):
+            return result_fn(
+                FailoverReason.provider_policy_blocked,
+                retryable=False,
+                should_fallback=False,
+            )
         if any(p in error_msg for p in _MODEL_NOT_FOUND_PATTERNS):
             return result_fn(
                 FailoverReason.model_not_found,
@@ -631,6 +738,15 @@ def _classify_400(
 ) -> ClassifiedError:
     """Classify 400 Bad Request — context overflow, format error, or generic."""
 
+    # Image-too-large from 400 (Anthropic's 5 MB per-image check fires this way).
+    # Must be checked BEFORE context_overflow because messages can trip both
+    # patterns ("exceeds" + "image") and image-shrink is a cheaper recovery.
+    if any(p in error_msg for p in _IMAGE_TOO_LARGE_PATTERNS):
+        return result_fn(
+            FailoverReason.image_too_large,
+            retryable=True,
+        )
+
     # Context overflow from 400
     if any(p in error_msg for p in _CONTEXT_OVERFLOW_PATTERNS):
         return result_fn(
@@ -640,6 +756,12 @@ def _classify_400(
         )
 
     # Some providers return model-not-found as 400 instead of 404 (e.g. OpenRouter).
+    if any(p in error_msg for p in _PROVIDER_POLICY_BLOCKED_PATTERNS):
+        return result_fn(
+            FailoverReason.provider_policy_blocked,
+            retryable=False,
+            should_fallback=False,
+        )
     if any(p in error_msg for p in _MODEL_NOT_FOUND_PATTERNS):
         return result_fn(
             FailoverReason.model_not_found,
@@ -675,7 +797,12 @@ def _classify_400(
         if not err_body_msg:
             err_body_msg = str(body.get("message") or "").strip().lower()
     is_generic = len(err_body_msg) < 30 or err_body_msg in ("error", "")
-    is_large = approx_tokens > context_length * 0.4 or approx_tokens > 80000 or num_messages > 80
+    # Absolute token/message-count thresholds are only a proxy for smaller
+    # context windows.  Large-context sessions can have many messages while
+    # still being far below their actual token budget.
+    is_large = approx_tokens > context_length * 0.4 or (
+        context_length <= 256000 and (approx_tokens > 80000 or num_messages > 80)
+    )
 
     if is_generic and is_large:
         return result_fn(
@@ -752,6 +879,13 @@ def _classify_by_message(
             should_compress=True,
         )
 
+    # Image-too-large patterns (from message text when no status_code)
+    if any(p in error_msg for p in _IMAGE_TOO_LARGE_PATTERNS):
+        return result_fn(
+            FailoverReason.image_too_large,
+            retryable=True,
+        )
+
     # Usage-limit patterns need the same disambiguation as 402: some providers
     # surface "usage limit" errors without an HTTP status code.  A transient
     # signal ("try again", "resets at", …) means it's a periodic quota, not
@@ -812,6 +946,15 @@ def _classify_by_message(
             should_fallback=True,
         )
 
+    # Provider policy-block (aggregator-side guardrail) — check before
+    # model_not_found so we don't mis-label as a missing model.
+    if any(p in error_msg for p in _PROVIDER_POLICY_BLOCKED_PATTERNS):
+        return result_fn(
+            FailoverReason.provider_policy_blocked,
+            retryable=False,
+            should_fallback=False,
+        )
+
     # Model not found patterns
     if any(p in error_msg for p in _MODEL_NOT_FOUND_PATTERNS):
         return result_fn(

agent/gemini_cloudcode_adapter.py

@@ -30,7 +30,6 @@ from __future__ import annotations
 
 import json
 import logging
-import os
 import time
 import uuid
 from types import SimpleNamespace
@@ -42,7 +41,6 @@ from agent import google_oauth
 from agent.gemini_schema import sanitize_gemini_tool_parameters
 from agent.google_code_assist import (
     CODE_ASSIST_ENDPOINT,
-    FREE_TIER_ID,
     CodeAssistError,
     ProjectContext,
     resolve_project_context,

agent/gemini_native_adapter.py

@@ -44,6 +44,97 @@ def is_native_gemini_base_url(base_url: str) -> bool:
     return not normalized.endswith("/openai")
 
 
+def probe_gemini_tier(
+    api_key: str,
+    base_url: str = DEFAULT_GEMINI_BASE_URL,
+    *,
+    model: str = "gemini-2.5-flash",
+    timeout: float = 10.0,
+) -> str:
+    """Probe a Google AI Studio API key and return its tier.
+
+    Returns one of:
+
+    - ``"free"``    -- key is on the free tier (unusable with Hermes)
+    - ``"paid"``    -- key is on a paid tier
+    - ``"unknown"`` -- probe failed; callers should proceed without blocking.
+    """
+    key = (api_key or "").strip()
+    if not key:
+        return "unknown"
+
+    normalized_base = str(base_url or DEFAULT_GEMINI_BASE_URL).strip().rstrip("/")
+    if not normalized_base:
+        normalized_base = DEFAULT_GEMINI_BASE_URL
+    if normalized_base.lower().endswith("/openai"):
+        normalized_base = normalized_base[: -len("/openai")]
+
+    url = f"{normalized_base}/models/{model}:generateContent"
+    payload = {
+        "contents": [{"role": "user", "parts": [{"text": "hi"}]}],
+        "generationConfig": {"maxOutputTokens": 1},
+    }
+
+    try:
+        with httpx.Client(timeout=timeout) as client:
+            resp = client.post(
+                url,
+                params={"key": key},
+                json=payload,
+                headers={"Content-Type": "application/json"},
+            )
+    except Exception as exc:
+        logger.debug("probe_gemini_tier: network error: %s", exc)
+        return "unknown"
+
+    headers_lower = {k.lower(): v for k, v in resp.headers.items()}
+    rpd_header = headers_lower.get("x-ratelimit-limit-requests-per-day")
+    if rpd_header:
+        try:
+            rpd_val = int(rpd_header)
+        except (TypeError, ValueError):
+            rpd_val = None
+        # Published free-tier daily caps (Dec 2025):
+        #   gemini-2.5-pro: 100, gemini-2.5-flash: 250, flash-lite: 1000
+        # Tier 1 starts at ~1500+ for Flash. We treat <= 1000 as free.
+        if rpd_val is not None and rpd_val <= 1000:
+            return "free"
+        if rpd_val is not None and rpd_val > 1000:
+            return "paid"
+
+    if resp.status_code == 429:
+        body_text = ""
+        try:
+            body_text = resp.text or ""
+        except Exception:
+            body_text = ""
+        if "free_tier" in body_text.lower():
+            return "free"
+        return "paid"
+
+    if 200 <= resp.status_code < 300:
+        return "paid"
+
+    return "unknown"
+
+
+def is_free_tier_quota_error(error_message: str) -> bool:
+    """Return True when a Gemini 429 message indicates free-tier exhaustion."""
+    if not error_message:
+        return False
+    return "free_tier" in error_message.lower()
+
+
+_FREE_TIER_GUIDANCE = (
+    "\n\nYour Google API key is on the free tier (<= 250 requests/day for "
+    "gemini-2.5-flash). Hermes typically makes 3-10 API calls per user turn, "
+    "so the free tier is exhausted in a handful of messages and cannot sustain "
+    "an agent session. Enable billing on your Google Cloud project and "
+    "regenerate the key in a billing-enabled project: "
+    "https://aistudio.google.com/apikey"
+)
+
+
 class GeminiAPIError(Exception):
     """Error shape compatible with Hermes retry/error classification."""
 
@@ -588,7 +679,21 @@ def translate_stream_event(event: Dict[str, Any], model: str, tool_call_indices:
     finish_reason_raw = str(cand.get("finishReason") or "")
     if finish_reason_raw:
         mapped = "tool_calls" if tool_call_indices else _map_gemini_finish_reason(finish_reason_raw)
-        chunks.append(_make_stream_chunk(model=model, finish_reason=mapped))
+        finish_chunk = _make_stream_chunk(model=model, finish_reason=mapped)
+        # Attach usage from this event's usageMetadata so the streaming
+        # loop in run_agent.py can record token counts (mirrors the
+        # non-streaming path in translate_gemini_response).
+        usage_meta = event.get("usageMetadata") or {}
+        if usage_meta:
+            finish_chunk.usage = SimpleNamespace(
+                prompt_tokens=int(usage_meta.get("promptTokenCount") or 0),
+                completion_tokens=int(usage_meta.get("candidatesTokenCount") or 0),
+                total_tokens=int(usage_meta.get("totalTokenCount") or 0),
+                prompt_tokens_details=SimpleNamespace(
+                    cached_tokens=int(usage_meta.get("cachedContentTokenCount") or 0),
+                ),
+            )
+        chunks.append(finish_chunk)
     return chunks
 
 
@@ -650,6 +755,12 @@ def gemini_http_error(response: httpx.Response) -> GeminiAPIError:
     else:
         message = f"Gemini returned HTTP {status}: {body_text[:500]}"
 
+    # Free-tier quota exhaustion -> append actionable guidance so users who
+    # bypassed the setup wizard (direct GOOGLE_API_KEY in .env) still learn
+    # that the free tier cannot sustain an agent session.
+    if status == 429 and is_free_tier_quota_error(err_message or body_text):
+        message = message + _FREE_TIER_GUIDANCE
+
     return GeminiAPIError(
         message,
         code=code,
@@ -704,6 +815,13 @@ class GeminiNativeClient:
         http_client: Optional[httpx.Client] = None,
         **_: Any,
     ) -> None:
+        if not (api_key or "").strip():
+            raise RuntimeError(
+                "Gemini native client requires an API key, but none was provided. "
+                "Set GOOGLE_API_KEY or GEMINI_API_KEY in your environment / ~/.hermes/.env "
+                "(get one at https://aistudio.google.com/app/apikey), or run `hermes setup` "
+                "to configure the Google provider."
+            )
         self.api_key = api_key
         normalized_base = (base_url or DEFAULT_GEMINI_BASE_URL).rstrip("/")
         if normalized_base.endswith("/openai"):

agent/gemini_schema.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from typing import Any, Dict, List
+from typing import Any, Dict
 
 # Gemini's ``FunctionDeclaration.parameters`` field accepts the ``Schema``
 # object, which is only a subset of OpenAPI 3.0 / JSON Schema.  Strip fields
@@ -73,6 +73,20 @@ def sanitize_gemini_schema(schema: Any) -> Dict[str, Any]:
             ]
             continue
         cleaned[key] = value
+
+    # Gemini's Schema validator requires every ``enum`` entry to be a string,
+    # even when the parent ``type`` is ``integer`` / ``number`` / ``boolean``.
+    # OpenAI / OpenRouter / Anthropic accept typed enums (e.g. Discord's
+    # ``auto_archive_duration: {type: integer, enum: [60, 1440, 4320, 10080]}``),
+    # so we only drop the ``enum`` when it would collide with Gemini's rule.
+    # Keeping ``type: integer`` plus the human-readable description gives the
+    # model enough guidance; the tool handler still validates the value.
+    enum_val = cleaned.get("enum")
+    type_val = cleaned.get("type")
+    if isinstance(enum_val, list) and type_val in {"integer", "number", "boolean"}:
+        if any(not isinstance(item, str) for item in enum_val):
+            cleaned.pop("enum", None)
+
     return cleaned
 
 

agent/google_code_assist.py

@@ -29,7 +29,6 @@ from __future__ import annotations
 
 import json
 import logging
-import os
 import time
 import urllib.error
 import urllib.parse

agent/google_oauth.py

@@ -49,14 +49,13 @@ import json
 import logging
 import os
 import secrets
-import socket
 import stat
 import threading
 import time
 import urllib.error
 import urllib.parse
 import urllib.request
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from pathlib import Path
 from typing import Any, Dict, Optional, Tuple
 
@@ -98,6 +97,7 @@ _DEFAULT_CLIENT_SECRET = f"GOCSPX-{_PUBLIC_CLIENT_SECRET_SUFFIX}"
 
 # Regex patterns for fallback scraping from an installed gemini-cli.
 import re as _re
+from utils import atomic_replace
 _CLIENT_ID_PATTERN = _re.compile(
     r"OAUTH_CLIENT_ID\s*=\s*['\"]([0-9]+-[a-z0-9]+\.apps\.googleusercontent\.com)['\"]"
 )
@@ -489,17 +489,30 @@ def save_credentials(creds: GoogleCredentials) -> Path:
     """Atomically write creds to disk with 0o600 permissions."""
     path = _credentials_path()
     path.parent.mkdir(parents=True, exist_ok=True)
+    # Tighten parent dir to 0o700 so siblings can't traverse to the creds file.
+    # On Windows this is a no-op (POSIX mode bits aren't enforced); ignore failures.
+    try:
+        os.chmod(path.parent, 0o700)
+    except OSError:
+        pass
     payload = json.dumps(creds.to_dict(), indent=2, sort_keys=True) + "\n"
 
     with _credentials_lock():
         tmp_path = path.with_suffix(f".tmp.{os.getpid()}.{secrets.token_hex(4)}")
         try:
-            with open(tmp_path, "w", encoding="utf-8") as fh:
+            # Create with 0o600 atomically to close the TOCTOU window where the
+            # default umask (often 0o644) would briefly expose tokens to other
+            # local users between open() and chmod().
+            fd = os.open(
+                str(tmp_path),
+                os.O_WRONLY | os.O_CREAT | os.O_EXCL,
+                stat.S_IRUSR | stat.S_IWUSR,
+            )
+            with os.fdopen(fd, "w", encoding="utf-8") as fh:
                 fh.write(payload)
                 fh.flush()
                 os.fsync(fh.fileno())
-            os.chmod(tmp_path, stat.S_IRUSR | stat.S_IWUSR)
-            os.replace(tmp_path, path)
+            atomic_replace(tmp_path, path)
         finally:
             try:
                 if tmp_path.exists():

agent/i18n.py

@@ -0,0 +1,230 @@
+"""Lightweight internationalization (i18n) for Hermes static user-facing messages.
+
+Scope (thin slice, by design): only the highest-impact static strings shown
+to the user by Hermes itself -- approval prompts, a handful of gateway slash
+command replies, restart-drain notices.  Agent-generated output, log lines,
+error tracebacks, tool outputs, and slash-command descriptions all stay in
+English.
+
+Catalog files live under ``locales/<lang>.yaml`` at the repo root.  Each
+catalog is a flat dict keyed by dotted paths (e.g. ``approval.choose`` or
+``gateway.approval_expired``).  Missing keys fall back to English; if English
+is missing too, the key path itself is returned so a broken catalog never
+crashes the agent.
+
+Usage::
+
+    from agent.i18n import t
+    print(t("approval.choose_long"))                       # current lang
+    print(t("gateway.draining", count=3))                  # {count} formatted
+    print(t("approval.choose_long", lang="zh"))            # explicit override
+
+Language resolution order:
+    1. Explicit ``lang=`` argument passed to :func:`t`
+    2. ``HERMES_LANGUAGE`` environment variable (for tests / quick override)
+    3. ``display.language`` from config.yaml
+    4. ``"en"`` (baseline)
+
+Supported languages: en, zh, ja, de, es.  Unknown values fall back to en.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import threading
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+SUPPORTED_LANGUAGES: tuple[str, ...] = ("en", "zh", "ja", "de", "es")
+DEFAULT_LANGUAGE = "en"
+
+# Accept a few natural aliases so users who type "chinese" / "zh-CN" / "jp"
+# get the right catalog instead of silently falling back to English.
+_LANGUAGE_ALIASES: dict[str, str] = {
+    "english": "en", "en-us": "en", "en-gb": "en",
+    "chinese": "zh", "mandarin": "zh", "zh-cn": "zh", "zh-tw": "zh", "zh-hans": "zh", "zh-hant": "zh",
+    "japanese": "ja", "jp": "ja", "ja-jp": "ja",
+    "german": "de", "deutsch": "de", "de-de": "de",
+    "spanish": "es", "español": "es", "espanol": "es", "es-es": "es", "es-mx": "es",
+}
+
+_catalog_cache: dict[str, dict[str, str]] = {}
+_catalog_lock = threading.Lock()
+
+
+def _locales_dir() -> Path:
+    """Return the directory containing locale YAML files.
+
+    Lives next to the repo root so both the bundled install and editable
+    checkouts find it without PYTHONPATH gymnastics.
+    """
+    # agent/i18n.py -> agent/ -> repo root
+    return Path(__file__).resolve().parent.parent / "locales"
+
+
+def _normalize_lang(value: Any) -> str:
+    """Normalize a user-supplied language value to a supported code.
+
+    Accepts supported codes directly, common aliases (``chinese`` -> ``zh``),
+    and case-insensitive regional tags (``zh-CN`` -> ``zh``).  Returns the
+    default language for unknown values.
+    """
+    if not isinstance(value, str):
+        return DEFAULT_LANGUAGE
+    key = value.strip().lower()
+    if not key:
+        return DEFAULT_LANGUAGE
+    if key in SUPPORTED_LANGUAGES:
+        return key
+    if key in _LANGUAGE_ALIASES:
+        return _LANGUAGE_ALIASES[key]
+    # Try stripping a region suffix (e.g. "pt-br" -> "pt" won't be supported,
+    # but "zh-CN" -> "zh" will).
+    base = key.split("-", 1)[0]
+    if base in SUPPORTED_LANGUAGES:
+        return base
+    return DEFAULT_LANGUAGE
+
+
+def _load_catalog(lang: str) -> dict[str, str]:
+    """Load and flatten one locale YAML file into a dotted-key dict.
+
+    YAML files can be nested for human readability; this produces the flat
+    key space :func:`t` expects.  Cached per-language for the process.
+    """
+    with _catalog_lock:
+        cached = _catalog_cache.get(lang)
+        if cached is not None:
+            return cached
+
+    path = _locales_dir() / f"{lang}.yaml"
+    if not path.is_file():
+        logger.debug("i18n catalog missing for %s at %s", lang, path)
+        with _catalog_lock:
+            _catalog_cache[lang] = {}
+        return {}
+
+    try:
+        import yaml  # PyYAML is already a hermes dependency
+        with path.open("r", encoding="utf-8") as f:
+            raw = yaml.safe_load(f) or {}
+    except Exception as exc:
+        logger.warning("Failed to load i18n catalog %s: %s", path, exc)
+        with _catalog_lock:
+            _catalog_cache[lang] = {}
+        return {}
+
+    flat: dict[str, str] = {}
+    _flatten_into(raw, "", flat)
+    with _catalog_lock:
+        _catalog_cache[lang] = flat
+    return flat
+
+
+def _flatten_into(node: Any, prefix: str, out: dict[str, str]) -> None:
+    if isinstance(node, dict):
+        for key, value in node.items():
+            child_key = f"{prefix}.{key}" if prefix else str(key)
+            _flatten_into(value, child_key, out)
+    elif isinstance(node, str):
+        out[prefix] = node
+    # Non-string, non-dict leaves are ignored -- catalogs are text-only.
+
+
+@lru_cache(maxsize=1)
+def _config_language_cached() -> str | None:
+    """Read ``display.language`` from config.yaml once per process.
+
+    Cached because ``t()`` is called in hot paths (every approval prompt,
+    every gateway reply) and re-reading YAML each call would be wasteful.
+    ``reset_language_cache()`` clears this when config changes at runtime
+    (e.g. after the setup wizard).
+    """
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config()
+        lang = (cfg.get("display") or {}).get("language")
+        if lang:
+            return _normalize_lang(lang)
+    except Exception as exc:
+        logger.debug("Could not read display.language from config: %s", exc)
+    return None
+
+
+def reset_language_cache() -> None:
+    """Invalidate cached language resolution and catalogs.
+
+    Call after :func:`hermes_cli.config.save_config` if a running process
+    needs to pick up a changed ``display.language`` without restart.
+    """
+    _config_language_cached.cache_clear()
+    with _catalog_lock:
+        _catalog_cache.clear()
+
+
+def get_language() -> str:
+    """Resolve the active language using env > config > default order."""
+    env_lang = os.environ.get("HERMES_LANGUAGE")
+    if env_lang:
+        return _normalize_lang(env_lang)
+    cfg_lang = _config_language_cached()
+    if cfg_lang:
+        return cfg_lang
+    return DEFAULT_LANGUAGE
+
+
+def t(key: str, lang: str | None = None, **format_kwargs: Any) -> str:
+    """Translate a dotted key to the active language.
+
+    Parameters
+    ----------
+    key
+        Dotted path into the catalog, e.g. ``"approval.choose_long"``.
+    lang
+        Explicit language override.  Takes precedence over env + config.
+    **format_kwargs
+        ``str.format`` substitution arguments (``t("gateway.drain", count=3)``
+        expects a catalog entry with a ``{count}`` placeholder).
+
+    Returns
+    -------
+    The translated string, or the English fallback if the key is missing in
+    the target language, or the bare key if English is also missing.
+    """
+    target = _normalize_lang(lang) if lang else get_language()
+    catalog = _load_catalog(target)
+    value = catalog.get(key)
+
+    if value is None and target != DEFAULT_LANGUAGE:
+        # Fall through to English rather than showing a key path to the user.
+        value = _load_catalog(DEFAULT_LANGUAGE).get(key)
+
+    if value is None:
+        # Last-ditch: return the key itself.  A broken catalog should not
+        # crash anything; it just looks ugly until someone fixes it.
+        logger.debug("i18n miss: key=%r lang=%r", key, target)
+        value = key
+
+    if format_kwargs:
+        try:
+            return value.format(**format_kwargs)
+        except (KeyError, IndexError, ValueError) as exc:
+            logger.warning(
+                "i18n format failed for key=%r lang=%r kwargs=%r: %s",
+                key, target, format_kwargs, exc,
+            )
+            return value
+    return value
+
+
+__all__ = [
+    "SUPPORTED_LANGUAGES",
+    "DEFAULT_LANGUAGE",
+    "t",
+    "get_language",
+    "reset_language_cache",
+]

agent/image_routing.py

@@ -0,0 +1,236 @@
+"""Routing helpers for inbound user-attached images.
+
+Two modes:
+
+  native  — attach images as OpenAI-style ``image_url`` content parts on the
+            user turn. Provider adapters (Anthropic, Gemini, Bedrock, Codex,
+            OpenAI chat.completions) already translate these into their
+            vendor-specific multimodal formats.
+
+  text    — run ``vision_analyze`` on each image up-front and prepend the
+            description to the user's text. The model never sees the pixels;
+            it only sees a lossy text summary. This is the pre-existing
+            behaviour and still the right choice for non-vision models.
+
+The decision is made once per message turn by :func:`decide_image_input_mode`.
+It reads ``agent.image_input_mode`` from config.yaml (``auto`` | ``native``
+| ``text``, default ``auto``) and the active model's capability metadata.
+
+In ``auto`` mode:
+  - If the user has explicitly configured ``auxiliary.vision.provider``
+    (i.e. not ``auto`` and not empty), we assume they want the text pipeline
+    regardless of the main model — they've opted in to a specific vision
+    backend for a reason (cost, quality, local-only, etc.).
+  - Otherwise, if the active model reports ``supports_vision=True`` in its
+    models.dev metadata, we attach natively.
+  - Otherwise (non-vision model, no explicit override), we fall back to text.
+
+This keeps ``vision_analyze`` surfaced as a tool in every session — skills
+and agent flows that chain it (browser screenshots, deeper inspection of
+URL-referenced images, style-gating loops) keep working. The routing only
+affects *how user-attached images on the current turn* are presented to the
+main model.
+"""
+
+from __future__ import annotations
+
+import base64
+import logging
+import mimetypes
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple
+
+logger = logging.getLogger(__name__)
+
+
+_VALID_MODES = frozenset({"auto", "native", "text"})
+
+
+def _coerce_mode(raw: Any) -> str:
+    """Normalize a config value into one of the valid modes."""
+    if not isinstance(raw, str):
+        return "auto"
+    val = raw.strip().lower()
+    if val in _VALID_MODES:
+        return val
+    return "auto"
+
+
+def _explicit_aux_vision_override(cfg: Optional[Dict[str, Any]]) -> bool:
+    """True when the user configured a specific auxiliary vision backend.
+
+    An explicit override means the user *wants* the text pipeline (they're
+    paying for a dedicated vision model), so we don't silently bypass it.
+    """
+    if not isinstance(cfg, dict):
+        return False
+    aux = cfg.get("auxiliary") or {}
+    if not isinstance(aux, dict):
+        return False
+    vision = aux.get("vision") or {}
+    if not isinstance(vision, dict):
+        return False
+
+    provider = str(vision.get("provider") or "").strip().lower()
+    model = str(vision.get("model") or "").strip()
+    base_url = str(vision.get("base_url") or "").strip()
+
+    # "auto" / "" / blank = not explicit
+    if provider in ("", "auto") and not model and not base_url:
+        return False
+    return True
+
+
+def _lookup_supports_vision(provider: str, model: str) -> Optional[bool]:
+    """Return True/False if we can resolve caps, None if unknown."""
+    if not provider or not model:
+        return None
+    try:
+        from agent.models_dev import get_model_capabilities
+        caps = get_model_capabilities(provider, model)
+    except Exception as exc:  # pragma: no cover - defensive
+        logger.debug("image_routing: caps lookup failed for %s:%s — %s", provider, model, exc)
+        return None
+    if caps is None:
+        return None
+    return bool(caps.supports_vision)
+
+
+def decide_image_input_mode(
+    provider: str,
+    model: str,
+    cfg: Optional[Dict[str, Any]],
+) -> str:
+    """Return ``"native"`` or ``"text"`` for the given turn.
+
+    Args:
+      provider: active inference provider ID (e.g. ``"anthropic"``, ``"openrouter"``).
+      model:    active model slug as it would be sent to the provider.
+      cfg:      loaded config.yaml dict, or None. When None, behaves as auto.
+    """
+    mode_cfg = "auto"
+    if isinstance(cfg, dict):
+        agent_cfg = cfg.get("agent") or {}
+        if isinstance(agent_cfg, dict):
+            mode_cfg = _coerce_mode(agent_cfg.get("image_input_mode"))
+
+    if mode_cfg == "native":
+        return "native"
+    if mode_cfg == "text":
+        return "text"
+
+    # auto
+    if _explicit_aux_vision_override(cfg):
+        return "text"
+
+    supports = _lookup_supports_vision(provider, model)
+    if supports is True:
+        return "native"
+    return "text"
+
+
+# Image size handling is REACTIVE rather than proactive: we attempt native
+# attachment at full size regardless of provider, and rely on
+# ``run_agent._try_shrink_image_parts_in_messages`` to shrink + retry if
+# the provider rejects the request (e.g. Anthropic's hard 5 MB per-image
+# ceiling returned as HTTP 400 "image exceeds 5 MB maximum").
+#
+# Why reactive: our knowledge of provider ceilings is partial and evolving
+# (OpenAI accepts 49 MB+, Anthropic 5 MB, Gemini 100 MB, others unknown).
+# A proactive per-provider table would be stale the moment a provider raises
+# or lowers its limit, and silently degrading quality for users on providers
+# that would have accepted the full image is the worse failure mode.
+# The shrink-on-reject path loses 1 API call + maybe 1s of Pillow work when
+# it fires, which is cheaper than permanent quality loss.
+
+
+def _guess_mime(path: Path) -> str:
+    mime, _ = mimetypes.guess_type(str(path))
+    if mime and mime.startswith("image/"):
+        return mime
+    # mimetypes on some Linux distros mis-maps .jpg; default to jpeg when
+    # the suffix looks imagey.
+    suffix = path.suffix.lower()
+    return {
+        ".jpg": "image/jpeg",
+        ".jpeg": "image/jpeg",
+        ".png": "image/png",
+        ".gif": "image/gif",
+        ".webp": "image/webp",
+        ".bmp": "image/bmp",
+    }.get(suffix, "image/jpeg")
+
+
+def _file_to_data_url(path: Path) -> Optional[str]:
+    """Encode a local image as a base64 data URL at its native size.
+
+    Size limits are NOT enforced here — the agent retry loop
+    (``run_agent._try_shrink_image_parts_in_messages``) shrinks on the
+    provider's first rejection. Keeping this simple means providers that
+    accept large images (OpenAI 49 MB+, Gemini 100 MB) don't pay a silent
+    quality tax just because one other provider is stricter.
+
+    Returns None only if the file can't be read (missing, permission
+    denied, etc.); the caller reports those paths in ``skipped``.
+    """
+    try:
+        raw = path.read_bytes()
+    except Exception as exc:
+        logger.warning("image_routing: failed to read %s — %s", path, exc)
+        return None
+    mime = _guess_mime(path)
+    b64 = base64.b64encode(raw).decode("ascii")
+    return f"data:{mime};base64,{b64}"
+
+
+def build_native_content_parts(
+    user_text: str,
+    image_paths: List[str],
+) -> Tuple[List[Dict[str, Any]], List[str]]:
+    """Build an OpenAI-style ``content`` list for a user turn.
+
+    Shape:
+      [{"type": "text", "text": "..."},
+       {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}},
+       ...]
+
+    Images are attached at their native size. If a provider rejects the
+    request because an image is too large (e.g. Anthropic's 5 MB per-image
+    ceiling), the agent's retry loop transparently shrinks and retries
+    once — see ``run_agent._try_shrink_image_parts_in_messages``.
+
+    Returns (content_parts, skipped_paths). Skipped paths are files that
+    couldn't be read from disk.
+    """
+    parts: List[Dict[str, Any]] = []
+    skipped: List[str] = []
+
+    text = (user_text or "").strip()
+    if text:
+        parts.append({"type": "text", "text": text})
+
+    for raw_path in image_paths:
+        p = Path(raw_path)
+        if not p.exists() or not p.is_file():
+            skipped.append(str(raw_path))
+            continue
+        data_url = _file_to_data_url(p)
+        if not data_url:
+            skipped.append(str(raw_path))
+            continue
+        parts.append({
+            "type": "image_url",
+            "image_url": {"url": data_url},
+        })
+
+    # If the text was empty, add a neutral prompt so the turn isn't just images.
+    if not text and any(p.get("type") == "image_url" for p in parts):
+        parts.insert(0, {"type": "text", "text": "What do you see in this image?"})
+
+    return parts, skipped
+
+
+__all__ = [
+    "decide_image_input_mode",
+    "build_native_content_parts",
+]

agent/lmstudio_reasoning.py

@@ -0,0 +1,48 @@
+"""LM Studio reasoning-effort resolution shared by the chat-completions
+transport and run_agent's iteration-limit summary path.
+
+LM Studio publishes per-model ``capabilities.reasoning.allowed_options`` (e.g.
+``["off","on"]`` for toggle-style models, ``["off","minimal","low"]`` for
+graduated models). We map the user's ``reasoning_config`` onto LM Studio's
+OpenAI-compatible vocabulary, then clamp against the model's allowed set so
+the server doesn't 400 on an unsupported effort.
+"""
+
+from __future__ import annotations
+
+from typing import List, Optional
+
+# LM Studio accepts these top-level reasoning_effort values via its
+# OpenAI-compatible chat.completions endpoint.
+_LM_VALID_EFFORTS = {"none", "minimal", "low", "medium", "high", "xhigh"}
+
+# Toggle-style models publish allowed_options as ["off","on"] in /api/v1/models.
+# Map them onto the OpenAI-compatible request vocabulary.
+_LM_EFFORT_ALIASES = {"off": "none", "on": "medium"}
+
+
+def resolve_lmstudio_effort(
+    reasoning_config: Optional[dict],
+    allowed_options: Optional[List[str]],
+) -> Optional[str]:
+    """Return the ``reasoning_effort`` string to send to LM Studio, or ``None``.
+
+    ``None`` means "omit the field": the user picked a level the model can't
+    honor, so let LM Studio fall back to the model's declared default rather
+    than silently substituting a different effort. When ``allowed_options`` is
+    falsy (probe failed), skip clamping and send the resolved effort anyway.
+    """
+    effort = "medium"
+    if reasoning_config and isinstance(reasoning_config, dict):
+        if reasoning_config.get("enabled") is False:
+            effort = "none"
+        else:
+            raw = (reasoning_config.get("effort") or "").strip().lower()
+            raw = _LM_EFFORT_ALIASES.get(raw, raw)
+            if raw in _LM_VALID_EFFORTS:
+                effort = raw
+    if allowed_options:
+        allowed = {_LM_EFFORT_ALIASES.get(opt, opt) for opt in allowed_options}
+        if effort not in allowed:
+            return None
+    return effort

agent/manual_compression_feedback.py

@@ -20,25 +20,25 @@ def summarize_manual_compression(
         headline = f"No changes from compression: {before_count} messages"
         if after_tokens == before_tokens:
             token_line = (
-                f"Rough transcript estimate: ~{before_tokens:,} tokens (unchanged)"
+                f"Approx request size: ~{before_tokens:,} tokens (unchanged)"
             )
         else:
             token_line = (
-                f"Rough transcript estimate: ~{before_tokens:,} → "
+                f"Approx request size: ~{before_tokens:,} → "
                 f"~{after_tokens:,} tokens"
             )
     else:
         headline = f"Compressed: {before_count} → {after_count} messages"
         token_line = (
-            f"Rough transcript estimate: ~{before_tokens:,} → "
+            f"Approx request size: ~{before_tokens:,} → "
             f"~{after_tokens:,} tokens"
         )
 
     note = None
     if not noop and after_count < before_count and after_tokens > before_tokens:
         note = (
-            "Note: fewer messages can still raise this rough transcript estimate "
-            "when compression rewrites the transcript into denser summaries."
+            "Note: fewer messages can still raise this estimate when "
+            "compression rewrites the transcript into denser summaries."
         )
 
     return {

agent/memory_manager.py

@@ -28,9 +28,9 @@ Usage in run_agent.py:
 
 from __future__ import annotations
 
-import json
 import logging
 import re
+import inspect
 from typing import Any, Dict, List, Optional
 
 from agent.memory_provider import MemoryProvider
@@ -62,15 +62,124 @@ def sanitize_context(text: str) -> str:
     return text
 
 
-def build_memory_context_block(raw_context: str) -> str:
-    """Wrap prefetched memory in a fenced block with system note.
+class StreamingContextScrubber:
+    """Stateful scrubber for streaming text that may contain split memory-context spans.
 
-    The fence prevents the model from treating recalled context as user
-    discourse.  Injected at API-call time only — never persisted.
+    The one-shot ``sanitize_context`` regex cannot survive chunk boundaries:
+    a ``<memory-context>`` opened in one delta and closed in a later delta
+    leaks its payload to the UI because the non-greedy block regex needs
+    both tags in one string.  This scrubber runs a small state machine
+    across deltas, holding back partial-tag tails and discarding
+    everything inside a span (including the system-note line).
+
+    Usage::
+
+        scrubber = StreamingContextScrubber()
+        for delta in stream:
+            visible = scrubber.feed(delta)
+            if visible:
+                emit(visible)
+        trailing = scrubber.flush()  # at end of stream
+        if trailing:
+            emit(trailing)
+
+    The scrubber is re-entrant per agent instance.  Callers building new
+    top-level responses (new turn) should create a fresh scrubber or call
+    ``reset()``.
     """
+
+    _OPEN_TAG = "<memory-context>"
+    _CLOSE_TAG = "</memory-context>"
+
+    def __init__(self) -> None:
+        self._in_span: bool = False
+        self._buf: str = ""
+
+    def reset(self) -> None:
+        self._in_span = False
+        self._buf = ""
+
+    def feed(self, text: str) -> str:
+        """Return the visible portion of ``text`` after scrubbing.
+
+        Any trailing fragment that could be the start of an open/close tag
+        is held back in the internal buffer and surfaced on the next
+        ``feed()`` call or discarded/emitted by ``flush()``.
+        """
+        if not text:
+            return ""
+        buf = self._buf + text
+        self._buf = ""
+        out: list[str] = []
+
+        while buf:
+            if self._in_span:
+                idx = buf.lower().find(self._CLOSE_TAG)
+                if idx == -1:
+                    # Hold back a potential partial close tag; drop the rest
+                    held = self._max_partial_suffix(buf, self._CLOSE_TAG)
+                    self._buf = buf[-held:] if held else ""
+                    return "".join(out)
+                # Found close — skip span content + tag, continue
+                buf = buf[idx + len(self._CLOSE_TAG):]
+                self._in_span = False
+            else:
+                idx = buf.lower().find(self._OPEN_TAG)
+                if idx == -1:
+                    # No open tag — hold back a potential partial open tag
+                    held = self._max_partial_suffix(buf, self._OPEN_TAG)
+                    if held:
+                        out.append(buf[:-held])
+                        self._buf = buf[-held:]
+                    else:
+                        out.append(buf)
+                    return "".join(out)
+                # Emit text before the tag, enter span
+                if idx > 0:
+                    out.append(buf[:idx])
+                buf = buf[idx + len(self._OPEN_TAG):]
+                self._in_span = True
+
+        return "".join(out)
+
+    def flush(self) -> str:
+        """Emit any held-back buffer at end-of-stream.
+
+        If we're still inside an unterminated span the remaining content is
+        discarded (safer: leaking partial memory context is worse than a
+        truncated answer).  Otherwise the held-back partial-tag tail is
+        emitted verbatim (it turned out not to be a real tag).
+        """
+        if self._in_span:
+            self._buf = ""
+            self._in_span = False
+            return ""
+        tail = self._buf
+        self._buf = ""
+        return tail
+
+    @staticmethod
+    def _max_partial_suffix(buf: str, tag: str) -> int:
+        """Return the length of the longest buf-suffix that is a tag-prefix.
+
+        Case-insensitive.  Returns 0 if no suffix could start the tag.
+        """
+        tag_lower = tag.lower()
+        buf_lower = buf.lower()
+        max_check = min(len(buf_lower), len(tag_lower) - 1)
+        for i in range(max_check, 0, -1):
+            if tag_lower.startswith(buf_lower[-i:]):
+                return i
+        return 0
+
+
+def build_memory_context_block(raw_context: str) -> str:
+    """Wrap prefetched memory in a fenced block with system note."""
     if not raw_context or not raw_context.strip():
         return ""
     clean = sanitize_context(raw_context)
+    if clean != raw_context:
+        logger.warning("memory provider returned pre-wrapped context; stripped")
     return (
         "<memory-context>\n"
         "[System note: The following is recalled memory context, "
@@ -293,6 +402,41 @@ class MemoryManager:
                     provider.name, e,
                 )
 
+    def on_session_switch(
+        self,
+        new_session_id: str,
+        *,
+        parent_session_id: str = "",
+        reset: bool = False,
+        **kwargs,
+    ) -> None:
+        """Notify all providers that the agent's session_id has rotated.
+
+        Fires on ``/resume``, ``/branch``, ``/reset``, ``/new``, and
+        context compression — any path that reassigns
+        ``AIAgent.session_id`` without tearing the provider down.
+
+        Providers keep running; they only need to refresh cached
+        per-session state so subsequent writes land in the correct
+        session's record. See ``MemoryProvider.on_session_switch`` for
+        the full contract.
+        """
+        if not new_session_id:
+            return
+        for provider in self._providers:
+            try:
+                provider.on_session_switch(
+                    new_session_id,
+                    parent_session_id=parent_session_id,
+                    reset=reset,
+                    **kwargs,
+                )
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_session_switch failed: %s",
+                    provider.name, e,
+                )
+
     def on_pre_compress(self, messages: List[Dict[str, Any]]) -> str:
         """Notify all providers before context compression.
 
@@ -312,7 +456,39 @@ class MemoryManager:
                 )
         return "\n\n".join(parts)
 
-    def on_memory_write(self, action: str, target: str, content: str) -> None:
+    @staticmethod
+    def _provider_memory_write_metadata_mode(provider: MemoryProvider) -> str:
+        """Return how to pass metadata to a provider's memory-write hook."""
+        try:
+            signature = inspect.signature(provider.on_memory_write)
+        except (TypeError, ValueError):
+            return "keyword"
+
+        params = list(signature.parameters.values())
+        if any(p.kind == inspect.Parameter.VAR_KEYWORD for p in params):
+            return "keyword"
+        if "metadata" in signature.parameters:
+            return "keyword"
+
+        accepted = [
+            p for p in params
+            if p.kind in (
+                inspect.Parameter.POSITIONAL_ONLY,
+                inspect.Parameter.POSITIONAL_OR_KEYWORD,
+                inspect.Parameter.KEYWORD_ONLY,
+            )
+        ]
+        if len(accepted) >= 4:
+            return "positional"
+        return "legacy"
+
+    def on_memory_write(
+        self,
+        action: str,
+        target: str,
+        content: str,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
         """Notify external providers when the built-in memory tool writes.
 
         Skips the builtin provider itself (it's the source of the write).
@@ -321,7 +497,15 @@ class MemoryManager:
             if provider.name == "builtin":
                 continue
             try:
-                provider.on_memory_write(action, target, content)
+                metadata_mode = self._provider_memory_write_metadata_mode(provider)
+                if metadata_mode == "keyword":
+                    provider.on_memory_write(
+                        action, target, content, metadata=dict(metadata or {})
+                    )
+                elif metadata_mode == "positional":
+                    provider.on_memory_write(action, target, content, dict(metadata or {}))
+                else:
+                    provider.on_memory_write(action, target, content)
             except Exception as e:
                 logger.debug(
                     "Memory provider '%s' on_memory_write failed: %s",

agent/memory_provider.py

@@ -25,8 +25,9 @@ Lifecycle (called by MemoryManager, wired in run_agent.py):
 Optional hooks (override to opt in):
   on_turn_start(turn, message, **kwargs) — per-turn tick with runtime context
   on_session_end(messages)               — end-of-session extraction
+  on_session_switch(new_session_id, **kwargs) — mid-process session_id rotation
   on_pre_compress(messages) -> str       — extract before context compression
-  on_memory_write(action, target, content) — mirror built-in memory writes
+  on_memory_write(action, target, content, metadata=None) — mirror built-in memory writes
   on_delegation(task, result, **kwargs)  — parent-side observation of subagent work
 """
 
@@ -34,7 +35,7 @@ from __future__ import annotations
 
 import logging
 from abc import ABC, abstractmethod
-from typing import Any, Dict, List
+from typing import Any, Dict, List, Optional
 
 logger = logging.getLogger(__name__)
 
@@ -160,6 +161,45 @@ class MemoryProvider(ABC):
         (CLI exit, /reset, gateway session expiry).
         """
 
+    def on_session_switch(
+        self,
+        new_session_id: str,
+        *,
+        parent_session_id: str = "",
+        reset: bool = False,
+        **kwargs,
+    ) -> None:
+        """Called when the agent switches session_id mid-process.
+
+        Fires on ``/resume``, ``/branch``, ``/reset``, ``/new`` (CLI), the
+        gateway equivalents, and context compression — any path that
+        reassigns ``AIAgent.session_id`` without tearing the provider down.
+
+        Providers that cache per-session state in ``initialize()``
+        (``_session_id``, ``_document_id``, accumulated turn buffers,
+        counters) should update or reset that state here so subsequent
+        writes land in the correct session's record.
+
+        Parameters
+        ----------
+        new_session_id:
+            The session_id the agent just switched to.
+        parent_session_id:
+            The previous session_id, if meaningful — set for ``/branch``
+            (fork lineage), context compression (continuation lineage),
+            and ``/resume`` (the session we're leaving). Empty string
+            when no lineage applies.
+        reset:
+            ``True`` when this is a genuinely new conversation, not a
+            resumption of an existing one. Fired by ``/reset`` / ``/new``.
+            Providers should flush accumulated per-session buffers
+            (``_session_turns``, ``_turn_counter``, etc.) when this is
+            set. ``False`` for ``/resume`` / ``/branch`` / compression
+            where the logical conversation continues under the new id.
+
+        Default is no-op for backward compatibility.
+        """
+
     def on_pre_compress(self, messages: List[Dict[str, Any]]) -> str:
         """Called before context compression discards old messages.
 
@@ -220,12 +260,21 @@ class MemoryProvider(ABC):
           should all have ``env_var`` set and this method stays no-op).
         """
 
-    def on_memory_write(self, action: str, target: str, content: str) -> None:
+    def on_memory_write(
+        self,
+        action: str,
+        target: str,
+        content: str,
+        metadata: Optional[Dict[str, Any]] = None,
+    ) -> None:
         """Called when the built-in memory tool writes an entry.
 
         action: 'add', 'replace', or 'remove'
         target: 'memory' or 'user'
         content: the entry content
+        metadata: structured provenance for the write, when available. Common
+          keys include ``write_origin``, ``execution_context``, ``session_id``,
+          ``parent_session_id``, ``platform``, and ``tool_name``.
 
         Use to mirror built-in memory writes to your backend.
         """

agent/model_metadata.py

@@ -6,6 +6,7 @@ and run_agent.py for pre-flight context checks.
 
 import ipaddress
 import logging
+import os
 import re
 import time
 from pathlib import Path
@@ -21,16 +22,37 @@ from hermes_constants import OPENROUTER_MODELS_URL
 
 logger = logging.getLogger(__name__)
 
+
+def _resolve_requests_verify() -> bool | str:
+    """Resolve SSL verify setting for `requests` calls from env vars.
+
+    The `requests` library only honours REQUESTS_CA_BUNDLE / CURL_CA_BUNDLE
+    by default. Hermes also honours HERMES_CA_BUNDLE (its own convention)
+    and SSL_CERT_FILE (used by the stdlib `ssl` module and by httpx), so
+    that a single env var can cover both `requests` and `httpx` callsites
+    inside the same process.
+
+    Returns either a filesystem path to a CA bundle, or True to defer to
+    the requests default (certifi).
+    """
+    for env_var in ("HERMES_CA_BUNDLE", "REQUESTS_CA_BUNDLE", "SSL_CERT_FILE"):
+        val = os.getenv(env_var)
+        if val and os.path.isfile(val):
+            return val
+    return True
+
 # Provider names that can appear as a "provider:" prefix before a model ID.
 # Only these are stripped — Ollama-style "model:tag" colons (e.g. "qwen3.5:27b")
 # are preserved so the full model name reaches cache lookups and server queries.
 _PROVIDER_PREFIXES: frozenset[str] = frozenset({
     "openrouter", "nous", "openai-codex", "copilot", "copilot-acp",
-    "gemini", "ollama-cloud", "zai", "kimi-coding", "kimi-coding-cn", "stepfun", "minimax", "minimax-cn", "anthropic", "deepseek",
+    "gemini", "ollama-cloud", "zai", "kimi-coding", "kimi-coding-cn", "stepfun", "minimax", "minimax-oauth", "minimax-cn", "anthropic", "deepseek",
     "opencode-zen", "opencode-go", "ai-gateway", "kilocode", "alibaba",
     "qwen-oauth",
     "xiaomi",
     "arcee",
+    "gmi",
+    "tencent-tokenhub",
     "custom", "local",
     # Common aliases
     "google", "google-gemini", "google-ai-studio",
@@ -39,7 +61,9 @@ _PROVIDER_PREFIXES: frozenset[str] = frozenset({
     "ollama",
     "stepfun", "opencode", "zen", "go", "vercel", "kilo", "dashscope", "aliyun", "qwen",
     "mimo", "xiaomi-mimo",
+    "tencent", "tokenhub", "tencent-cloud", "tencentmaas",
     "arcee-ai", "arceeai",
+    "gmi-cloud", "gmicloud",
     "xai", "x-ai", "x.ai", "grok",
     "nvidia", "nim", "nvidia-nim", "nemotron",
     "qwen-portal",
@@ -86,9 +110,11 @@ _endpoint_model_metadata_cache_time: Dict[str, float] = {}
 _ENDPOINT_MODEL_CACHE_TTL = 300
 
 # Descending tiers for context length probing when the model is unknown.
-# We start at 128K (a safe default for most modern models) and step down
-# on context-length errors until one works.
+# We start at 256K (covers GPT-5.x, many current large-context models) and
+# step down on context-length errors until one works.  Tier[0] is also the
+# default fallback when no detection method succeeds.
 CONTEXT_PROBE_TIERS = [
+    256_000,
     128_000,
     64_000,
     32_000,
@@ -123,6 +149,11 @@ DEFAULT_CONTEXT_LENGTHS = {
     "claude": 200000,
     # OpenAI — GPT-5 family (most have 400k; specific overrides first)
     # Source: https://developers.openai.com/api/docs/models
+    # GPT-5.5 (launched Apr 23 2026) is 1.05M on the direct OpenAI API and
+    # ChatGPT Codex OAuth caps it at 272K; both paths resolve via their own
+    # provider-aware branches (_resolve_codex_oauth_context_length + models.dev).
+    # This hardcoded value is only reached when every probe misses.
+    "gpt-5.5": 1050000,
     "gpt-5.4-nano": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4-mini": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4": 1050000,               # GPT-5.4, GPT-5.4 Pro (1.05M context)
@@ -138,7 +169,17 @@ DEFAULT_CONTEXT_LENGTHS = {
     "gemma-4-31b": 256000,
     "gemma-3": 131072,
     "gemma": 8192,  # fallback for older gemma models
-    # DeepSeek
+    # DeepSeek — V4 family ships with a 1M context window. The legacy
+    # aliases ``deepseek-chat`` / ``deepseek-reasoner`` are server-side
+    # mapped to the non-thinking / thinking modes of ``deepseek-v4-flash``
+    # and inherit the same 1M window. The ``deepseek`` substring entry
+    # below remains as a 128K fallback for older / unknown DeepSeek model
+    # ids (e.g. via custom endpoints).
+    # https://api-docs.deepseek.com/zh-cn/quick_start/pricing
+    "deepseek-v4-pro": 1_000_000,
+    "deepseek-v4-flash": 1_000_000,
+    "deepseek-chat": 1_000_000,
+    "deepseek-reasoner": 1_000_000,
     "deepseek": 128000,
     # Meta
     "llama": 131072,
@@ -169,6 +210,8 @@ DEFAULT_CONTEXT_LENGTHS = {
     "grok": 131072,             # catch-all (grok-beta, unknown grok-*)
     # Kimi
     "kimi": 262144,
+    # Tencent — Hy3 Preview (Hunyuan) with 256K context window
+    "hy3-preview": 256000,
     # Nemotron — NVIDIA's open-weights series (128K context across all sizes)
     "nemotron": 131072,
     # Arcee
@@ -183,12 +226,12 @@ DEFAULT_CONTEXT_LENGTHS = {
     "moonshotai/Kimi-K2.6": 262144,
     "moonshotai/Kimi-K2-Thinking": 262144,
     "MiniMaxAI/MiniMax-M2.5": 204800,
-    "XiaomiMiMo/MiMo-V2-Flash": 256000,
-    "mimo-v2-pro": 1000000,
-    "mimo-v2-omni": 256000,
-    "mimo-v2-flash": 256000,
-    "mimo-v2.5-pro": 1000000,
-    "mimo-v2.5": 1000000,
+    "XiaomiMiMo/MiMo-V2-Flash": 262144,
+    "mimo-v2-pro": 1048576,
+    "mimo-v2.5-pro": 1048576,
+    "mimo-v2.5": 1048576,
+    "mimo-v2-omni": 262144,
+    "mimo-v2-flash": 262144,
     "zai-org/GLM-5": 202752,
 }
 
@@ -270,6 +313,8 @@ _URL_TO_PROVIDER: Dict[str, str] = {
     "integrate.api.nvidia.com": "nvidia",
     "api.xiaomimimo.com": "xiaomi",
     "xiaomimimo.com": "xiaomi",
+    "api.gmi-serving.com": "gmi",
+    "tokenhub.tencentmaas.com": "tencent-tokenhub",
     "ollama.com": "ollama-cloud",
 }
 
@@ -491,7 +536,7 @@ def fetch_model_metadata(force_refresh: bool = False) -> Dict[str, Dict[str, Any
         return _model_metadata_cache
 
     try:
-        response = requests.get(OPENROUTER_MODELS_URL, timeout=10)
+        response = requests.get(OPENROUTER_MODELS_URL, timeout=10, verify=_resolve_requests_verify())
         response.raise_for_status()
         data = response.json()
 
@@ -558,6 +603,7 @@ def fetch_endpoint_model_metadata(
                     server_url.rstrip("/") + "/api/v1/models",
                     headers=headers,
                     timeout=10,
+                    verify=_resolve_requests_verify(),
                 )
                 response.raise_for_status()
                 payload = response.json()
@@ -579,8 +625,6 @@ def fetch_endpoint_model_metadata(
                         if isinstance(ctx, int) and ctx > 0:
                             context_length = ctx
                             break
-                    if context_length is None:
-                        context_length = _extract_context_length(model)
                     if context_length is not None:
                         entry["context_length"] = context_length
 
@@ -606,7 +650,7 @@ def fetch_endpoint_model_metadata(
     for candidate in candidates:
         url = candidate.rstrip("/") + "/models"
         try:
-            response = requests.get(url, headers=headers, timeout=10)
+            response = requests.get(url, headers=headers, timeout=10, verify=_resolve_requests_verify())
             response.raise_for_status()
             payload = response.json()
             cache: Dict[str, Dict[str, Any]] = {}
@@ -637,9 +681,10 @@ def fetch_endpoint_model_metadata(
                 try:
                     # Try /v1/props first (current llama.cpp); fall back to /props for older builds
                     base = candidate.rstrip("/").replace("/v1", "")
-                    props_resp = requests.get(base + "/v1/props", headers=headers, timeout=5)
+                    _verify = _resolve_requests_verify()
+                    props_resp = requests.get(base + "/v1/props", headers=headers, timeout=5, verify=_verify)
                     if not props_resp.ok:
-                        props_resp = requests.get(base + "/props", headers=headers, timeout=5)
+                        props_resp = requests.get(base + "/props", headers=headers, timeout=5, verify=_verify)
                     if props_resp.ok:
                         props = props_resp.json()
                         gen_settings = props.get("default_generation_settings", {})
@@ -663,6 +708,29 @@ def fetch_endpoint_model_metadata(
     return {}
 
 
+def _resolve_endpoint_context_length(
+    model: str,
+    base_url: str,
+    api_key: str = "",
+) -> Optional[int]:
+    """Resolve context length from an endpoint's live ``/models`` metadata."""
+    endpoint_metadata = fetch_endpoint_model_metadata(base_url, api_key=api_key)
+    matched = endpoint_metadata.get(model)
+    if not matched:
+        if len(endpoint_metadata) == 1:
+            matched = next(iter(endpoint_metadata.values()))
+        else:
+            for key, entry in endpoint_metadata.items():
+                if model in key or key in model:
+                    matched = entry
+                    break
+    if matched:
+        context_length = matched.get("context_length")
+        if isinstance(context_length, int):
+            return context_length
+    return None
+
+
 def _get_context_cache_path() -> Path:
     """Return path to the persistent context length cache file."""
     from hermes_constants import get_hermes_home
@@ -711,6 +779,22 @@ def get_cached_context_length(model: str, base_url: str) -> Optional[int]:
     return cache.get(key)
 
 
+def _invalidate_cached_context_length(model: str, base_url: str) -> None:
+    """Drop a stale cache entry so it gets re-resolved on the next lookup."""
+    key = f"{model}@{base_url}"
+    cache = _load_context_cache()
+    if key not in cache:
+        return
+    del cache[key]
+    path = _get_context_cache_path()
+    try:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with open(path, "w") as f:
+            yaml.dump({"context_lengths": cache}, f, default_flow_style=False)
+    except Exception as e:
+        logger.debug("Failed to invalidate context length cache entry %s: %s", key, e)
+
+
 def get_next_probe_tier(current_length: int) -> Optional[int]:
     """Return the next lower probe tier, or None if already at minimum."""
     for tier in CONTEXT_PROBE_TIERS:
@@ -930,10 +1014,7 @@ def _query_local_context_length(model: str, base_url: str, api_key: str = "") ->
                                 ctx = cfg.get("context_length")
                                 if ctx and isinstance(ctx, (int, float)):
                                     return int(ctx)
-                            # Fall back to max_context_length (theoretical model max)
-                            ctx = m.get("max_context_length") or m.get("context_length")
-                            if ctx and isinstance(ctx, (int, float)):
-                                return int(ctx)
+                            break
 
             # LM Studio / vLLM / llama.cpp: try /v1/models/{model}
             resp = client.get(f"{server_url}/v1/models/{model}")
@@ -988,7 +1069,7 @@ def _query_anthropic_context_length(model: str, base_url: str, api_key: str) ->
             "x-api-key": api_key,
             "anthropic-version": "2023-06-01",
         }
-        resp = requests.get(url, headers=headers, timeout=10)
+        resp = requests.get(url, headers=headers, timeout=10, verify=_resolve_requests_verify())
         if resp.status_code != 200:
             return None
         data = resp.json()
@@ -1002,6 +1083,116 @@ def _query_anthropic_context_length(model: str, base_url: str, api_key: str) ->
     return None
 
 
+# Known ChatGPT Codex OAuth context windows (observed via live
+# chatgpt.com/backend-api/codex/models probe, Apr 2026). These are the
+# `context_window` values, which are what Codex actually enforces — the
+# direct OpenAI API has larger limits for the same slugs, but Codex OAuth
+# caps lower (e.g. gpt-5.5 is 1.05M on the API, 272K on Codex).
+#
+# Used as a fallback when the live probe fails (no token, network error).
+# Longest keys first so substring match picks the most specific entry.
+_CODEX_OAUTH_CONTEXT_FALLBACK: Dict[str, int] = {
+    "gpt-5.1-codex-max": 272_000,
+    "gpt-5.1-codex-mini": 272_000,
+    "gpt-5.3-codex": 272_000,
+    "gpt-5.2-codex": 272_000,
+    "gpt-5.4-mini": 272_000,
+    "gpt-5.5": 272_000,
+    "gpt-5.4": 272_000,
+    "gpt-5.2": 272_000,
+    "gpt-5": 272_000,
+}
+
+
+_codex_oauth_context_cache: Dict[str, int] = {}
+_codex_oauth_context_cache_time: float = 0.0
+_CODEX_OAUTH_CONTEXT_CACHE_TTL = 3600  # 1 hour
+
+
+def _fetch_codex_oauth_context_lengths(access_token: str) -> Dict[str, int]:
+    """Probe the ChatGPT Codex /models endpoint for per-slug context windows.
+
+    Codex OAuth imposes its own context limits that differ from the direct
+    OpenAI API (e.g. gpt-5.5 is 1.05M on the API, 272K on Codex). The
+    `context_window` field in each model entry is the authoritative source.
+
+    Returns a ``{slug: context_window}`` dict. Empty on failure.
+    """
+    global _codex_oauth_context_cache, _codex_oauth_context_cache_time
+    now = time.time()
+    if (
+        _codex_oauth_context_cache
+        and now - _codex_oauth_context_cache_time < _CODEX_OAUTH_CONTEXT_CACHE_TTL
+    ):
+        return _codex_oauth_context_cache
+
+    try:
+        resp = requests.get(
+            "https://chatgpt.com/backend-api/codex/models?client_version=1.0.0",
+            headers={"Authorization": f"Bearer {access_token}"},
+            timeout=10,
+            verify=_resolve_requests_verify(),
+        )
+        if resp.status_code != 200:
+            logger.debug(
+                "Codex /models probe returned HTTP %s; falling back to hardcoded defaults",
+                resp.status_code,
+            )
+            return {}
+        data = resp.json()
+    except Exception as exc:
+        logger.debug("Codex /models probe failed: %s", exc)
+        return {}
+
+    entries = data.get("models", []) if isinstance(data, dict) else []
+    result: Dict[str, int] = {}
+    for item in entries:
+        if not isinstance(item, dict):
+            continue
+        slug = item.get("slug")
+        ctx = item.get("context_window")
+        if isinstance(slug, str) and isinstance(ctx, int) and ctx > 0:
+            result[slug.strip()] = ctx
+
+    if result:
+        _codex_oauth_context_cache = result
+        _codex_oauth_context_cache_time = now
+    return result
+
+
+def _resolve_codex_oauth_context_length(
+    model: str, access_token: str = ""
+) -> Optional[int]:
+    """Resolve a Codex OAuth model's real context window.
+
+    Prefers a live probe of chatgpt.com/backend-api/codex/models (when we
+    have a bearer token), then falls back to ``_CODEX_OAUTH_CONTEXT_FALLBACK``.
+    """
+    model_bare = _strip_provider_prefix(model).strip()
+    if not model_bare:
+        return None
+
+    if access_token:
+        live = _fetch_codex_oauth_context_lengths(access_token)
+        if model_bare in live:
+            return live[model_bare]
+        # Case-insensitive match in case casing drifts
+        model_lower = model_bare.lower()
+        for slug, ctx in live.items():
+            if slug.lower() == model_lower:
+                return ctx
+
+    # Fallback: longest-key-first substring match over hardcoded defaults.
+    model_lower = model_bare.lower()
+    for slug, ctx in sorted(
+        _CODEX_OAUTH_CONTEXT_FALLBACK.items(), key=lambda x: len(x[0]), reverse=True
+    ):
+        if slug in model_lower:
+            return ctx
+
+    return None
+
+
 def _resolve_nous_context_length(model: str) -> Optional[int]:
     """Resolve Nous Portal model context length via OpenRouter metadata.
 
@@ -1041,12 +1232,14 @@ def get_model_context_length(
     api_key: str = "",
     config_context_length: int | None = None,
     provider: str = "",
+    custom_providers: list | None = None,
 ) -> int:
     """Get the context length for a model.
 
     Resolution order:
     0. Explicit config override (model.context_length or custom_providers per-model)
     1. Persistent cache (previously discovered via probing)
+    1b. AWS Bedrock static table (must precede custom-endpoint probe)
     2. Active endpoint metadata (/models for explicit custom endpoints)
     3. Local server query (for local endpoints)
     4. Anthropic /v1/models API (API-key users only, not OAuth)
@@ -1054,22 +1247,76 @@ def get_model_context_length(
     6. Nous suffix-match via OpenRouter cache
     7. models.dev registry lookup (provider-aware)
     8. Thin hardcoded defaults (broad family patterns)
-    9. Default fallback (128K)
+    9. Default fallback (256K)
     """
     # 0. Explicit config override — user knows best
     if config_context_length is not None and isinstance(config_context_length, int) and config_context_length > 0:
         return config_context_length
 
+    # 0b. custom_providers per-model override — check before any probe.
+    # This closes the gap where /model switch and display paths used to fall
+    # back to 128K despite the user having a per-model context_length set.
+    # See #15779.
+    if custom_providers and base_url and model:
+        try:
+            from hermes_cli.config import get_custom_provider_context_length
+            cp_ctx = get_custom_provider_context_length(
+                model=model,
+                base_url=base_url,
+                custom_providers=custom_providers,
+            )
+            if cp_ctx:
+                return cp_ctx
+        except Exception:
+            pass  # fall through to probing
+
     # Normalise provider-prefixed model names (e.g. "local:model-name" →
     # "model-name") so cache lookups and server queries use the bare ID that
     # local servers actually know about.  Ollama "model:tag" colons are preserved.
     model = _strip_provider_prefix(model)
 
     # 1. Check persistent cache (model+provider)
-    if base_url:
+    # LM Studio is excluded — its loaded context length is transient (the
+    # user can reload the model with a different context_length at any time
+    # via /api/v1/models/load), so a stale cached value would mask reloads.
+    if base_url and provider != "lmstudio":
         cached = get_cached_context_length(model, base_url)
         if cached is not None:
-            return cached
+            # Invalidate stale Codex OAuth cache entries: pre-PR #14935 builds
+            # resolved gpt-5.x to the direct-API value (e.g. 1.05M) via
+            # models.dev and persisted it. Codex OAuth caps at 272K for every
+            # slug, so any cached Codex entry at or above 400K is a leftover
+            # from the old resolution path. Drop it and fall through to the
+            # live /models probe in step 5 below.
+            if provider == "openai-codex" and cached >= 400_000:
+                logger.info(
+                    "Dropping stale Codex cache entry %s@%s -> %s (pre-fix value); "
+                    "re-resolving via live /models probe",
+                    model, base_url, f"{cached:,}",
+                )
+                _invalidate_cached_context_length(model, base_url)
+            else:
+                return cached
+
+    # 1b. AWS Bedrock — use static context length table.
+    # Bedrock's ListFoundationModels API doesn't expose context window sizes,
+    # so we maintain a curated table in bedrock_adapter.py that reflects
+    # AWS-imposed limits (e.g. 200K for Claude models vs 1M on the native
+    # Anthropic API).  This must run BEFORE the custom-endpoint probe at
+    # step 2 — bedrock-runtime.<region>.amazonaws.com is not in
+    # _URL_TO_PROVIDER, so it would otherwise be treated as a custom endpoint,
+    # fail the /models probe (Bedrock doesn't expose that shape), and fall
+    # back to the 128K default before reaching the original step 4b branch.
+    if provider == "bedrock" or (
+        base_url
+        and base_url_hostname(base_url).startswith("bedrock-runtime.")
+        and base_url_host_matches(base_url, "amazonaws.com")
+    ):
+        try:
+            from agent.bedrock_adapter import get_bedrock_context_length
+            return get_bedrock_context_length(model)
+        except ImportError:
+            pass  # boto3 not installed — fall through to generic resolution
 
     # 2. Active endpoint metadata for truly custom/unknown endpoints.
     # Known providers (Copilot, OpenAI, Anthropic, etc.) skip this — their
@@ -1077,28 +1324,16 @@ def get_model_context_length(
     # returns 128k) instead of the model's full context (400k).  models.dev
     # has the correct per-provider values and is checked at step 5+.
     if _is_custom_endpoint(base_url) and not _is_known_provider_base_url(base_url):
-        endpoint_metadata = fetch_endpoint_model_metadata(base_url, api_key=api_key)
-        matched = endpoint_metadata.get(model)
-        if not matched:
-            # Single-model servers: if only one model is loaded, use it
-            if len(endpoint_metadata) == 1:
-                matched = next(iter(endpoint_metadata.values()))
-            else:
-                # Fuzzy match: substring in either direction
-                for key, entry in endpoint_metadata.items():
-                    if model in key or key in model:
-                        matched = entry
-                        break
-        if matched:
-            context_length = matched.get("context_length")
-            if isinstance(context_length, int):
-                return context_length
+        context_length = _resolve_endpoint_context_length(model, base_url, api_key=api_key)
+        if context_length is not None:
+            return context_length
         if not _is_known_provider_base_url(base_url):
             # 3. Try querying local server directly
             if is_local_endpoint(base_url):
                 local_ctx = _query_local_context_length(model, base_url, api_key=api_key)
                 if local_ctx and local_ctx > 0:
-                    save_context_length(model, base_url, local_ctx)
+                    if provider != "lmstudio":
+                        save_context_length(model, base_url, local_ctx)
                     return local_ctx
             logger.info(
                 "Could not detect context length for model %r at %s — "
@@ -1116,19 +1351,7 @@ def get_model_context_length(
         if ctx:
             return ctx
 
-    # 4b. AWS Bedrock — use static context length table.
-    # Bedrock's ListFoundationModels doesn't expose context window sizes,
-    # so we maintain a curated table in bedrock_adapter.py.
-    if provider == "bedrock" or (
-        base_url
-        and base_url_hostname(base_url).startswith("bedrock-runtime.")
-        and base_url_host_matches(base_url, "amazonaws.com")
-    ):
-        try:
-            from agent.bedrock_adapter import get_bedrock_context_length
-            return get_bedrock_context_length(model)
-        except ImportError:
-            pass  # boto3 not installed — fall through to generic resolution
+    # 4b. (Bedrock handled earlier at step 1b — before custom-endpoint probe.)
 
     # 5. Provider-aware lookups (before generic OpenRouter cache)
     # These are provider-specific and take priority over the generic OR cache,
@@ -1142,10 +1365,38 @@ def get_model_context_length(
             if inferred:
                 effective_provider = inferred
 
+    # 5a. Copilot live /models API — max_prompt_tokens from the user's account.
+    # This catches account-specific models (e.g. claude-opus-4.6-1m) that
+    # don't exist in models.dev. For models that ARE in models.dev, this
+    # returns the provider-enforced limit which is what users can actually use.
+    if effective_provider in ("copilot", "copilot-acp", "github-copilot"):
+        try:
+            from hermes_cli.models import get_copilot_model_context
+            ctx = get_copilot_model_context(model, api_key=api_key)
+            if ctx:
+                return ctx
+        except Exception:
+            pass  # Fall through to models.dev
+
     if effective_provider == "nous":
         ctx = _resolve_nous_context_length(model)
         if ctx:
             return ctx
+    if effective_provider == "openai-codex":
+        # Codex OAuth enforces lower context limits than the direct OpenAI
+        # API for the same slug (e.g. gpt-5.5 is 1.05M on the API but 272K
+        # on Codex). Authoritative source is Codex's own /models endpoint.
+        codex_ctx = _resolve_codex_oauth_context_length(model, access_token=api_key or "")
+        if codex_ctx:
+            if base_url:
+                save_context_length(model, base_url, codex_ctx)
+            return codex_ctx
+    if effective_provider == "gmi" and base_url:
+        # GMI exposes authoritative context_length via /models, but it is not
+        # in models.dev yet. Preserve that higher-fidelity endpoint lookup.
+        ctx = _resolve_endpoint_context_length(model, base_url, api_key=api_key)
+        if ctx is not None:
+            return ctx
     if effective_provider:
         from agent.models_dev import lookup_models_dev_context
         ctx = lookup_models_dev_context(effective_provider, model)
@@ -1155,7 +1406,7 @@ def get_model_context_length(
     # 6. OpenRouter live API metadata (provider-unaware fallback)
     metadata = fetch_model_metadata()
     if model in metadata:
-        return metadata[model].get("context_length", 128000)
+        return metadata[model].get("context_length", DEFAULT_FALLBACK_CONTEXT)
 
     # 8. Hardcoded defaults (fuzzy match — longest key first for specificity)
     # Only check `default_model in model` (is the key a substring of the input).
@@ -1172,10 +1423,11 @@ def get_model_context_length(
     if base_url and is_local_endpoint(base_url):
         local_ctx = _query_local_context_length(model, base_url, api_key=api_key)
         if local_ctx and local_ctx > 0:
-            save_context_length(model, base_url, local_ctx)
+            if provider != "lmstudio":
+                save_context_length(model, base_url, local_ctx)
             return local_ctx
 
-    # 10. Default fallback — 128K
+    # 10. Default fallback — 256K
     return DEFAULT_FALLBACK_CONTEXT
 
 

agent/models_dev.py

@@ -149,6 +149,7 @@ PROVIDER_TO_MODELS_DEV: Dict[str, str] = {
     "stepfun": "stepfun",
     "kimi-coding-cn": "kimi-for-coding",
     "minimax": "minimax",
+    "minimax-oauth": "minimax",
     "minimax-cn": "minimax-cn",
     "deepseek": "deepseek",
     "alibaba": "alibaba",

agent/moonshot_schema.py

@@ -0,0 +1,231 @@
+"""Helpers for translating OpenAI-style tool schemas to Moonshot's schema subset.
+
+Moonshot (Kimi) accepts a stricter subset of JSON Schema than standard OpenAI
+tool calling.  Requests that violate it fail with HTTP 400:
+
+    tools.function.parameters is not a valid moonshot flavored json schema,
+    details: <...>
+
+Known rejection modes documented at
+https://forum.moonshot.ai/t/tool-calling-specification-violation-on-moonshot-api/102
+and MoonshotAI/kimi-cli#1595:
+
+1. Every property schema must carry a ``type``.  Standard JSON Schema allows
+   type to be omitted (the value is then unconstrained); Moonshot refuses.
+2. When ``anyOf`` is used, ``type`` must be on the ``anyOf`` children, not
+   the parent.  Presence of both causes "type should be defined in anyOf
+   items instead of the parent schema".
+
+The ``#/definitions/...`` → ``#/$defs/...`` rewrite for draft-07 refs is
+handled separately in ``tools/mcp_tool._normalize_mcp_input_schema`` so it
+applies at MCP registration time for all providers.
+"""
+
+from __future__ import annotations
+
+import copy
+from typing import Any, Dict, List
+
+# Keys whose values are maps of name → schema (not schemas themselves).
+# When we recurse, we walk the values of these maps as schemas, but we do
+# NOT apply the missing-type repair to the map itself.
+_SCHEMA_MAP_KEYS = frozenset({"properties", "patternProperties", "$defs", "definitions"})
+
+# Keys whose values are lists of schemas.
+_SCHEMA_LIST_KEYS = frozenset({"anyOf", "oneOf", "allOf", "prefixItems"})
+
+# Keys whose values are a single nested schema.
+_SCHEMA_NODE_KEYS = frozenset({"items", "contains", "not", "additionalProperties", "propertyNames"})
+
+
+def _repair_schema(node: Any, is_schema: bool = True) -> Any:
+    """Recursively apply Moonshot repairs to a schema node.
+
+    ``is_schema=True`` means this dict is a JSON Schema node and gets the
+    missing-type + anyOf-parent repairs applied.  ``is_schema=False`` means
+    it's a container map (e.g. the value of ``properties``) and we only
+    recurse into its values.
+    """
+    if isinstance(node, list):
+        # Lists only show up under schema-list keys (anyOf/oneOf/allOf), so
+        # every element is itself a schema.
+        return [_repair_schema(item, is_schema=True) for item in node]
+    if not isinstance(node, dict):
+        return node
+
+    # Walk the dict, deciding per-key whether recursion is into a schema
+    # node, a container map, or a scalar.
+    repaired: Dict[str, Any] = {}
+    for key, value in node.items():
+        if key in _SCHEMA_MAP_KEYS and isinstance(value, dict):
+            # Map of name → schema.  Don't treat the map itself as a schema
+            # (it has no type / properties of its own), but each value is.
+            repaired[key] = {
+                sub_key: _repair_schema(sub_val, is_schema=True)
+                for sub_key, sub_val in value.items()
+            }
+        elif key in _SCHEMA_LIST_KEYS and isinstance(value, list):
+            repaired[key] = [_repair_schema(v, is_schema=True) for v in value]
+        elif key in _SCHEMA_NODE_KEYS:
+            # items / not / additionalProperties: single nested schema.
+            # additionalProperties can also be a bool — leave those alone.
+            if isinstance(value, dict):
+                repaired[key] = _repair_schema(value, is_schema=True)
+            else:
+                repaired[key] = value
+        else:
+            # Scalars (description, title, format, enum values, etc.) pass through.
+            repaired[key] = value
+
+    if not is_schema:
+        return repaired
+
+    # Rule 2: when anyOf is present, type belongs only on the children.
+    # Additionally, Moonshot rejects null-type branches inside anyOf
+    # (enum value (<nil>) does not match any type in [string]).
+    # Collapse the anyOf to the first non-null branch and infer its type.
+    if "anyOf" in repaired and isinstance(repaired["anyOf"], list):
+        repaired.pop("type", None)
+        non_null = [b for b in repaired["anyOf"]
+                    if isinstance(b, dict) and b.get("type") != "null"]
+        if non_null and len(non_null) < len(repaired["anyOf"]):
+            # Drop the anyOf wrapper — keep only the non-null branch.
+            # If there's a single non-null branch, promote it and fall
+            # through to Rules 1/3 so nullable/enum cleanup still applies
+            # to the merged node.
+            if len(non_null) == 1:
+                merge = {k: v for k, v in repaired.items() if k != "anyOf"}
+                merge.update(non_null[0])
+                repaired = merge
+            else:
+                repaired["anyOf"] = non_null
+                return repaired
+        else:
+            # Nothing to collapse — parent type stripped, children already
+            # repaired by the recursive walk above.
+            return repaired
+
+    # Moonshot also rejects non-standard keywords like ``nullable`` on
+    # parameter schemas — strip it.
+    repaired.pop("nullable", None)
+
+    # Rule 1: property schemas without type need one.  $ref nodes are exempt
+    # — their type comes from the referenced definition.
+    # Fill missing type BEFORE Rule 3 so enum cleanup can check the type.
+    if "$ref" not in repaired:
+        repaired = _fill_missing_type(repaired)
+
+    # Rule 3: Moonshot rejects null/empty-string values inside enum arrays
+    # when the parent type is a scalar (string, integer, etc.).  The error:
+    #   "enum value (<nil>) does not match any type in [string]"
+    # Strip null and empty-string from enum values, and if the enum becomes
+    # empty, drop it entirely.
+    if "enum" in repaired and isinstance(repaired["enum"], list):
+        node_type = repaired.get("type")
+        if node_type in ("string", "integer", "number", "boolean"):
+            cleaned = [v for v in repaired["enum"]
+                       if v is not None and v != ""]
+            if cleaned:
+                repaired["enum"] = cleaned
+            else:
+                repaired.pop("enum")
+
+    return repaired
+
+
+def _fill_missing_type(node: Dict[str, Any]) -> Dict[str, Any]:
+    """Infer a reasonable ``type`` if this schema node has none."""
+    if "type" in node and node["type"] not in (None, ""):
+        return node
+
+    # Heuristic: presence of ``properties`` → object, ``items`` → array, ``enum``
+    # → type of first enum value, else fall back to ``string`` (safest scalar).
+    if "properties" in node or "required" in node or "additionalProperties" in node:
+        inferred = "object"
+    elif "items" in node or "prefixItems" in node:
+        inferred = "array"
+    elif "enum" in node and isinstance(node["enum"], list) and node["enum"]:
+        sample = node["enum"][0]
+        if isinstance(sample, bool):
+            inferred = "boolean"
+        elif isinstance(sample, int):
+            inferred = "integer"
+        elif isinstance(sample, float):
+            inferred = "number"
+        else:
+            inferred = "string"
+    else:
+        inferred = "string"
+
+    return {**node, "type": inferred}
+
+
+def sanitize_moonshot_tool_parameters(parameters: Any) -> Dict[str, Any]:
+    """Normalize tool parameters to a Moonshot-compatible object schema.
+
+    Returns a deep-copied schema with the two flavored-JSON-Schema repairs
+    applied.  Input is not mutated.
+    """
+    if not isinstance(parameters, dict):
+        return {"type": "object", "properties": {}}
+
+    repaired = _repair_schema(copy.deepcopy(parameters), is_schema=True)
+    if not isinstance(repaired, dict):
+        return {"type": "object", "properties": {}}
+
+    # Top-level must be an object schema
+    if repaired.get("type") != "object":
+        repaired["type"] = "object"
+    if "properties" not in repaired:
+        repaired["properties"] = {}
+
+    return repaired
+
+
+def sanitize_moonshot_tools(tools: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """Apply ``sanitize_moonshot_tool_parameters`` to every tool's parameters."""
+    if not tools:
+        return tools
+
+    sanitized: List[Dict[str, Any]] = []
+    any_change = False
+    for tool in tools:
+        if not isinstance(tool, dict):
+            sanitized.append(tool)
+            continue
+        fn = tool.get("function")
+        if not isinstance(fn, dict):
+            sanitized.append(tool)
+            continue
+        params = fn.get("parameters")
+        repaired = sanitize_moonshot_tool_parameters(params)
+        if repaired is not params:
+            any_change = True
+            new_fn = {**fn, "parameters": repaired}
+            sanitized.append({**tool, "function": new_fn})
+        else:
+            sanitized.append(tool)
+
+    return sanitized if any_change else tools
+
+
+def is_moonshot_model(model: str | None) -> bool:
+    """True for any Kimi / Moonshot model slug, regardless of aggregator prefix.
+
+    Matches bare names (``kimi-k2.6``, ``moonshotai/Kimi-K2.6``) and aggregator-
+    prefixed slugs (``nous/moonshotai/kimi-k2.6``, ``openrouter/moonshotai/...``).
+    Detection by model name covers Nous / OpenRouter / other aggregators that
+    route to Moonshot's inference, where the base URL is the aggregator's, not
+    ``api.moonshot.ai``.
+    """
+    if not model:
+        return False
+    bare = model.strip().lower()
+    # Last path segment (covers aggregator-prefixed slugs)
+    tail = bare.rsplit("/", 1)[-1]
+    if tail.startswith("kimi-") or tail == "kimi":
+        return True
+    # Vendor-prefixed forms commonly used on aggregators
+    if "moonshot" in bare or "/kimi" in bare or bare.startswith("kimi"):
+        return True
+    return False

agent/nous_rate_guard.py

@@ -18,6 +18,7 @@ import os
 import tempfile
 import time
 from typing import Any, Mapping, Optional
+from utils import atomic_replace
 
 logger = logging.getLogger(__name__)
 
@@ -118,7 +119,7 @@ def record_nous_rate_limit(
         try:
             with os.fdopen(fd, "w") as f:
                 json.dump(state, f)
-            os.replace(tmp_path, path)
+            atomic_replace(tmp_path, path)
         except Exception:
             # Clean up temp file on failure
             try:
@@ -180,3 +181,145 @@ def format_remaining(seconds: float) -> str:
     h, remainder = divmod(s, 3600)
     m = remainder // 60
     return f"{h}h {m}m" if m else f"{h}h"
+
+
+# Buckets with reset windows shorter than this are treated as transient
+# (upstream jitter, secondary throttling) rather than a genuine quota
+# exhaustion worth a cross-session breaker trip.
+_MIN_RESET_FOR_BREAKER_SECONDS = 60.0
+
+
+def is_genuine_nous_rate_limit(
+    *,
+    headers: Optional[Mapping[str, str]] = None,
+    last_known_state: Optional[Any] = None,
+) -> bool:
+    """Decide whether a 429 from Nous Portal is a real account rate limit.
+
+    Nous Portal multiplexes multiple upstream providers (DeepSeek, Kimi,
+    MiMo, Hermes, ...) behind one endpoint.  A 429 can mean either:
+
+      (a) The caller's own RPM / RPH / TPM / TPH bucket on Nous is
+          exhausted — a genuine rate limit that will last until the
+          bucket resets.
+      (b) The upstream provider is out of capacity for a specific model
+          — transient, clears in seconds, and has nothing to do with
+          the caller's quota on Nous.
+
+    Tripping the cross-session breaker on (b) blocks ALL Nous requests
+    (and all models, since Nous is one provider key) for minutes even
+    though the caller's account is healthy and a different model would
+    have worked.  That's the bug users hit when DeepSeek V4 Pro 429s
+    trigger a breaker that then blocks Kimi 2.6 and MiMo V2.5 Pro.
+
+    We tell the two apart by looking at:
+
+      1. The 429 response's own ``x-ratelimit-*`` headers.  Nous emits
+         the full suite on every response including 429s.  An exhausted
+         bucket (``remaining == 0`` with a reset window >= 60s) is
+         proof of (a).
+      2. The last-known-good rate-limit state captured by
+         ``_capture_rate_limits()`` on the previous successful
+         response.  If any bucket there was already near-exhausted with
+         a substantial reset window, the current 429 is almost
+         certainly (a) continuing from that condition.
+
+    If neither signal fires, we treat the 429 as (b): fail the single
+    request, let the retry loop or model-switch proceed, and do NOT
+    write the cross-session breaker file.
+
+    Returns True when the evidence points at (a).
+    """
+    # Signal 1: current 429 response headers.
+    state = _parse_buckets_from_headers(headers)
+    if _has_exhausted_bucket(state):
+        return True
+
+    # Signal 2: last-known-good state from a recent successful response.
+    # Accepts either a RateLimitState (dataclass from rate_limit_tracker)
+    # or a dict of bucket snapshots.
+    if last_known_state is not None and _has_exhausted_bucket_in_object(last_known_state):
+        return True
+
+    return False
+
+
+def _parse_buckets_from_headers(
+    headers: Optional[Mapping[str, str]],
+) -> dict[str, tuple[Optional[int], Optional[float]]]:
+    """Extract (remaining, reset_seconds) per bucket from x-ratelimit-* headers.
+
+    Returns empty dict when no rate-limit headers are present.
+    """
+    if not headers:
+        return {}
+
+    lowered = {k.lower(): v for k, v in headers.items()}
+    if not any(k.startswith("x-ratelimit-") for k in lowered):
+        return {}
+
+    def _maybe_int(raw: Optional[str]) -> Optional[int]:
+        if raw is None:
+            return None
+        try:
+            return int(float(raw))
+        except (TypeError, ValueError):
+            return None
+
+    def _maybe_float(raw: Optional[str]) -> Optional[float]:
+        if raw is None:
+            return None
+        try:
+            return float(raw)
+        except (TypeError, ValueError):
+            return None
+
+    result: dict[str, tuple[Optional[int], Optional[float]]] = {}
+    for tag in ("requests", "requests-1h", "tokens", "tokens-1h"):
+        remaining = _maybe_int(lowered.get(f"x-ratelimit-remaining-{tag}"))
+        reset = _maybe_float(lowered.get(f"x-ratelimit-reset-{tag}"))
+        if remaining is not None or reset is not None:
+            result[tag] = (remaining, reset)
+    return result
+
+
+def _has_exhausted_bucket(
+    buckets: Mapping[str, tuple[Optional[int], Optional[float]]],
+) -> bool:
+    """Return True when any bucket has remaining == 0 AND a meaningful reset window."""
+    for remaining, reset in buckets.values():
+        if remaining is None or remaining > 0:
+            continue
+        if reset is None:
+            continue
+        if reset >= _MIN_RESET_FOR_BREAKER_SECONDS:
+            return True
+    return False
+
+
+def _has_exhausted_bucket_in_object(state: Any) -> bool:
+    """Check a RateLimitState-like object for an exhausted bucket.
+
+    Accepts the dataclass from ``agent.rate_limit_tracker`` (buckets
+    exposed as attributes ``requests_min``, ``requests_hour``,
+    ``tokens_min``, ``tokens_hour``) and falls back gracefully for any
+    object missing those attributes.
+    """
+    for attr in ("requests_min", "requests_hour", "tokens_min", "tokens_hour"):
+        bucket = getattr(state, attr, None)
+        if bucket is None:
+            continue
+        limit = getattr(bucket, "limit", 0) or 0
+        remaining = getattr(bucket, "remaining", 0) or 0
+        # Prefer the adjusted "remaining_seconds_now" property when present;
+        # fall back to raw reset_seconds.
+        reset = getattr(bucket, "remaining_seconds_now", None)
+        if reset is None:
+            reset = getattr(bucket, "reset_seconds", 0.0) or 0.0
+        if limit <= 0:
+            continue
+        if remaining > 0:
+            continue
+        if reset >= _MIN_RESET_FOR_BREAKER_SECONDS:
+            return True
+    return False

agent/onboarding.py

@@ -0,0 +1,193 @@
+"""
+Contextual first-touch onboarding hints.
+
+Instead of blocking first-run questionnaires, show a one-time hint the *first*
+time a user hits a behavior fork — message-while-running, first long-running
+tool, etc.  Each hint is shown once per install (tracked in ``config.yaml`` under
+``onboarding.seen.<flag>``) and then never again.
+
+Keep this module tiny and dependency-free so both the CLI and gateway can import
+it without pulling in heavy modules.
+"""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Any, Mapping, Optional
+
+logger = logging.getLogger(__name__)
+
+
+# -------------------------------------------------------------------------
+# Flag names (stable — used as config.yaml keys under onboarding.seen)
+# -------------------------------------------------------------------------
+
+BUSY_INPUT_FLAG = "busy_input_prompt"
+TOOL_PROGRESS_FLAG = "tool_progress_prompt"
+OPENCLAW_RESIDUE_FLAG = "openclaw_residue_cleanup"
+
+
+# -------------------------------------------------------------------------
+# Hint content
+# -------------------------------------------------------------------------
+
+def busy_input_hint_gateway(mode: str) -> str:
+    """Hint shown the first time a user messages while the agent is busy.
+
+    ``mode`` is the effective busy_input_mode that was just applied, so the
+    message matches reality ("I just interrupted…" vs "I just queued…").
+    """
+    if mode == "queue":
+        return (
+            "💡 First-time tip — I queued your message instead of interrupting. "
+            "Send `/busy interrupt` to make new messages stop the current task "
+            "immediately, or `/busy status` to check. This notice won't appear again."
+        )
+    if mode == "steer":
+        return (
+            "💡 First-time tip — I steered your message into the current run; "
+            "it will arrive after the next tool call instead of interrupting. "
+            "Send `/busy interrupt` or `/busy queue` to change this, or "
+            "`/busy status` to check. This notice won't appear again."
+        )
+    return (
+        "💡 First-time tip — I just interrupted my current task to answer you. "
+        "Send `/busy queue` to queue follow-ups for after the current task instead, "
+        "`/busy steer` to inject them mid-run without interrupting, or "
+        "`/busy status` to check. This notice won't appear again."
+    )
+
+
+def busy_input_hint_cli(mode: str) -> str:
+    """CLI version of the busy-input hint (plain text, no markdown)."""
+    if mode == "queue":
+        return (
+            "(tip) Your message was queued for the next turn. "
+            "Use /busy interrupt to make Enter stop the current run instead, "
+            "or /busy steer to inject mid-run. This tip only shows once."
+        )
+    if mode == "steer":
+        return (
+            "(tip) Your message was steered into the current run; it arrives "
+            "after the next tool call. Use /busy interrupt or /busy queue to "
+            "change this. This tip only shows once."
+        )
+    return (
+        "(tip) Your message interrupted the current run. "
+        "Use /busy queue to queue messages for the next turn instead, "
+        "or /busy steer to inject mid-run. This tip only shows once."
+    )
+
+
+def tool_progress_hint_gateway() -> str:
+    return (
+        "💡 First-time tip — that tool took a while and I'm streaming every step. "
+        "If the progress messages feel noisy, send `/verbose` to cycle modes "
+        "(all → new → off). This notice won't appear again."
+    )
+
+
+def tool_progress_hint_cli() -> str:
+    return (
+        "(tip) That tool ran for a while. Use /verbose to cycle tool-progress "
+        "display modes (all -> new -> off -> verbose). This tip only shows once."
+    )
+
+
+def openclaw_residue_hint_cli() -> str:
+    """Banner shown the first time Hermes starts and finds ``~/.openclaw/``.
+
+    Points users at ``hermes claw migrate`` (non-destructive port of config,
+    memory, and skills) first. ``hermes claw cleanup`` is mentioned as the
+    follow-up step for users who have already migrated and want to archive
+    the old directory — with a warning that archiving breaks OpenClaw.
+    """
+    return (
+        "A legacy OpenClaw directory was detected at ~/.openclaw/.\n"
+        "To port your config, memory, and skills over to Hermes, run "
+        "`hermes claw migrate`.\n"
+        "If you've already migrated and want to archive the old directory, "
+        "run `hermes claw cleanup` (renames it to ~/.openclaw.pre-migration — "
+        "OpenClaw will stop working after this).\n"
+        "This tip only shows once."
+    )
+
+
+def detect_openclaw_residue(home: Optional[Path] = None) -> bool:
+    """Return True if an OpenClaw workspace directory is present in ``$HOME``.
+
+    Pure filesystem check — no side effects. ``home`` override exists for tests.
+    """
+    base = home or Path.home()
+    try:
+        return (base / ".openclaw").is_dir()
+    except OSError:
+        return False
+
+
+# -------------------------------------------------------------------------
+# State read / write
+# -------------------------------------------------------------------------
+
+def _get_seen_dict(config: Mapping[str, Any]) -> Mapping[str, Any]:
+    onboarding = config.get("onboarding") if isinstance(config, Mapping) else None
+    if not isinstance(onboarding, Mapping):
+        return {}
+    seen = onboarding.get("seen")
+    return seen if isinstance(seen, Mapping) else {}
+
+
+def is_seen(config: Mapping[str, Any], flag: str) -> bool:
+    """Return True if the user has already been shown this first-touch hint."""
+    return bool(_get_seen_dict(config).get(flag))
+
+
+def mark_seen(config_path: Path, flag: str) -> bool:
+    """Persist ``onboarding.seen.<flag> = True`` to ``config_path``.
+
+    Uses the atomic YAML writer so a concurrent process can't observe a
+    partially-written file.  Returns True on success, False on any error
+    (including the config file being absent — onboarding is best-effort).
+    """
+    try:
+        import yaml
+        from utils import atomic_yaml_write
+    except Exception as e:  # pragma: no cover — dependency issue
+        logger.debug("onboarding: failed to import yaml/utils: %s", e)
+        return False
+
+    try:
+        cfg: dict = {}
+        if config_path.exists():
+            with open(config_path, encoding="utf-8") as f:
+                cfg = yaml.safe_load(f) or {}
+        if not isinstance(cfg.get("onboarding"), dict):
+            cfg["onboarding"] = {}
+        seen = cfg["onboarding"].get("seen")
+        if not isinstance(seen, dict):
+            seen = {}
+            cfg["onboarding"]["seen"] = seen
+        if seen.get(flag) is True:
+            return True  # already marked — nothing to do
+        seen[flag] = True
+        atomic_yaml_write(config_path, cfg)
+        return True
+    except Exception as e:
+        logger.debug("onboarding: failed to mark flag %s: %s", flag, e)
+        return False
+
+
+__all__ = [
+    "BUSY_INPUT_FLAG",
+    "TOOL_PROGRESS_FLAG",
+    "OPENCLAW_RESIDUE_FLAG",
+    "busy_input_hint_gateway",
+    "busy_input_hint_cli",
+    "tool_progress_hint_gateway",
+    "tool_progress_hint_cli",
+    "openclaw_residue_hint_cli",
+    "detect_openclaw_residue",
+    "is_seen",
+    "mark_seen",
+]

agent/prompt_builder.py

@@ -141,6 +141,12 @@ DEFAULT_AGENT_IDENTITY = (
     "Be targeted and efficient in your exploration and investigations."
 )
 
+HERMES_AGENT_HELP_GUIDANCE = (
+    "If the user asks about configuring, setting up, or using Hermes Agent "
+    "itself, load the `hermes-agent` skill with skill_view(name='hermes-agent') "
+    "before answering. Docs: https://hermes-agent.nousresearch.com/docs"
+)
+
 MEMORY_GUIDANCE = (
     "You have persistent memory across sessions. Save durable facts using the memory "
     "tool: user preferences, environment details, tool quirks, and stable conventions. "
@@ -176,6 +182,64 @@ SKILLS_GUIDANCE = (
     "Skills that aren't maintained become liabilities."
 )
 
+KANBAN_GUIDANCE = (
+    "# Kanban task execution protocol\n"
+    "You have been assigned ONE task from "
+    "the shared board at `~/.hermes/kanban.db`. Your task id is in "
+    "`$HERMES_KANBAN_TASK`; your workspace is `$HERMES_KANBAN_WORKSPACE`. "
+    "The `kanban_*` tools in your schema are your primary coordination surface — "
+    "they write directly to the shared SQLite DB and work regardless of terminal "
+    "backend (local/docker/modal/ssh).\n"
+    "\n"
+    "## Lifecycle\n"
+    "\n"
+    "1. **Orient.** Call `kanban_show()` first (no args — it defaults to your "
+    "task). The response includes title, body, parent-task handoffs (summary + "
+    "metadata), any prior attempts on this task if you're a retry, the full "
+    "comment thread, and a pre-formatted `worker_context` you can treat as "
+    "ground truth.\n"
+    "2. **Work inside the workspace.** `cd $HERMES_KANBAN_WORKSPACE` before "
+    "any file operations. The workspace is yours for this run. Don't modify "
+    "files outside it unless the task explicitly asks.\n"
+    "3. **Heartbeat on long operations.** Call `kanban_heartbeat(note=...)` "
+    "every few minutes during long subprocesses (training, encoding, crawling). "
+    "Skip heartbeats for short tasks.\n"
+    "4. **Block on genuine ambiguity.** If you need a human decision you cannot "
+    "infer (missing credentials, UX choice, paywalled source, peer output you "
+    "need first), call `kanban_block(reason=\"...\")` and stop. Don't guess. "
+    "The user will unblock with context and the dispatcher will respawn you.\n"
+    "5. **Complete with structured handoff.** Call `kanban_complete(summary=..., "
+    "metadata=...)`. `summary` is 1–3 human-readable sentences naming concrete "
+    "artifacts. `metadata` is machine-readable facts "
+    "(`{changed_files: [...], tests_run: N, decisions: [...]}`). Downstream "
+    "workers read both via their own `kanban_show`. Never put secrets / "
+    "tokens / raw PII in either field — run rows are durable forever.\n"
+    "6. **If follow-up work appears, create it; don't do it.** Use "
+    "`kanban_create(title=..., assignee=<right-profile>, parents=[your-task-id])` "
+    "to spawn a child task for the appropriate specialist profile instead of "
+    "scope-creeping into the next thing.\n"
+    "\n"
+    "## Orchestrator mode\n"
+    "\n"
+    "If your task is itself a decomposition task (e.g. a planner profile given "
+    "a high-level goal), use `kanban_create` to fan out into child tasks — one "
+    "per specialist, each with an explicit `assignee` and `parents=[...]` to "
+    "express dependencies. Then `kanban_complete` your own task with a summary "
+    "of the decomposition. Do NOT execute the work yourself; your job is "
+    "routing, not implementation.\n"
+    "\n"
+    "## Do NOT\n"
+    "\n"
+    "- Do not shell out to `hermes kanban <verb>` for board operations. Use "
+    "the `kanban_*` tools — they work across all terminal backends.\n"
+    "- Do not complete a task you didn't actually finish. Block it.\n"
+    "- Do not assign follow-up work to yourself. Assign it to the right "
+    "specialist profile.\n"
+    "- Do not call `delegate_task` as a board substitute. `delegate_task` is "
+    "for short reasoning subtasks inside your own run; board tasks are for "
+    "cross-agent handoffs that outlive one API loop."
+)
+
 TOOL_USE_ENFORCEMENT_GUIDANCE = (
     "# Tool-use enforcement\n"
     "You MUST use your tools to take action — do not describe what you would do "
@@ -304,6 +368,10 @@ PLATFORM_HINTS = {
         "Standard markdown is automatically converted to Telegram format. "
         "Supported: **bold**, *italic*, ~~strikethrough~~, ||spoiler||, "
         "`inline code`, ```code blocks```, [links](url), and ## headers. "
+        "Telegram has NO table syntax — prefer bullet lists or labeled "
+        "key: value pairs over pipe tables (any tables you do emit are "
+        "auto-rewritten into row-group bullets, which you can produce "
+        "directly for cleaner output). "
         "You can send media files natively: to deliver a file to the user, "
         "include MEDIA:/absolute/path/to/file in your response. Images "
         "(.png, .jpg, .webp) appear as photos, audio (.ogg) sends as voice "
@@ -422,6 +490,35 @@ PLATFORM_HINTS = {
         "your response. Images are sent as native photos, and other files arrive as downloadable "
         "documents."
     ),
+    "yuanbao": (
+        "You are on Yuanbao (腾讯元宝), a Chinese AI assistant platform. "
+        "Markdown formatting is supported (code blocks, tables, bold/italic). "
+        "You CAN send media files natively — to deliver a file to the user, include "
+        "MEDIA:/absolute/path/to/file in your response. The file will be sent as a native "
+        "Yuanbao attachment: images (.jpg, .png, .webp, .gif) are sent as photos, "
+        "and other files (.pdf, .docx, .txt, .zip, etc.) arrive as downloadable documents "
+        "(max 50 MB). You can also include image URLs in markdown format ![alt](url) and "
+        "they will be downloaded and sent as native photos. "
+        "Do NOT tell the user you lack file-sending capability — use MEDIA: syntax "
+        "whenever a file delivery is appropriate.\n\n"
+        "Stickers (贴纸 / 表情包 / TIM face): Yuanbao has a built-in sticker catalogue. "
+        "When the user sends a sticker (you see '[emoji: 名称]' in their message) or asks "
+        "you to send/reply-with a 贴纸/表情/表情包, you MUST use the sticker tools:\n"
+        "  1. Call yb_search_sticker with a Chinese keyword (e.g. '666', '比心', '吃瓜', "
+        "     '捂脸', '合十') to discover matching sticker_ids.\n"
+        "  2. Call yb_send_sticker with the chosen sticker_id or name — this sends a real "
+        "     TIMFaceElem that renders as a native sticker in the chat.\n"
+        "DO NOT draw sticker-like PNGs with execute_code/Pillow/matplotlib and then send "
+        "them via MEDIA: or send_image_file. That produces a fake low-quality 'sticker' "
+        "image and is the WRONG path. Bare Unicode emoji in text is also not a substitute "
+        "— when a sticker is the right response, use yb_send_sticker."
+    ),
+    "api_server": (
+        "You're responding through an API server. The rendering layer is unknown — "
+        "assume plain text. No markdown formatting (no asterisks, bullets, headers, "
+        "code fences). Treat this like a conversation, not a document. Keep responses "
+        "brief and natural."
+    ),
 }
 
 # ---------------------------------------------------------------------------
@@ -825,6 +922,11 @@ def build_skills_system_prompt(
             "Skills also encode the user's preferred approach, conventions, and quality standards "
             "for tasks like code review, planning, and testing — load them even for tasks you "
             "already know how to do, because the skill defines how it should be done here.\n"
+            "Whenever the user asks you to configure, set up, install, enable, disable, modify, "
+            "or troubleshoot Hermes Agent itself — its CLI, config, models, providers, tools, "
+            "skills, voice, gateway, plugins, or any feature — load the `hermes-agent` skill "
+            "first. It has the actual commands (e.g. `hermes config set …`, `hermes tools`, "
+            "`hermes setup`) so you don't have to guess or invent workarounds.\n"
             "If a skill has issues, fix it with skill_manage(action='patch').\n"
             "After difficult/iterative tasks, offer to save as a skill. "
             "If a skill you loaded was missing steps, had wrong commands, or needed "

agent/redact.py

@@ -56,8 +56,12 @@ _SENSITIVE_BODY_KEYS = frozenset({
 })
 
 # Snapshot at import time so runtime env mutations (e.g. LLM-generated
-# `export HERMES_REDACT_SECRETS=false`) cannot disable redaction mid-session.
-_REDACT_ENABLED = os.getenv("HERMES_REDACT_SECRETS", "").lower() not in ("0", "false", "no", "off")
+# `export HERMES_REDACT_SECRETS=true`) cannot enable/disable redaction
+# mid-session.  OFF by default — user must opt in via
+# `security.redact_secrets: true` in config.yaml (bridged to this env var
+# in hermes_cli/main.py and gateway/run.py) or `HERMES_REDACT_SECRETS=true`
+# in ~/.hermes/.env.
+_REDACT_ENABLED = os.getenv("HERMES_REDACT_SECRETS", "").lower() in ("1", "true", "yes", "on")
 
 # Known API key prefixes -- match the prefix + contiguous token chars
 _PREFIX_PATTERNS = [
@@ -180,11 +184,59 @@ _PREFIX_RE = re.compile(
 )
 
 
+def mask_secret(
+    value: str,
+    *,
+    head: int = 4,
+    tail: int = 4,
+    floor: int = 12,
+    placeholder: str = "***",
+    empty: str = "",
+) -> str:
+    """Mask a secret for display, preserving ``head`` and ``tail`` characters.
+
+    Canonical helper for display-time redaction across Hermes — used by
+    ``hermes config``, ``hermes status``, ``hermes dump``, and anywhere
+    a secret needs to be shown truncated for debuggability while still
+    keeping the bulk hidden.
+
+    Args:
+        value:       The secret to mask. ``None``/empty returns ``empty``.
+        head:        Leading characters to preserve. Default 4.
+        tail:        Trailing characters to preserve. Default 4.
+        floor:       Values shorter than ``head + tail + floor_margin`` are
+                     fully masked (returns ``placeholder``). Default 12 —
+                     matches the existing config/status/dump convention.
+        placeholder: Value returned for too-short inputs. Default ``"***"``.
+        empty:       Value returned when ``value`` is falsy (None, ""). The
+                     caller can override this to e.g. ``color("(not set)",
+                     Colors.DIM)`` for user-facing display.
+
+    Examples:
+        >>> mask_secret("sk-proj-abcdef1234567890")
+        'sk-p...7890'
+        >>> mask_secret("short")                         # fully masked
+        '***'
+        >>> mask_secret("")                              # empty default
+        ''
+        >>> mask_secret("", empty="(not set)")           # empty override
+        '(not set)'
+        >>> mask_secret("long-token", head=6, tail=4, floor=18)
+        '***'
+    """
+    if not value:
+        return empty
+    if len(value) < floor:
+        return placeholder
+    return f"{value[:head]}...{value[-tail:]}"
+
+
 def _mask_token(token: str) -> str:
-    """Mask a token, preserving prefix for long tokens."""
-    if len(token) < 18:
+    """Mask a log token — conservative 18-char floor, preserves 6 prefix / 4 suffix."""
+    # Empty input: historically this returned "***" rather than "". Preserve.
+    if not token:
         return "***"
-    return f"{token[:6]}...{token[-4:]}"
+    return mask_secret(token, head=6, tail=4, floor=18)
 
 
 def _redact_query_string(query: str) -> str:
@@ -253,11 +305,18 @@ def _redact_form_body(text: str) -> str:
     return _redact_query_string(text.strip())
 
 
-def redact_sensitive_text(text: str) -> str:
+def redact_sensitive_text(text: str, *, force: bool = False, code_file: bool = False) -> str:
     """Apply all redaction patterns to a block of text.
 
     Safe to call on any string -- non-matching text passes through unchanged.
-    Disabled when security.redact_secrets is false in config.yaml.
+    Disabled by default — enable via security.redact_secrets: true in config.yaml.
+    Set force=True for safety boundaries that must never return raw secrets
+    regardless of the user's global logging redaction preference.
+
+    Set code_file=True to skip the ENV-assignment and JSON-field regex
+    patterns when the text is known to be source code (e.g. MAX_TOKENS=***
+    constants, "apiKey": "test" fixtures). Prefix patterns, auth headers,
+    private keys, DB connstrings, JWTs, and URL secrets are still redacted.
     """
     if text is None:
         return None
@@ -265,23 +324,24 @@ def redact_sensitive_text(text: str) -> str:
         text = str(text)
     if not text:
         return text
-    if not _REDACT_ENABLED:
+    if not (force or _REDACT_ENABLED):
         return text
 
     # Known prefixes (sk-, ghp_, etc.)
     text = _PREFIX_RE.sub(lambda m: _mask_token(m.group(1)), text)
 
-    # ENV assignments: OPENAI_API_KEY=sk-abc...
-    def _redact_env(m):
-        name, quote, value = m.group(1), m.group(2), m.group(3)
-        return f"{name}={quote}{_mask_token(value)}{quote}"
-    text = _ENV_ASSIGN_RE.sub(_redact_env, text)
+    # ENV assignments: OPENAI_API_KEY=***  (skip for code files — false positives)
+    if not code_file:
+        def _redact_env(m):
+            name, quote, value = m.group(1), m.group(2), m.group(3)
+            return f"{name}={quote}{_mask_token(value)}{quote}"
+        text = _ENV_ASSIGN_RE.sub(_redact_env, text)
 
-    # JSON fields: "apiKey": "value"
-    def _redact_json(m):
-        key, value = m.group(1), m.group(2)
-        return f'{key}: "{_mask_token(value)}"'
-    text = _JSON_FIELD_RE.sub(_redact_json, text)
+        # JSON fields: "apiKey": "***"  (skip for code files — false positives)
+        def _redact_json(m):
+            key, value = m.group(1), m.group(2)
+            return f'{key}: "{_mask_token(value)}"'
+        text = _JSON_FIELD_RE.sub(_redact_json, text)
 
     # Authorization headers
     text = _AUTH_HEADER_RE.sub(

agent/shell_hooks.py

@@ -76,6 +76,7 @@ except ImportError:  # pragma: no cover
     fcntl = None  # type: ignore[assignment]
 
 from hermes_constants import get_hermes_home
+from utils import atomic_replace
 
 logger = logging.getLogger(__name__)
 
@@ -568,7 +569,7 @@ def save_allowlist(data: Dict[str, Any]) -> None:
         try:
             with os.fdopen(fd, "w") as fh:
                 fh.write(json.dumps(data, indent=2, sort_keys=True))
-            os.replace(tmp_path, p)
+            atomic_replace(tmp_path, p)
         except Exception:
             try:
                 os.unlink(tmp_path)
@@ -754,7 +755,11 @@ def _resolve_effective_accept(
     if env in ("1", "true", "yes", "on"):
         return True
     cfg_val = cfg.get("hooks_auto_accept", False)
-    return bool(cfg_val)
+    if isinstance(cfg_val, bool):
+        return cfg_val
+    if isinstance(cfg_val, str):
+        return cfg_val.strip().lower() in ("1", "true", "yes", "on")
+    return False
 
 
 # ---------------------------------------------------------------------------

agent/skill_commands.py

@@ -1,153 +1,54 @@
-"""Shared slash command helpers for skills and built-in prompt-style modes.
+"""Shared slash command helpers for skills.
 
 Shared between CLI (cli.py) and gateway (gateway/run.py) so both surfaces
-can invoke skills via /skill-name commands and prompt-only built-ins like
-/plan.
+can invoke skills via /skill-name commands.
 """
 
 import json
 import logging
+import os
 import re
-import subprocess
-from datetime import datetime
 from pathlib import Path
 from typing import Any, Dict, Optional
 
 from hermes_constants import display_hermes_home
+from agent.skill_preprocessing import (
+    expand_inline_shell as _expand_inline_shell,
+    load_skills_config as _load_skills_config,
+    substitute_template_vars as _substitute_template_vars,
+)
 
 logger = logging.getLogger(__name__)
 
 _skill_commands: Dict[str, Dict[str, Any]] = {}
-_PLAN_SLUG_RE = re.compile(r"[^a-z0-9]+")
+_skill_commands_platform: Optional[str] = None
 # Patterns for sanitizing skill names into clean hyphen-separated slugs.
 _SKILL_INVALID_CHARS = re.compile(r"[^a-z0-9-]")
 _SKILL_MULTI_HYPHEN = re.compile(r"-{2,}")
 
-# Matches ${HERMES_SKILL_DIR} / ${HERMES_SESSION_ID} tokens in SKILL.md.
-# Tokens that don't resolve (e.g. ${HERMES_SESSION_ID} with no session) are
-# left as-is so the user can debug them.
-_SKILL_TEMPLATE_RE = re.compile(r"\$\{(HERMES_SKILL_DIR|HERMES_SESSION_ID)\}")
 
-# Matches inline shell snippets like:  !`date +%Y-%m-%d`
-# Non-greedy, single-line only — no newlines inside the backticks.
-_INLINE_SHELL_RE = re.compile(r"!`([^`\n]+)`")
+def _resolve_skill_commands_platform() -> Optional[str]:
+    """Return the current platform scope used for disabled-skill filtering.
 
-# Cap inline-shell output so a runaway command can't blow out the context.
-_INLINE_SHELL_MAX_OUTPUT = 4000
+    Used to detect when the active platform has shifted so
+    :func:`get_skill_commands` can drop a stale cache that was populated
+    for a different platform's ``skills.platform_disabled`` view (#14536).
 
-
-def _load_skills_config() -> dict:
-    """Load the ``skills`` section of config.yaml (best-effort)."""
-    try:
-        from hermes_cli.config import load_config
-
-        cfg = load_config() or {}
-        skills_cfg = cfg.get("skills")
-        if isinstance(skills_cfg, dict):
-            return skills_cfg
-    except Exception:
-        logger.debug("Could not read skills config", exc_info=True)
-    return {}
-
-
-def _substitute_template_vars(
-    content: str,
-    skill_dir: Path | None,
-    session_id: str | None,
-) -> str:
-    """Replace ${HERMES_SKILL_DIR} / ${HERMES_SESSION_ID} in skill content.
-
-    Only substitutes tokens for which a concrete value is available —
-    unresolved tokens are left in place so the author can spot them.
-    """
-    if not content:
-        return content
-
-    skill_dir_str = str(skill_dir) if skill_dir else None
-
-    def _replace(match: re.Match) -> str:
-        token = match.group(1)
-        if token == "HERMES_SKILL_DIR" and skill_dir_str:
-            return skill_dir_str
-        if token == "HERMES_SESSION_ID" and session_id:
-            return str(session_id)
-        return match.group(0)
-
-    return _SKILL_TEMPLATE_RE.sub(_replace, content)
-
-
-def _run_inline_shell(command: str, cwd: Path | None, timeout: int) -> str:
-    """Execute a single inline-shell snippet and return its stdout (trimmed).
-
-    Failures return a short ``[inline-shell error: ...]`` marker instead of
-    raising, so one bad snippet can't wreck the whole skill message.
+    Resolves from (in order) ``HERMES_PLATFORM`` env var and
+    ``HERMES_SESSION_PLATFORM`` from the gateway session context. Returns
+    ``None`` when no platform scope is active (e.g. classic CLI, RL
+    rollouts, standalone scripts).
     """
     try:
-        completed = subprocess.run(
-            ["bash", "-c", command],
-            cwd=str(cwd) if cwd else None,
-            capture_output=True,
-            text=True,
-            timeout=max(1, int(timeout)),
-            check=False,
+        from gateway.session_context import get_session_env
+
+        resolved_platform = (
+            os.getenv("HERMES_PLATFORM")
+            or get_session_env("HERMES_SESSION_PLATFORM")
         )
-    except subprocess.TimeoutExpired:
-        return f"[inline-shell timeout after {timeout}s: {command}]"
-    except FileNotFoundError:
-        return f"[inline-shell error: bash not found]"
-    except Exception as exc:
-        return f"[inline-shell error: {exc}]"
-
-    output = (completed.stdout or "").rstrip("\n")
-    if not output and completed.stderr:
-        output = completed.stderr.rstrip("\n")
-    if len(output) > _INLINE_SHELL_MAX_OUTPUT:
-        output = output[:_INLINE_SHELL_MAX_OUTPUT] + "…[truncated]"
-    return output
-
-
-def _expand_inline_shell(
-    content: str,
-    skill_dir: Path | None,
-    timeout: int,
-) -> str:
-    """Replace every !`cmd` snippet in ``content`` with its stdout.
-
-    Runs each snippet with the skill directory as CWD so relative paths in
-    the snippet work the way the author expects.
-    """
-    if "!`" not in content:
-        return content
-
-    def _replace(match: re.Match) -> str:
-        cmd = match.group(1).strip()
-        if not cmd:
-            return ""
-        return _run_inline_shell(cmd, skill_dir, timeout)
-
-    return _INLINE_SHELL_RE.sub(_replace, content)
-
-
-def build_plan_path(
-    user_instruction: str = "",
-    *,
-    now: datetime | None = None,
-) -> Path:
-    """Return the default workspace-relative markdown path for a /plan invocation.
-
-    Relative paths are intentional: file tools are task/backend-aware and resolve
-    them against the active working directory for local, docker, ssh, modal,
-    daytona, and similar terminal backends. That keeps the plan with the active
-    workspace instead of the Hermes host's global home directory.
-    """
-    slug_source = (user_instruction or "").strip().splitlines()[0] if user_instruction else ""
-    slug = _PLAN_SLUG_RE.sub("-", slug_source.lower()).strip("-")
-    if slug:
-        slug = "-".join(part for part in slug.split("-")[:8] if part)[:48].strip("-")
-    slug = slug or "conversation-plan"
-    timestamp = (now or datetime.now()).strftime("%Y-%m-%d_%H%M%S")
-    return Path(".hermes") / "plans" / f"{timestamp}-{slug}.md"
-
+    except Exception:
+        resolved_platform = os.getenv("HERMES_PLATFORM")
+    return resolved_platform or None
 
 def _load_skill_payload(skill_identifier: str, task_id: str | None = None) -> tuple[dict[str, Any], Path | None, str] | None:
     """Load a skill by name/path and return (loaded_payload, skill_dir, display_name)."""
@@ -167,7 +68,9 @@ def _load_skill_payload(skill_identifier: str, task_id: str | None = None) -> tu
         else:
             normalized = raw_identifier.lstrip("/")
 
-        loaded_skill = json.loads(skill_view(normalized, task_id=task_id))
+        loaded_skill = json.loads(
+            skill_view(normalized, task_id=task_id, preprocess=False)
+        )
     except Exception:
         return None
 
@@ -341,11 +244,12 @@ def scan_skill_commands() -> Dict[str, Dict[str, Any]]:
     Returns:
         Dict mapping "/skill-name" to {name, description, skill_md_path, skill_dir}.
     """
-    global _skill_commands
+    global _skill_commands, _skill_commands_platform
+    _skill_commands_platform = _resolve_skill_commands_platform()
     _skill_commands = {}
     try:
         from tools.skills_tool import SKILLS_DIR, _parse_frontmatter, skill_matches_platform, _get_disabled_skill_names
-        from agent.skill_utils import get_external_skills_dirs
+        from agent.skill_utils import get_external_skills_dirs, iter_skill_index_files
         disabled = _get_disabled_skill_names()
         seen_names: set = set()
 
@@ -356,8 +260,8 @@ def scan_skill_commands() -> Dict[str, Dict[str, Any]]:
         dirs_to_scan.extend(get_external_skills_dirs())
 
         for scan_dir in dirs_to_scan:
-            for skill_md in scan_dir.rglob("SKILL.md"):
-                if any(part in ('.git', '.github', '.hub') for part in skill_md.parts):
+            for skill_md in iter_skill_index_files(scan_dir, "SKILL.md"):
+                if any(part in ('.git', '.github', '.hub', '.archive') for part in skill_md.parts):
                     continue
                 try:
                     content = skill_md.read_text(encoding='utf-8')
@@ -401,12 +305,85 @@ def scan_skill_commands() -> Dict[str, Dict[str, Any]]:
 
 
 def get_skill_commands() -> Dict[str, Dict[str, Any]]:
-    """Return the current skill commands mapping (scan first if empty)."""
-    if not _skill_commands:
+    """Return the current skill commands mapping (scan first if empty).
+
+    Rescans when the active platform scope changes (e.g. a gateway
+    process serving Telegram and Discord concurrently) so each platform
+    sees its own ``skills.platform_disabled`` view (#14536).
+    """
+    if (
+        not _skill_commands
+        or _skill_commands_platform != _resolve_skill_commands_platform()
+    ):
         scan_skill_commands()
     return _skill_commands
 
 
+def reload_skills() -> Dict[str, Any]:
+    """Re-scan the skills directory and return a diff of what changed.
+
+    Rescans ``~/.hermes/skills/`` and any ``skills.external_dirs`` so the
+    slash-command map (``agent.skill_commands._skill_commands``) reflects
+    skills added or removed on disk.
+
+    This does NOT invalidate the skills system-prompt cache. Skills are
+    called by name via ``/skill-name``, ``skills_list``, or ``skill_view``
+    — they don't need to be in the system prompt for the model to use them.
+    Keeping the prompt cache intact preserves prefix caching across the
+    reload, so a user invoking ``/reload-skills`` pays no cache-reset cost.
+
+    Returns:
+        Dict with keys::
+
+            {
+              "added":      [{"name": str, "description": str}, ...],
+              "removed":    [{"name": str, "description": str}, ...],
+              "unchanged":  [skill names present before and after],
+              "total":      total skill count after rescan,
+              "commands":   total /slash-skill count after rescan,
+            }
+
+        ``description`` is the skill's full SKILL.md frontmatter
+        ``description:`` field — the same string the system prompt renders
+        as ``    - name: description`` for pre-existing skills.
+    """
+    # Snapshot pre-reload state (name -> description) from the current
+    # slash-command cache. Using dicts lets the post-rescan diff carry
+    # descriptions for newly-visible or just-removed skills without a
+    # second disk walk.
+    def _snapshot(cmds: Dict[str, Dict[str, Any]]) -> Dict[str, str]:
+        out: Dict[str, str] = {}
+        for slash_key, info in cmds.items():
+            bare = slash_key.lstrip("/")
+            out[bare] = (info or {}).get("description") or ""
+        return out
+
+    before = _snapshot(_skill_commands)
+
+    # Rescan the skills dir. ``scan_skill_commands`` resets
+    # ``_skill_commands = {}`` internally and repopulates it.
+    new_commands = scan_skill_commands()
+
+    after = _snapshot(new_commands)
+
+    added_names = sorted(set(after) - set(before))
+    removed_names = sorted(set(before) - set(after))
+    unchanged = sorted(set(after) & set(before))
+
+    added = [{"name": n, "description": after[n]} for n in added_names]
+    # For removed skills, use the description we had cached pre-rescan
+    # (the skill file is gone so we can't re-read it).
+    removed = [{"name": n, "description": before[n]} for n in removed_names]
+
+    return {
+        "added": added,
+        "removed": removed,
+        "unchanged": unchanged,
+        "total": len(after),
+        "commands": len(new_commands),
+    }
+
+
 def resolve_skill_command_key(command: str) -> Optional[str]:
     """Resolve a user-typed /command to its canonical skill_cmds key.
 
@@ -451,8 +428,16 @@ def build_skill_invocation_message(
         return f"[Failed to load skill: {skill_info['name']}]"
 
     loaded_skill, skill_dir, skill_name = loaded
+
+    # Track active usage for Curator lifecycle management (#17782)
+    try:
+        from tools.skill_usage import bump_use
+        bump_use(skill_name)
+    except Exception:
+        pass  # Non-critical — skill invocation proceeds regardless
+
     activation_note = (
-        f'[SYSTEM: The user has invoked the "{skill_name}" skill, indicating they want '
+        f'[IMPORTANT: The user has invoked the "{skill_name}" skill, indicating they want '
         "you to follow its instructions. The full skill content is loaded below.]"
     )
     return _build_skill_message(
@@ -490,8 +475,16 @@ def build_preloaded_skills_prompt(
             continue
 
         loaded_skill, skill_dir, skill_name = loaded
+
+        # Track active usage for Curator lifecycle management (#17782)
+        try:
+            from tools.skill_usage import bump_use
+            bump_use(skill_name)
+        except Exception:
+            pass  # Non-critical
+
         activation_note = (
-            f'[SYSTEM: The user launched this CLI session with the "{skill_name}" skill '
+            f'[IMPORTANT: The user launched this CLI session with the "{skill_name}" skill '
             "preloaded. Treat its instructions as active guidance for the duration of this "
             "session unless the user overrides them.]"
         )

agent/skill_preprocessing.py

@@ -0,0 +1,131 @@
+"""Shared SKILL.md preprocessing helpers."""
+
+import logging
+import re
+import subprocess
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+# Matches ${HERMES_SKILL_DIR} / ${HERMES_SESSION_ID} tokens in SKILL.md.
+# Tokens that don't resolve (e.g. ${HERMES_SESSION_ID} with no session) are
+# left as-is so the user can debug them.
+_SKILL_TEMPLATE_RE = re.compile(r"\$\{(HERMES_SKILL_DIR|HERMES_SESSION_ID)\}")
+
+# Matches inline shell snippets like:  !`date +%Y-%m-%d`
+# Non-greedy, single-line only -- no newlines inside the backticks.
+_INLINE_SHELL_RE = re.compile(r"!`([^`\n]+)`")
+
+# Cap inline-shell output so a runaway command can't blow out the context.
+_INLINE_SHELL_MAX_OUTPUT = 4000
+
+
+def load_skills_config() -> dict:
+    """Load the ``skills`` section of config.yaml (best-effort)."""
+    try:
+        from hermes_cli.config import load_config
+
+        cfg = load_config() or {}
+        skills_cfg = cfg.get("skills")
+        if isinstance(skills_cfg, dict):
+            return skills_cfg
+    except Exception:
+        logger.debug("Could not read skills config", exc_info=True)
+    return {}
+
+
+def substitute_template_vars(
+    content: str,
+    skill_dir: Path | None,
+    session_id: str | None,
+) -> str:
+    """Replace ${HERMES_SKILL_DIR} / ${HERMES_SESSION_ID} in skill content.
+
+    Only substitutes tokens for which a concrete value is available --
+    unresolved tokens are left in place so the author can spot them.
+    """
+    if not content:
+        return content
+
+    skill_dir_str = str(skill_dir) if skill_dir else None
+
+    def _replace(match: re.Match) -> str:
+        token = match.group(1)
+        if token == "HERMES_SKILL_DIR" and skill_dir_str:
+            return skill_dir_str
+        if token == "HERMES_SESSION_ID" and session_id:
+            return str(session_id)
+        return match.group(0)
+
+    return _SKILL_TEMPLATE_RE.sub(_replace, content)
+
+
+def run_inline_shell(command: str, cwd: Path | None, timeout: int) -> str:
+    """Execute a single inline-shell snippet and return its stdout (trimmed).
+
+    Failures return a short ``[inline-shell error: ...]`` marker instead of
+    raising, so one bad snippet can't wreck the whole skill message.
+    """
+    try:
+        completed = subprocess.run(
+            ["bash", "-c", command],
+            cwd=str(cwd) if cwd else None,
+            capture_output=True,
+            text=True,
+            timeout=max(1, int(timeout)),
+            check=False,
+        )
+    except subprocess.TimeoutExpired:
+        return f"[inline-shell timeout after {timeout}s: {command}]"
+    except FileNotFoundError:
+        return "[inline-shell error: bash not found]"
+    except Exception as exc:
+        return f"[inline-shell error: {exc}]"
+
+    output = (completed.stdout or "").rstrip("\n")
+    if not output and completed.stderr:
+        output = completed.stderr.rstrip("\n")
+    if len(output) > _INLINE_SHELL_MAX_OUTPUT:
+        output = output[:_INLINE_SHELL_MAX_OUTPUT] + "...[truncated]"
+    return output
+
+
+def expand_inline_shell(
+    content: str,
+    skill_dir: Path | None,
+    timeout: int,
+) -> str:
+    """Replace every !`cmd` snippet in ``content`` with its stdout.
+
+    Runs each snippet with the skill directory as CWD so relative paths in
+    the snippet work the way the author expects.
+    """
+    if "!`" not in content:
+        return content
+
+    def _replace(match: re.Match) -> str:
+        cmd = match.group(1).strip()
+        if not cmd:
+            return ""
+        return run_inline_shell(cmd, skill_dir, timeout)
+
+    return _INLINE_SHELL_RE.sub(_replace, content)
+
+
+def preprocess_skill_content(
+    content: str,
+    skill_dir: Path | None,
+    session_id: str | None = None,
+    skills_cfg: dict | None = None,
+) -> str:
+    """Apply configured SKILL.md template and inline-shell preprocessing."""
+    if not content:
+        return content
+
+    cfg = skills_cfg if isinstance(skills_cfg, dict) else load_skills_config()
+    if cfg.get("template_vars", True):
+        content = substitute_template_vars(content, skill_dir, session_id)
+    if cfg.get("inline_shell", False):
+        timeout = int(cfg.get("inline_shell_timeout", 10) or 10)
+        content = expand_inline_shell(content, skill_dir, timeout)
+    return content

agent/skill_utils.py

@@ -24,7 +24,7 @@ PLATFORM_MAP = {
     "windows": "win32",
 }
 
-EXCLUDED_SKILL_DIRS = frozenset((".git", ".github", ".hub"))
+EXCLUDED_SKILL_DIRS = frozenset((".git", ".github", ".hub", ".archive"))
 
 # ── Lazy YAML loader ─────────────────────────────────────────────────────
 
@@ -200,6 +200,9 @@ def get_external_skills_dirs() -> List[Path]:
     if not isinstance(raw_dirs, list):
         return []
 
+    from hermes_constants import get_hermes_home
+
+    hermes_home = get_hermes_home()
     local_skills = get_skills_dir().resolve()
     seen: Set[Path] = set()
     result: List[Path] = []
@@ -210,7 +213,12 @@ def get_external_skills_dirs() -> List[Path]:
             continue
         # Expand ~ and environment variables
         expanded = os.path.expanduser(os.path.expandvars(entry))
-        p = Path(expanded).resolve()
+        p = Path(expanded)
+        # Resolve relative paths against HERMES_HOME, not cwd
+        if not p.is_absolute():
+            p = (hermes_home / p).resolve()
+        else:
+            p = p.resolve()
         if p == local_skills:
             continue
         if p in seen:
@@ -432,7 +440,7 @@ def extract_skill_description(frontmatter: Dict[str, Any]) -> str:
 def iter_skill_index_files(skills_dir: Path, filename: str):
     """Walk skills_dir yielding sorted paths matching *filename*.
 
-    Excludes ``.git``, ``.github``, ``.hub`` directories.
+    Excludes ``.git``, ``.github``, ``.hub``, ``.archive`` directories.
     """
     matches = []
     for root, dirs, files in os.walk(skills_dir, followlinks=True):

agent/think_scrubber.py

@@ -0,0 +1,386 @@
+"""Stateful scrubber for reasoning/thinking blocks in streamed assistant text.
+
+``run_agent._strip_think_blocks`` is regex-based and correct for a complete
+string, but when it runs *per-delta* in ``_fire_stream_delta`` it destroys
+the state that downstream consumers (CLI ``_stream_delta``, gateway
+``GatewayStreamConsumer._filter_and_accumulate``) rely on.
+
+Concretely, when MiniMax-M2.7 streams
+
+    delta1 = "<think>"
+    delta2 = "Let me check their config"
+    delta3 = "</think>"
+
+the per-delta regex erases delta1 entirely (case 2: unterminated-open at
+boundary matches ``^<think>...``), so the downstream state machine never
+sees the open tag, treats delta2 as regular content, and leaks reasoning
+to the user.  Consumers that don't run their own state machine (ACP,
+api_server, TTS) never had any defence at all — they just emitted
+whatever survived the upstream regex.
+
+This module centralises the tag-suppression state machine at the
+upstream layer so every stream_delta_callback sees text that has
+already had reasoning blocks removed.  Partial tags at delta
+boundaries are held back until the next delta resolves them, and
+end-of-stream flushing surfaces any held-back prose that turned out
+not to be a real tag.
+
+Usage::
+
+    scrubber = StreamingThinkScrubber()
+    for delta in stream:
+        visible = scrubber.feed(delta)
+        if visible:
+            emit(visible)
+    tail = scrubber.flush()  # at end of stream
+    if tail:
+        emit(tail)
+
+The scrubber is re-entrant per agent instance.  Call ``reset()`` at
+the top of each new turn so a hung block from an interrupted prior
+stream cannot taint the next turn's output.
+
+Tag variants handled (case-insensitive):
+  ``<think>``, ``<thinking>``, ``<reasoning>``, ``<thought>``,
+  ``<REASONING_SCRATCHPAD>``.
+
+Block-boundary rule for opens: an opening tag is only treated as a
+reasoning-block opener when it appears at the start of the stream,
+after a newline (optionally followed by whitespace), or when only
+whitespace has been emitted on the current line.  This prevents prose
+that *mentions* the tag name (e.g. ``"use <think> tags here"``) from
+being incorrectly suppressed.  Closed pairs (``<think>X</think>``) are
+always suppressed regardless of boundary; a closed pair is an
+intentional, bounded construct.
+"""
+
+from __future__ import annotations
+
+from typing import Tuple
+
+__all__ = ["StreamingThinkScrubber"]
+
+
+class StreamingThinkScrubber:
+    """Stateful scrubber for streaming reasoning/thinking blocks.
+
+    State machine:
+      - ``_in_block``: True while inside an opened block, waiting for
+        a close tag.  All text inside is discarded.
+      - ``_buf``: held-back partial-tag tail.  Emitted / discarded on
+        the next ``feed()`` call or by ``flush()``.
+      - ``_last_emitted_ended_newline``: True iff the most recent
+        emission to the consumer ended with ``\\n``, or nothing has
+        been emitted yet (start-of-stream counts as a boundary).  Used
+        to decide whether an open tag at buffer position 0 is at a
+        block boundary.
+    """
+
+    _OPEN_TAG_NAMES: Tuple[str, ...] = (
+        "think",
+        "thinking",
+        "reasoning",
+        "thought",
+        "REASONING_SCRATCHPAD",
+    )
+
+    # Materialise literal tag strings so the hot path does string
+    # operations, not regex compilation per feed().
+    _OPEN_TAGS: Tuple[str, ...] = tuple(f"<{name}>" for name in _OPEN_TAG_NAMES)
+    _CLOSE_TAGS: Tuple[str, ...] = tuple(f"</{name}>" for name in _OPEN_TAG_NAMES)
+
+    # Pre-compute the longest tag (for partial-tag hold-back bound).
+    _MAX_TAG_LEN: int = max(len(tag) for tag in _OPEN_TAGS + _CLOSE_TAGS)
+
+    def __init__(self) -> None:
+        self._in_block: bool = False
+        self._buf: str = ""
+        self._last_emitted_ended_newline: bool = True
+
+    def reset(self) -> None:
+        """Reset all state.  Call at the top of every new turn."""
+        self._in_block = False
+        self._buf = ""
+        self._last_emitted_ended_newline = True
+
+    def feed(self, text: str) -> str:
+        """Feed one delta; return the scrubbed visible portion.
+
+        May return an empty string when the entire delta is reasoning
+        content or is being held back pending resolution of a partial
+        tag at the boundary.
+        """
+        if not text:
+            return ""
+        buf = self._buf + text
+        self._buf = ""
+        out: list[str] = []
+
+        while buf:
+            if self._in_block:
+                # Hunt for the earliest close tag.
+                close_idx, close_len = self._find_first_tag(
+                    buf, self._CLOSE_TAGS,
+                )
+                if close_idx == -1:
+                    # No close yet — hold back a potential partial
+                    # close-tag prefix; discard everything else.
+                    held = self._max_partial_suffix(buf, self._CLOSE_TAGS)
+                    self._buf = buf[-held:] if held else ""
+                    return "".join(out)
+                # Found close: discard block content + tag, continue.
+                buf = buf[close_idx + close_len:]
+                self._in_block = False
+            else:
+                # Priority 1 — closed <tag>X</tag> pair anywhere in
+                # buf.  Closed pairs are always an intentional,
+                # bounded construct (even mid-line prose containing
+                # an open/close pair is almost certainly a model
+                # leaking reasoning inline), so no boundary gating.
+                pair = self._find_earliest_closed_pair(buf)
+                # Priority 2 — unterminated open tag at a block
+                # boundary.  Boundary-gated so prose that mentions
+                # '<think>' isn't over-stripped.
+                open_idx, open_len = self._find_open_at_boundary(
+                    buf, out,
+                )
+
+                # Pick whichever match comes earliest in the buffer.
+                if pair is not None and (
+                    open_idx == -1 or pair[0] <= open_idx
+                ):
+                    start_idx, end_idx = pair
+                    preceding = buf[:start_idx]
+                    if preceding:
+                        preceding = self._strip_orphan_close_tags(preceding)
+                        if preceding:
+                            out.append(preceding)
+                            self._last_emitted_ended_newline = (
+                                preceding.endswith("\n")
+                            )
+                    buf = buf[end_idx:]
+                    continue
+
+                if open_idx != -1:
+                    # Unterminated open at boundary — emit preceding,
+                    # enter block, continue loop with remainder.
+                    preceding = buf[:open_idx]
+                    if preceding:
+                        preceding = self._strip_orphan_close_tags(preceding)
+                        if preceding:
+                            out.append(preceding)
+                            self._last_emitted_ended_newline = (
+                                preceding.endswith("\n")
+                            )
+                    self._in_block = True
+                    buf = buf[open_idx + open_len:]
+                    continue
+
+                # No resolvable tag structure in buf.  Hold back any
+                # partial-tag prefix at the tail so a split tag
+                # across deltas isn't missed, then emit the rest.
+                held = self._max_partial_suffix(buf, self._OPEN_TAGS)
+                held_close = self._max_partial_suffix(
+                    buf, self._CLOSE_TAGS,
+                )
+                held = max(held, held_close)
+                if held:
+                    emit_text = buf[:-held]
+                    self._buf = buf[-held:]
+                else:
+                    emit_text = buf
+                    self._buf = ""
+                if emit_text:
+                    emit_text = self._strip_orphan_close_tags(emit_text)
+                    if emit_text:
+                        out.append(emit_text)
+                        self._last_emitted_ended_newline = (
+                            emit_text.endswith("\n")
+                        )
+                return "".join(out)
+
+        return "".join(out)
+
+    def flush(self) -> str:
+        """End-of-stream flush.
+
+        If still inside an unterminated block, held-back content is
+        discarded — leaking partial reasoning is worse than a
+        truncated answer.  Otherwise the held-back partial-tag tail is
+        emitted verbatim (it turned out not to be a real tag prefix).
+        """
+        if self._in_block:
+            self._buf = ""
+            self._in_block = False
+            return ""
+        tail = self._buf
+        self._buf = ""
+        if not tail:
+            return ""
+        tail = self._strip_orphan_close_tags(tail)
+        if tail:
+            self._last_emitted_ended_newline = tail.endswith("\n")
+        return tail
+
+    # ── internal helpers ───────────────────────────────────────────────
+
+    @staticmethod
+    def _find_first_tag(
+        buf: str, tags: Tuple[str, ...],
+    ) -> Tuple[int, int]:
+        """Return (earliest_index, tag_length) over *tags*, or (-1, 0).
+
+        Case-insensitive match.
+        """
+        buf_lower = buf.lower()
+        best_idx = -1
+        best_len = 0
+        for tag in tags:
+            idx = buf_lower.find(tag.lower())
+            if idx != -1 and (best_idx == -1 or idx < best_idx):
+                best_idx = idx
+                best_len = len(tag)
+        return best_idx, best_len
+
+    def _find_earliest_closed_pair(self, buf: str):
+        """Return (start_idx, end_idx) of the earliest closed pair, else None.
+
+        A closed pair is ``<tag>...</tag>`` of any variant.  Matches are
+        case-insensitive and non-greedy (the closest close tag after
+        an open tag wins), matching the regex ``<tag>.*?</tag>``
+        semantics of ``_strip_think_blocks`` case 1.  When two tag
+        variants could both match, the one whose open tag appears
+        earlier wins.
+        """
+        buf_lower = buf.lower()
+        best: "tuple[int, int] | None" = None
+        for open_tag, close_tag in zip(self._OPEN_TAGS, self._CLOSE_TAGS):
+            open_lower = open_tag.lower()
+            close_lower = close_tag.lower()
+            open_idx = buf_lower.find(open_lower)
+            if open_idx == -1:
+                continue
+            close_idx = buf_lower.find(
+                close_lower, open_idx + len(open_lower),
+            )
+            if close_idx == -1:
+                continue
+            end_idx = close_idx + len(close_lower)
+            if best is None or open_idx < best[0]:
+                best = (open_idx, end_idx)
+        return best
+
+    def _find_open_at_boundary(
+        self, buf: str, already_emitted: list[str],
+    ) -> Tuple[int, int]:
+        """Return the earliest block-boundary open-tag (idx, len).
+
+        Returns (-1, 0) if no boundary-legal opener is present.
+        """
+        buf_lower = buf.lower()
+        best_idx = -1
+        best_len = 0
+        for tag in self._OPEN_TAGS:
+            tag_lower = tag.lower()
+            search_start = 0
+            while True:
+                idx = buf_lower.find(tag_lower, search_start)
+                if idx == -1:
+                    break
+                if self._is_block_boundary(buf, idx, already_emitted):
+                    if best_idx == -1 or idx < best_idx:
+                        best_idx = idx
+                        best_len = len(tag)
+                    break  # first boundary hit for this tag is enough
+                search_start = idx + 1
+        return best_idx, best_len
+
+    def _is_block_boundary(
+        self, buf: str, idx: int, already_emitted: list[str],
+    ) -> bool:
+        """True iff position *idx* in *buf* is a block boundary.
+
+        A block boundary is:
+          - buf position 0 AND the most recent emission ended with
+            a newline (or nothing has been emitted yet)
+          - any position whose preceding text on the current line
+            (since the last newline in buf) is whitespace-only, AND
+            if there is no newline in the preceding buf portion, the
+            most recent prior emission ended with a newline
+        """
+        if idx == 0:
+            # Check whether the last already-emitted chunk in THIS
+            # feed() call ended with a newline, otherwise fall back
+            # to the cross-feed flag.
+            if already_emitted:
+                return already_emitted[-1].endswith("\n")
+            return self._last_emitted_ended_newline
+        preceding = buf[:idx]
+        last_nl = preceding.rfind("\n")
+        if last_nl == -1:
+            # No newline in buf before the tag — boundary only if the
+            # prior emission ended with a newline AND everything since
+            # is whitespace.
+            if already_emitted:
+                prior_newline = already_emitted[-1].endswith("\n")
+            else:
+                prior_newline = self._last_emitted_ended_newline
+            return prior_newline and preceding.strip() == ""
+        # Newline present — text between it and the tag must be
+        # whitespace-only.
+        return preceding[last_nl + 1:].strip() == ""
+
+    @classmethod
+    def _max_partial_suffix(
+        cls, buf: str, tags: Tuple[str, ...],
+    ) -> int:
+        """Return the longest buf-suffix that is a prefix of any tag.
+
+        Only prefixes strictly shorter than the tag itself count
+        (full-length suffixes are the tag and are handled as matches,
+        not held-back partials).  Case-insensitive.
+        """
+        if not buf:
+            return 0
+        buf_lower = buf.lower()
+        max_check = min(len(buf_lower), cls._MAX_TAG_LEN - 1)
+        for i in range(max_check, 0, -1):
+            suffix = buf_lower[-i:]
+            for tag in tags:
+                tag_lower = tag.lower()
+                if len(tag_lower) > i and tag_lower.startswith(suffix):
+                    return i
+        return 0
+
+    @classmethod
+    def _strip_orphan_close_tags(cls, text: str) -> str:
+        """Remove any close tags from *text* (orphan-close handling).
+
+        An orphan close tag has no matching open in the current
+        scrubber state; it's always noise, stripped with any trailing
+        whitespace so the surrounding prose flows naturally.
+        """
+        if "</" not in text:
+            return text
+        text_lower = text.lower()
+        out: list[str] = []
+        i = 0
+        while i < len(text):
+            matched = False
+            if text_lower[i:i + 2] == "</":
+                for tag in cls._CLOSE_TAGS:
+                    tag_lower = tag.lower()
+                    tag_len = len(tag_lower)
+                    if text_lower[i:i + tag_len] == tag_lower:
+                        # Skip the tag and any trailing whitespace,
+                        # matching _strip_think_blocks case 3.
+                        j = i + tag_len
+                        while j < len(text) and text[j] in " \t\n\r":
+                            j += 1
+                        i = j
+                        matched = True
+                        break
+            if not matched:
+                out.append(text[i])
+                i += 1
+        return "".join(out)

agent/title_generator.py

@@ -6,12 +6,19 @@ adds latency to the user-facing reply.
 
 import logging
 import threading
-from typing import Optional
+from typing import Callable, Optional
 
 from agent.auxiliary_client import call_llm
 
 logger = logging.getLogger(__name__)
 
+# Callback signature: (task_name, exception) -> None. Used to surface
+# auxiliary failures to the user through AIAgent._emit_auxiliary_failure
+# so silent-drops (e.g. OpenRouter 402 exhausting the fallback chain)
+# become visible instead of piling up as NULL session titles.
+FailureCallback = Callable[[str, BaseException], None]
+TitleCallback = Callable[[str], None]
+
 _TITLE_PROMPT = (
     "Generate a short, descriptive title (3-7 words) for a conversation that starts with the "
     "following exchange. The title should capture the main topic or intent. "
@@ -19,11 +26,23 @@ _TITLE_PROMPT = (
 )
 
 
-def generate_title(user_message: str, assistant_response: str, timeout: float = 30.0) -> Optional[str]:
+def generate_title(
+    user_message: str,
+    assistant_response: str,
+    timeout: float = 30.0,
+    failure_callback: Optional[FailureCallback] = None,
+    main_runtime: dict = None,
+) -> Optional[str]:
     """Generate a session title from the first exchange.
 
-    Uses the auxiliary LLM client (cheapest/fastest available model).
+    Uses the main runtime's model when available, falling back to the
+    auxiliary LLM client (cheapest/fastest available model).
     Returns the title string or None on failure.
+
+    ``failure_callback`` is invoked with ``(task, exception)`` when the
+    auxiliary call raises — the caller typically wires this to
+    ``AIAgent._emit_auxiliary_failure`` so the user sees a warning instead
+    of silently accumulating untitled sessions.
     """
     # Truncate long messages to keep the request small
     user_snippet = user_message[:500] if user_message else ""
@@ -41,6 +60,7 @@ def generate_title(user_message: str, assistant_response: str, timeout: float =
             max_tokens=500,
             temperature=0.3,
             timeout=timeout,
+            main_runtime=main_runtime,
         )
         title = (response.choices[0].message.content or "").strip()
         # Clean up: remove quotes, trailing punctuation, prefixes like "Title: "
@@ -52,7 +72,15 @@ def generate_title(user_message: str, assistant_response: str, timeout: float =
             title = title[:77] + "..."
         return title if title else None
     except Exception as e:
-        logger.debug("Title generation failed: %s", e)
+        # Log at WARNING so this shows up in agent.log without debug mode.
+        # Full detail at debug level for operators who need the stack.
+        logger.warning("Title generation failed: %s", e)
+        logger.debug("Title generation traceback", exc_info=True)
+        if failure_callback is not None:
+            try:
+                failure_callback("title generation", e)
+            except Exception:
+                logger.debug("Title generation failure_callback raised", exc_info=True)
         return None
 
 
@@ -61,6 +89,9 @@ def auto_title_session(
     session_id: str,
     user_message: str,
     assistant_response: str,
+    failure_callback: Optional[FailureCallback] = None,
+    main_runtime: dict = None,
+    title_callback: Optional[TitleCallback] = None,
 ) -> None:
     """Generate and set a session title if one doesn't already exist.
 
@@ -81,13 +112,20 @@ def auto_title_session(
     except Exception:
         return
 
-    title = generate_title(user_message, assistant_response)
+    title = generate_title(
+        user_message, assistant_response, failure_callback=failure_callback, main_runtime=main_runtime
+    )
     if not title:
         return
 
     try:
         session_db.set_session_title(session_id, title)
         logger.debug("Auto-generated session title: %s", title)
+        if title_callback is not None:
+            try:
+                title_callback(title)
+            except Exception:
+                logger.debug("Auto-title callback failed", exc_info=True)
     except Exception as e:
         logger.debug("Failed to set auto-generated title: %s", e)
 
@@ -98,6 +136,9 @@ def maybe_auto_title(
     user_message: str,
     assistant_response: str,
     conversation_history: list,
+    failure_callback: Optional[FailureCallback] = None,
+    main_runtime: dict = None,
+    title_callback: Optional[TitleCallback] = None,
 ) -> None:
     """Fire-and-forget title generation after the first exchange.
 
@@ -119,6 +160,11 @@ def maybe_auto_title(
     thread = threading.Thread(
         target=auto_title_session,
         args=(session_db, session_id, user_message, assistant_response),
+        kwargs={
+            "failure_callback": failure_callback,
+            "main_runtime": main_runtime,
+            "title_callback": title_callback,
+        },
         daemon=True,
         name="auto-title",
     )

agent/tool_guardrails.py

@@ -0,0 +1,455 @@
+"""Pure tool-call loop guardrail primitives.
+
+The controller in this module is intentionally side-effect free: it tracks
+per-turn tool-call observations and returns decisions. Runtime code owns whether
+those decisions become warning guidance, synthetic tool results, or controlled
+turn halts.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import json
+from dataclasses import dataclass, field
+from typing import Any, Mapping
+
+from utils import safe_json_loads
+
+
+IDEMPOTENT_TOOL_NAMES = frozenset(
+    {
+        "read_file",
+        "search_files",
+        "web_search",
+        "web_extract",
+        "session_search",
+        "browser_snapshot",
+        "browser_console",
+        "browser_get_images",
+        "mcp_filesystem_read_file",
+        "mcp_filesystem_read_text_file",
+        "mcp_filesystem_read_multiple_files",
+        "mcp_filesystem_list_directory",
+        "mcp_filesystem_list_directory_with_sizes",
+        "mcp_filesystem_directory_tree",
+        "mcp_filesystem_get_file_info",
+        "mcp_filesystem_search_files",
+    }
+)
+
+MUTATING_TOOL_NAMES = frozenset(
+    {
+        "terminal",
+        "execute_code",
+        "write_file",
+        "patch",
+        "todo",
+        "memory",
+        "skill_manage",
+        "browser_click",
+        "browser_type",
+        "browser_press",
+        "browser_scroll",
+        "browser_navigate",
+        "send_message",
+        "cronjob",
+        "delegate_task",
+        "process",
+    }
+)
+
+
+@dataclass(frozen=True)
+class ToolCallGuardrailConfig:
+    """Thresholds for per-turn tool-call loop detection.
+
+    Warnings are enabled by default and never prevent tool execution. Hard stops
+    are explicit opt-in so interactive CLI/TUI sessions get a gentle nudge unless
+    the user enables circuit-breaker behavior in config.yaml.
+    """
+
+    warnings_enabled: bool = True
+    hard_stop_enabled: bool = False
+    exact_failure_warn_after: int = 2
+    exact_failure_block_after: int = 5
+    same_tool_failure_warn_after: int = 3
+    same_tool_failure_halt_after: int = 8
+    no_progress_warn_after: int = 2
+    no_progress_block_after: int = 5
+    idempotent_tools: frozenset[str] = field(default_factory=lambda: IDEMPOTENT_TOOL_NAMES)
+    mutating_tools: frozenset[str] = field(default_factory=lambda: MUTATING_TOOL_NAMES)
+
+    @classmethod
+    def from_mapping(cls, data: Mapping[str, Any] | None) -> "ToolCallGuardrailConfig":
+        """Build config from the `tool_loop_guardrails` config.yaml section."""
+        if not isinstance(data, Mapping):
+            return cls()
+
+        warn_after = data.get("warn_after")
+        if not isinstance(warn_after, Mapping):
+            warn_after = {}
+        hard_stop_after = data.get("hard_stop_after")
+        if not isinstance(hard_stop_after, Mapping):
+            hard_stop_after = {}
+
+        defaults = cls()
+        return cls(
+            warnings_enabled=_as_bool(data.get("warnings_enabled"), defaults.warnings_enabled),
+            hard_stop_enabled=_as_bool(data.get("hard_stop_enabled"), defaults.hard_stop_enabled),
+            exact_failure_warn_after=_positive_int(
+                warn_after.get("exact_failure", data.get("exact_failure_warn_after")),
+                defaults.exact_failure_warn_after,
+            ),
+            same_tool_failure_warn_after=_positive_int(
+                warn_after.get("same_tool_failure", data.get("same_tool_failure_warn_after")),
+                defaults.same_tool_failure_warn_after,
+            ),
+            no_progress_warn_after=_positive_int(
+                warn_after.get("idempotent_no_progress", data.get("no_progress_warn_after")),
+                defaults.no_progress_warn_after,
+            ),
+            exact_failure_block_after=_positive_int(
+                hard_stop_after.get("exact_failure", data.get("exact_failure_block_after")),
+                defaults.exact_failure_block_after,
+            ),
+            same_tool_failure_halt_after=_positive_int(
+                hard_stop_after.get("same_tool_failure", data.get("same_tool_failure_halt_after")),
+                defaults.same_tool_failure_halt_after,
+            ),
+            no_progress_block_after=_positive_int(
+                hard_stop_after.get("idempotent_no_progress", data.get("no_progress_block_after")),
+                defaults.no_progress_block_after,
+            ),
+        )
+
+
+@dataclass(frozen=True)
+class ToolCallSignature:
+    """Stable, non-reversible identity for a tool name plus canonical args."""
+
+    tool_name: str
+    args_hash: str
+
+    @classmethod
+    def from_call(cls, tool_name: str, args: Mapping[str, Any] | None) -> "ToolCallSignature":
+        canonical = canonical_tool_args(args or {})
+        return cls(tool_name=tool_name, args_hash=_sha256(canonical))
+
+    def to_metadata(self) -> dict[str, str]:
+        """Return public metadata without raw argument values."""
+        return {"tool_name": self.tool_name, "args_hash": self.args_hash}
+
+
+@dataclass(frozen=True)
+class ToolGuardrailDecision:
+    """Decision returned by the tool-call guardrail controller."""
+
+    action: str = "allow"  # allow | warn | block | halt
+    code: str = "allow"
+    message: str = ""
+    tool_name: str = ""
+    count: int = 0
+    signature: ToolCallSignature | None = None
+
+    @property
+    def allows_execution(self) -> bool:
+        return self.action in {"allow", "warn"}
+
+    @property
+    def should_halt(self) -> bool:
+        return self.action in {"block", "halt"}
+
+    def to_metadata(self) -> dict[str, Any]:
+        data: dict[str, Any] = {
+            "action": self.action,
+            "code": self.code,
+            "message": self.message,
+            "tool_name": self.tool_name,
+            "count": self.count,
+        }
+        if self.signature is not None:
+            data["signature"] = self.signature.to_metadata()
+        return data
+
+
+def canonical_tool_args(args: Mapping[str, Any]) -> str:
+    """Return sorted compact JSON for parsed tool arguments."""
+    if not isinstance(args, Mapping):
+        raise TypeError(f"tool args must be a mapping, got {type(args).__name__}")
+    return json.dumps(
+        args,
+        ensure_ascii=False,
+        sort_keys=True,
+        separators=(",", ":"),
+        default=str,
+    )
+
+
+def classify_tool_failure(tool_name: str, result: str | None) -> tuple[bool, str]:
+    """Safety-fallback classifier used only when callers don't pass ``failed``.
+
+    Mirrors ``agent.display._detect_tool_failure`` exactly so the guardrail
+    never disagrees with the CLI's user-visible ``[error]`` tag. Production
+    callers in ``run_agent.py`` always pass an explicit ``failed=`` derived
+    from ``_detect_tool_failure``; this function exists so standalone callers
+    (tests, tooling) still get consistent behavior.
+    """
+    if result is None:
+        return False, ""
+
+    if tool_name == "terminal":
+        data = safe_json_loads(result)
+        if isinstance(data, dict):
+            exit_code = data.get("exit_code")
+            if exit_code is not None and exit_code != 0:
+                return True, f" [exit {exit_code}]"
+        return False, ""
+
+    if tool_name == "memory":
+        data = safe_json_loads(result)
+        if isinstance(data, dict):
+            if data.get("success") is False and "exceed the limit" in data.get("error", ""):
+                return True, " [full]"
+
+    lower = result[:500].lower()
+    if '"error"' in lower or '"failed"' in lower or result.startswith("Error"):
+        return True, " [error]"
+
+    return False, ""
+
+
+class ToolCallGuardrailController:
+    """Per-turn controller for repeated failed/non-progressing tool calls."""
+
+    def __init__(self, config: ToolCallGuardrailConfig | None = None):
+        self.config = config or ToolCallGuardrailConfig()
+        self.reset_for_turn()
+
+    def reset_for_turn(self) -> None:
+        self._exact_failure_counts: dict[ToolCallSignature, int] = {}
+        self._same_tool_failure_counts: dict[str, int] = {}
+        self._no_progress: dict[ToolCallSignature, tuple[str, int]] = {}
+        self._halt_decision: ToolGuardrailDecision | None = None
+
+    @property
+    def halt_decision(self) -> ToolGuardrailDecision | None:
+        return self._halt_decision
+
+    def before_call(self, tool_name: str, args: Mapping[str, Any] | None) -> ToolGuardrailDecision:
+        signature = ToolCallSignature.from_call(tool_name, _coerce_args(args))
+        if not self.config.hard_stop_enabled:
+            return ToolGuardrailDecision(tool_name=tool_name, signature=signature)
+
+        exact_count = self._exact_failure_counts.get(signature, 0)
+        if exact_count >= self.config.exact_failure_block_after:
+            decision = ToolGuardrailDecision(
+                action="block",
+                code="repeated_exact_failure_block",
+                message=(
+                    f"Blocked {tool_name}: the same tool call failed {exact_count} "
+                    "times with identical arguments. Stop retrying it unchanged; "
+                    "change strategy or explain the blocker."
+                ),
+                tool_name=tool_name,
+                count=exact_count,
+                signature=signature,
+            )
+            self._halt_decision = decision
+            return decision
+
+        if self._is_idempotent(tool_name):
+            record = self._no_progress.get(signature)
+            if record is not None:
+                _result_hash, repeat_count = record
+                if repeat_count >= self.config.no_progress_block_after:
+                    decision = ToolGuardrailDecision(
+                        action="block",
+                        code="idempotent_no_progress_block",
+                        message=(
+                            f"Blocked {tool_name}: this read-only call returned the same "
+                            f"result {repeat_count} times. Stop repeating it unchanged; "
+                            "use the result already provided or try a different query."
+                        ),
+                        tool_name=tool_name,
+                        count=repeat_count,
+                        signature=signature,
+                    )
+                    self._halt_decision = decision
+                    return decision
+
+        return ToolGuardrailDecision(tool_name=tool_name, signature=signature)
+
+    def after_call(
+        self,
+        tool_name: str,
+        args: Mapping[str, Any] | None,
+        result: str | None,
+        *,
+        failed: bool | None = None,
+    ) -> ToolGuardrailDecision:
+        args = _coerce_args(args)
+        signature = ToolCallSignature.from_call(tool_name, args)
+        if failed is None:
+            failed, _ = classify_tool_failure(tool_name, result)
+
+        if failed:
+            exact_count = self._exact_failure_counts.get(signature, 0) + 1
+            self._exact_failure_counts[signature] = exact_count
+            self._no_progress.pop(signature, None)
+
+            same_count = self._same_tool_failure_counts.get(tool_name, 0) + 1
+            self._same_tool_failure_counts[tool_name] = same_count
+
+            if self.config.hard_stop_enabled and same_count >= self.config.same_tool_failure_halt_after:
+                decision = ToolGuardrailDecision(
+                    action="halt",
+                    code="same_tool_failure_halt",
+                    message=(
+                        f"Stopped {tool_name}: it failed {same_count} times this turn. "
+                        "Stop retrying the same failing tool path and choose a different approach."
+                    ),
+                    tool_name=tool_name,
+                    count=same_count,
+                    signature=signature,
+                )
+                self._halt_decision = decision
+                return decision
+
+            if self.config.warnings_enabled and exact_count >= self.config.exact_failure_warn_after:
+                return ToolGuardrailDecision(
+                    action="warn",
+                    code="repeated_exact_failure_warning",
+                    message=(
+                        f"{tool_name} has failed {exact_count} times with identical arguments. "
+                        "This looks like a loop; inspect the error and change strategy "
+                        "instead of retrying it unchanged."
+                    ),
+                    tool_name=tool_name,
+                    count=exact_count,
+                    signature=signature,
+                )
+
+            if self.config.warnings_enabled and same_count >= self.config.same_tool_failure_warn_after:
+                return ToolGuardrailDecision(
+                    action="warn",
+                    code="same_tool_failure_warning",
+                    message=(
+                        f"{tool_name} has failed {same_count} times this turn. "
+                        "This looks like a loop; change approach before retrying."
+                    ),
+                    tool_name=tool_name,
+                    count=same_count,
+                    signature=signature,
+                )
+
+            return ToolGuardrailDecision(tool_name=tool_name, count=exact_count, signature=signature)
+
+        self._exact_failure_counts.pop(signature, None)
+        self._same_tool_failure_counts.pop(tool_name, None)
+
+        if not self._is_idempotent(tool_name):
+            self._no_progress.pop(signature, None)
+            return ToolGuardrailDecision(tool_name=tool_name, signature=signature)
+
+        result_hash = _result_hash(result)
+        previous = self._no_progress.get(signature)
+        repeat_count = 1
+        if previous is not None and previous[0] == result_hash:
+            repeat_count = previous[1] + 1
+        self._no_progress[signature] = (result_hash, repeat_count)
+
+        if self.config.warnings_enabled and repeat_count >= self.config.no_progress_warn_after:
+            return ToolGuardrailDecision(
+                action="warn",
+                code="idempotent_no_progress_warning",
+                message=(
+                    f"{tool_name} returned the same result {repeat_count} times. "
+                    "Use the result already provided or change the query instead of "
+                    "repeating it unchanged."
+                ),
+                tool_name=tool_name,
+                count=repeat_count,
+                signature=signature,
+            )
+
+        return ToolGuardrailDecision(tool_name=tool_name, count=repeat_count, signature=signature)
+
+    def _is_idempotent(self, tool_name: str) -> bool:
+        if tool_name in self.config.mutating_tools:
+            return False
+        return tool_name in self.config.idempotent_tools
+
+
+def toolguard_synthetic_result(decision: ToolGuardrailDecision) -> str:
+    """Build a synthetic role=tool content string for a blocked tool call."""
+    return json.dumps(
+        {
+            "error": decision.message,
+            "guardrail": decision.to_metadata(),
+        },
+        ensure_ascii=False,
+    )
+
+
+def append_toolguard_guidance(result: str, decision: ToolGuardrailDecision) -> str:
+    """Append runtime guidance to the current tool result content."""
+    if decision.action not in {"warn", "halt"} or not decision.message:
+        return result
+    label = "Tool loop hard stop" if decision.action == "halt" else "Tool loop warning"
+    suffix = (
+        f"\n\n[{label}: "
+        f"{decision.code}; count={decision.count}; {decision.message}]"
+    )
+    return (result or "") + suffix
+
+
+def _coerce_args(args: Mapping[str, Any] | None) -> Mapping[str, Any]:
+    return args if isinstance(args, Mapping) else {}
+
+
+def _result_hash(result: str | None) -> str:
+    parsed = safe_json_loads(result or "")
+    if parsed is not None:
+        try:
+            canonical = json.dumps(
+                parsed,
+                ensure_ascii=False,
+                sort_keys=True,
+                separators=(",", ":"),
+                default=str,
+            )
+        except TypeError:
+            canonical = str(parsed)
+    else:
+        canonical = result or ""
+    return _sha256(canonical)
+
+
+def _as_bool(value: Any, default: bool) -> bool:
+    if value is None:
+        return default
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, (int, float)):
+        return bool(value)
+    if isinstance(value, str):
+        lowered = value.strip().lower()
+        if lowered in {"1", "true", "yes", "on", "enabled"}:
+            return True
+        if lowered in {"0", "false", "no", "off", "disabled"}:
+            return False
+    return default
+
+
+def _positive_int(value: Any, default: int) -> int:
+    if value is None:
+        return default
+    try:
+        parsed = int(value)
+    except (TypeError, ValueError):
+        return default
+    return parsed if parsed >= 1 else default
+
+
+def _sha256(value: str) -> str:
+    return hashlib.sha256(value.encode("utf-8")).hexdigest()

agent/transports/__init__.py

@@ -23,9 +23,14 @@ def get_transport(api_mode: str):
     This allows gradual migration — call sites can check for None
     and fall back to the legacy code path.
     """
-    if not _REGISTRY:
-        _discover_transports()
     cls = _REGISTRY.get(api_mode)
+    if cls is None:
+        # The registry can be partially populated when a specific transport
+        # module was imported directly (for example chat_completions before
+        # codex).  Discover on misses, not only when the registry is empty, so
+        # test/order-dependent imports do not make valid api_modes unavailable.
+        _discover_transports()
+        cls = _REGISTRY.get(api_mode)
     if cls is None:
         return None
     return cls()

agent/transports/anthropic.py

@@ -58,6 +58,7 @@ class AnthropicTransport(ProviderTransport):
             context_length: int | None
             base_url: str | None
             fast_mode: bool
+            drop_context_1m_beta: bool
         """
         from agent.anthropic_adapter import build_anthropic_kwargs
 
@@ -73,6 +74,7 @@ class AnthropicTransport(ProviderTransport):
             context_length=params.get("context_length"),
             base_url=params.get("base_url"),
             fast_mode=params.get("fast_mode", False),
+            drop_context_1m_beta=params.get("drop_context_1m_beta", False),
         )
 
     def normalize_response(self, response: Any, **kwargs) -> NormalizedResponse:

agent/transports/chat_completions.py

@@ -12,11 +12,93 @@ reasoning configuration, temperature handling, and extra_body assembly.
 import copy
 from typing import Any, Dict, List, Optional
 
+from agent.lmstudio_reasoning import resolve_lmstudio_effort
+from agent.moonshot_schema import is_moonshot_model, sanitize_moonshot_tools
 from agent.prompt_builder import DEVELOPER_ROLE_MODELS
 from agent.transports.base import ProviderTransport
 from agent.transports.types import NormalizedResponse, ToolCall, Usage
 
 
+def _build_gemini_thinking_config(model: str, reasoning_config: dict | None) -> dict | None:
+    """Translate Hermes/OpenRouter-style reasoning config to Gemini thinkingConfig."""
+    if reasoning_config is None or not isinstance(reasoning_config, dict):
+        return None
+
+    normalized_model = (model or "").strip().lower()
+    if normalized_model.startswith("google/"):
+        normalized_model = normalized_model.split("/", 1)[1]
+
+    # ``thinking_config`` is a Gemini-only request parameter. The same
+    # ``gemini`` provider also serves Gemma (and historically PaLM/Bard);
+    # those reject the field with HTTP 400 "Unknown name 'thinking_config':
+    # Cannot find field" — including the polite ``{"includeThoughts": False}``
+    # form. Omit the field entirely on non-Gemini models. (#17426)
+    if not normalized_model.startswith("gemini"):
+        return None
+
+    if reasoning_config.get("enabled") is False:
+        # Gemini can hide thought parts even when internal thinking still
+        # happens; omit thinkingLevel to avoid model-specific validation quirks.
+        return {"includeThoughts": False}
+
+    effort = str(reasoning_config.get("effort", "medium") or "medium").strip().lower()
+    if effort == "none":
+        return {"includeThoughts": False}
+
+    thinking_config: Dict[str, Any] = {"includeThoughts": True}
+
+    # Gemini 2.5 accepts thinkingBudget; don't guess a budget from Hermes'
+    # coarse effort levels. ``includeThoughts`` alone is enough to surface
+    # thought parts without risking request validation errors.
+    if normalized_model.startswith("gemini-2.5-"):
+        return thinking_config
+
+    if effort not in {"minimal", "low", "medium", "high", "xhigh"}:
+        effort = "medium"
+
+    # Gemini 3 Flash documents low/medium/high thinking levels; Gemini 3 Pro
+    # is stricter (low/high). Clamp Hermes' wider effort set to what each
+    # family accepts so we never forward an undocumented level verbatim.
+    if normalized_model.startswith(("gemini-3", "gemini-3.1")):
+        if "flash" in normalized_model:
+            if effort in {"minimal", "low"}:
+                thinking_config["thinkingLevel"] = "low"
+            elif effort in {"high", "xhigh"}:
+                thinking_config["thinkingLevel"] = "high"
+            else:
+                thinking_config["thinkingLevel"] = "medium"
+        elif "pro" in normalized_model:
+            thinking_config["thinkingLevel"] = (
+                "high" if effort in {"high", "xhigh"} else "low"
+            )
+
+    return thinking_config
+
+
+def _snake_case_gemini_thinking_config(config: dict | None) -> dict | None:
+    """Convert Gemini thinking config keys to the OpenAI-compat field names."""
+    if not isinstance(config, dict) or not config:
+        return None
+
+    translated: Dict[str, Any] = {}
+    if isinstance(config.get("includeThoughts"), bool):
+        translated["include_thoughts"] = config["includeThoughts"]
+    if isinstance(config.get("thinkingLevel"), str) and config["thinkingLevel"].strip():
+        translated["thinking_level"] = config["thinkingLevel"].strip().lower()
+    if isinstance(config.get("thinkingBudget"), (int, float)):
+        translated["thinking_budget"] = int(config["thinkingBudget"])
+    return translated or None
+
+
+def _is_gemini_openai_compat_base_url(base_url: Any) -> bool:
+    normalized = str(base_url or "").strip().rstrip("/").lower()
+    if not normalized:
+        return False
+    if "generativelanguage.googleapis.com" not in normalized:
+        return False
+    return normalized.endswith("/openai")
+
+
 class ChatCompletionsTransport(ProviderTransport):
     """Transport for api_mode='chat_completions'.
 
@@ -30,15 +112,15 @@ class ChatCompletionsTransport(ProviderTransport):
     def convert_messages(self, messages: List[Dict[str, Any]], **kwargs) -> List[Dict[str, Any]]:
         """Messages are already in OpenAI format — sanitize Codex leaks only.
 
-        Strips Codex Responses API fields (``codex_reasoning_items`` on the
-        message, ``call_id``/``response_item_id`` on tool_calls) that strict
-        chat-completions providers reject with 400/422.
+        Strips Codex Responses API fields (``codex_reasoning_items`` /
+        ``codex_message_items`` on the message, ``call_id``/``response_item_id``
+        on tool_calls) that strict chat-completions providers reject with 400/422.
         """
         needs_sanitize = False
         for msg in messages:
             if not isinstance(msg, dict):
                 continue
-            if "codex_reasoning_items" in msg:
+            if "codex_reasoning_items" in msg or "codex_message_items" in msg:
                 needs_sanitize = True
                 break
             tool_calls = msg.get("tool_calls")
@@ -58,6 +140,7 @@ class ChatCompletionsTransport(ProviderTransport):
             if not isinstance(msg, dict):
                 continue
             msg.pop("codex_reasoning_items", None)
+            msg.pop("codex_message_items", None)
             tool_calls = msg.get("tool_calls")
             if isinstance(tool_calls, list):
                 for tc in tool_calls:
@@ -99,6 +182,7 @@ class ChatCompletionsTransport(ProviderTransport):
             is_github_models: bool
             is_nvidia_nim: bool
             is_kimi: bool
+            is_lmstudio: bool
             is_custom_provider: bool
             ollama_num_ctx: int | None
             # Provider routing
@@ -112,6 +196,7 @@ class ChatCompletionsTransport(ProviderTransport):
             # Reasoning
             supports_reasoning: bool
             github_reasoning_extra: dict | None
+            lmstudio_reasoning_options: list[str] | None  # raw allowed_options from /api/v1/models
             # Claude on OpenRouter/Nous max output
             anthropic_max_output: int | None
             # Extra
@@ -172,6 +257,11 @@ class ChatCompletionsTransport(ProviderTransport):
 
         # Tools
         if tools:
+            # Moonshot/Kimi uses a stricter flavored JSON Schema.  Rewriting
+            # tool parameters here keeps aggregator routes (Nous, OpenRouter,
+            # etc.) compatible, in addition to direct moonshot.ai endpoints.
+            if is_moonshot_model(model):
+                tools = sanitize_moonshot_tools(tools)
             api_kwargs["tools"] = tools
 
         # max_tokens resolution — priority: ephemeral > user > provider default
@@ -181,6 +271,7 @@ class ChatCompletionsTransport(ProviderTransport):
         anthropic_max_out = params.get("anthropic_max_output")
         is_nvidia_nim = params.get("is_nvidia_nim", False)
         is_kimi = params.get("is_kimi", False)
+        is_tokenhub = params.get("is_tokenhub", False)
         reasoning_config = params.get("reasoning_config")
 
         if ephemeral is not None and max_tokens_fn:
@@ -212,12 +303,41 @@ class ChatCompletionsTransport(ProviderTransport):
                         _kimi_effort = _e
                 api_kwargs["reasoning_effort"] = _kimi_effort
 
+        # Tencent TokenHub: top-level reasoning_effort (unless thinking disabled)
+        if is_tokenhub:
+            _tokenhub_thinking_off = bool(
+                reasoning_config
+                and isinstance(reasoning_config, dict)
+                and reasoning_config.get("enabled") is False
+            )
+            if not _tokenhub_thinking_off:
+                _tokenhub_effort = "high"
+                if reasoning_config and isinstance(reasoning_config, dict):
+                    _e = (reasoning_config.get("effort") or "").strip().lower()
+                    if _e in ("low", "medium", "high"):
+                        _tokenhub_effort = _e
+                api_kwargs["reasoning_effort"] = _tokenhub_effort
+
+        # LM Studio: top-level reasoning_effort. Only emit when the model
+        # declares reasoning support via /api/v1/models capabilities (gated
+        # upstream by params["supports_reasoning"]). resolve_lmstudio_effort
+        # is shared with run_agent's summary path so both stay in sync.
+        if params.get("is_lmstudio", False) and params.get("supports_reasoning", False):
+            _lm_effort = resolve_lmstudio_effort(
+                reasoning_config,
+                params.get("lmstudio_reasoning_options"),
+            )
+            if _lm_effort is not None:
+                api_kwargs["reasoning_effort"] = _lm_effort
+
         # extra_body assembly
         extra_body: Dict[str, Any] = {}
 
         is_openrouter = params.get("is_openrouter", False)
         is_nous = params.get("is_nous", False)
         is_github_models = params.get("is_github_models", False)
+        provider_name = str(params.get("provider_name") or "").strip().lower()
+        base_url = params.get("base_url")
 
         provider_prefs = params.get("provider_preferences")
         if provider_prefs and is_openrouter:
@@ -233,8 +353,9 @@ class ChatCompletionsTransport(ProviderTransport):
                 "type": "enabled" if _kimi_thinking_enabled else "disabled",
             }
 
-        # Reasoning
-        if params.get("supports_reasoning", False):
+        # Reasoning. LM Studio is handled above via top-level reasoning_effort,
+        # so skip emitting extra_body.reasoning for it.
+        if params.get("supports_reasoning", False) and not params.get("is_lmstudio", False):
             if is_github_models:
                 gh_reasoning = params.get("github_reasoning_extra")
                 if gh_reasoning is not None:
@@ -270,6 +391,23 @@ class ChatCompletionsTransport(ProviderTransport):
         if is_qwen:
             extra_body["vl_high_resolution_images"] = True
 
+        if provider_name == "gemini":
+            raw_thinking_config = _build_gemini_thinking_config(model, reasoning_config)
+            if _is_gemini_openai_compat_base_url(base_url):
+                thinking_config = _snake_case_gemini_thinking_config(raw_thinking_config)
+                if thinking_config:
+                    openai_compat_extra = extra_body.get("extra_body", {})
+                    google_extra = openai_compat_extra.get("google", {})
+                    google_extra["thinking_config"] = thinking_config
+                    openai_compat_extra["google"] = google_extra
+                    extra_body["extra_body"] = openai_compat_extra
+            elif raw_thinking_config:
+                extra_body["thinking_config"] = raw_thinking_config
+        elif provider_name == "google-gemini-cli":
+            thinking_config = _build_gemini_thinking_config(model, reasoning_config)
+            if thinking_config:
+                extra_body["thinking_config"] = thinking_config
+
         # Merge any pre-built extra_body additions
         additions = params.get("extra_body_additions")
         if additions:
@@ -339,9 +477,13 @@ class ChatCompletionsTransport(ProviderTransport):
         # so keep them apart in provider_data rather than merging.
         reasoning = getattr(msg, "reasoning", None)
         reasoning_content = getattr(msg, "reasoning_content", None)
+        if reasoning_content is None and hasattr(msg, "model_extra"):
+            model_extra = getattr(msg, "model_extra", None) or {}
+            if isinstance(model_extra, dict) and "reasoning_content" in model_extra:
+                reasoning_content = model_extra["reasoning_content"]
 
         provider_data: Dict[str, Any] = {}
-        if reasoning_content:
+        if reasoning_content is not None:
             provider_data["reasoning_content"] = reasoning_content
         rd = getattr(msg, "reasoning_details", None)
         if rd:

agent/transports/codex.py

@@ -8,7 +8,7 @@ streaming, or the _run_codex_stream() call path.
 from typing import Any, Dict, List, Optional
 
 from agent.transports.base import ProviderTransport
-from agent.transports.types import NormalizedResponse, ToolCall, Usage
+from agent.transports.types import NormalizedResponse, ToolCall
 
 
 class ResponsesApiTransport(ProviderTransport):
@@ -120,12 +120,41 @@ class ResponsesApiTransport(ProviderTransport):
         if request_overrides:
             kwargs.update(request_overrides)
 
+        if is_codex_backend:
+            prompt_cache_key = kwargs.get("prompt_cache_key")
+            cache_scope_id = str(prompt_cache_key or session_id or "").strip()
+            if cache_scope_id:
+                existing_extra_headers = kwargs.get("extra_headers")
+                merged_extra_headers: Dict[str, str] = {}
+                if isinstance(existing_extra_headers, dict):
+                    merged_extra_headers.update(
+                        {
+                            str(key): str(value)
+                            for key, value in existing_extra_headers.items()
+                            if key and value is not None
+                        }
+                    )
+                merged_extra_headers["session_id"] = cache_scope_id
+                merged_extra_headers["x-client-request-id"] = cache_scope_id
+                kwargs["extra_headers"] = merged_extra_headers
+
         max_tokens = params.get("max_tokens")
         if max_tokens is not None and not is_codex_backend:
             kwargs["max_output_tokens"] = max_tokens
 
         if is_xai_responses and session_id:
-            kwargs["extra_headers"] = {"x-grok-conv-id": session_id}
+            existing_extra_headers = kwargs.get("extra_headers")
+            merged_extra_headers: Dict[str, str] = {}
+            if isinstance(existing_extra_headers, dict):
+                merged_extra_headers.update(
+                    {
+                        str(key): str(value)
+                        for key, value in existing_extra_headers.items()
+                        if key and value is not None
+                    }
+                )
+            merged_extra_headers["x-grok-conv-id"] = session_id
+            kwargs["extra_headers"] = merged_extra_headers
 
         return kwargs
 
@@ -133,8 +162,6 @@ class ResponsesApiTransport(ProviderTransport):
         """Normalize Codex Responses API response to NormalizedResponse."""
         from agent.codex_responses_adapter import (
             _normalize_codex_response,
-            _extract_responses_message_text,
-            _extract_responses_reasoning_text,
         )
 
         # _normalize_codex_response returns (SimpleNamespace, finish_reason_str)
@@ -160,6 +187,8 @@ class ResponsesApiTransport(ProviderTransport):
         provider_data = {}
         if msg and hasattr(msg, "codex_reasoning_items") and msg.codex_reasoning_items:
             provider_data["codex_reasoning_items"] = msg.codex_reasoning_items
+        if msg and hasattr(msg, "codex_message_items") and msg.codex_message_items:
+            provider_data["codex_message_items"] = msg.codex_message_items
         if msg and hasattr(msg, "reasoning_details") and msg.reasoning_details:
             provider_data["reasoning_details"] = msg.reasoning_details
 

agent/transports/types.py

@@ -61,6 +61,20 @@ class ToolCall:
         """Codex response_item_id from provider_data."""
         return (self.provider_data or {}).get("response_item_id")
 
+    @property
+    def extra_content(self) -> Optional[Dict[str, Any]]:
+        """Gemini extra_content (thought_signature) from provider_data.
+
+        Gemini 3 thinking models attach ``extra_content`` with a
+        ``thought_signature`` to each tool call.  This signature must be
+        replayed on subsequent API calls — without it the API rejects the
+        request with HTTP 400.  The chat_completions transport stores this
+        in ``provider_data["extra_content"]``; this property exposes it so
+        ``_build_assistant_message`` can ``getattr(tc, "extra_content")``
+        uniformly.
+        """
+        return (self.provider_data or {}).get("extra_content")
+
 
 @dataclass
 class Usage:
@@ -83,7 +97,7 @@ class NormalizedResponse:
     Response-level ``provider_data`` examples:
 
     * Anthropic: ``{"reasoning_details": [...]}``
-    * Codex: ``{"codex_reasoning_items": [...]}``
+    * Codex: ``{"codex_reasoning_items": [...], "codex_message_items": [...]}``
     * Others: ``None``
     """
 
@@ -112,6 +126,11 @@ class NormalizedResponse:
         pd = self.provider_data or {}
         return pd.get("codex_reasoning_items")
 
+    @property
+    def codex_message_items(self):
+        pd = self.provider_data or {}
+        return pd.get("codex_message_items")
+
 
 # ---------------------------------------------------------------------------
 # Factory helpers

agent/usage_pricing.py

@@ -359,6 +359,25 @@ _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
         source_url="https://aws.amazon.com/bedrock/pricing/",
         pricing_version="bedrock-pricing-2026-04",
     ),
+    # MiniMax
+    (
+        "minimax",
+        "minimax-m2.7",
+    ): PricingEntry(
+        input_cost_per_million=Decimal("0.30"),
+        output_cost_per_million=Decimal("1.20"),
+        source="official_docs_snapshot",
+        pricing_version="minimax-pricing-2026-04",
+    ),
+    (
+        "minimax-cn",
+        "minimax-m2.7",
+    ): PricingEntry(
+        input_cost_per_million=Decimal("0.30"),
+        output_cost_per_million=Decimal("1.20"),
+        source="official_docs_snapshot",
+        pricing_version="minimax-pricing-2026-04",
+    ),
 }
 
 
@@ -400,6 +419,8 @@ def resolve_billing_route(
         return BillingRoute(provider="anthropic", model=model.split("/")[-1], base_url=base_url or "", billing_mode="official_docs_snapshot")
     if provider_name == "openai":
         return BillingRoute(provider="openai", model=model.split("/")[-1], base_url=base_url or "", billing_mode="official_docs_snapshot")
+    if provider_name in {"minimax", "minimax-cn"}:
+        return BillingRoute(provider=provider_name, model=model.split("/")[-1], base_url=base_url or "", billing_mode="official_docs_snapshot")
     if provider_name in {"custom", "local"} or (base and "localhost" in base):
         return BillingRoute(provider=provider_name or "custom", model=model, base_url=base_url or "", billing_mode="unknown")
     return BillingRoute(provider=provider_name or "unknown", model=model.split("/")[-1] if model else "", base_url=base_url or "", billing_mode="unknown")

apps/dashboard/README.md

@@ -10,17 +10,34 @@ Browser-based dashboard for managing Hermes Agent configuration, API keys, and m
 
 ## Development
 
-```bash
-# Start the backend API server
-cd ../
-python -m hermes_cli.main web --no-open
+Install workspace dependencies from the repo root first:
 
-# In another terminal, start the Vite dev server (with HMR + API proxy)
-cd web/
+```bash
+npm install
+```
+
+Start the backend API server from the repo root:
+
+```bash
+hermes dashboard --tui --no-open
+```
+
+`--tui` exposes the in-browser Chat tab through `/api/pty`. Omit it if you only need the config/session dashboard.
+
+In another terminal, start the Vite dev server:
+
+```bash
+cd apps/dashboard
 npm run dev
 ```
 
-The Vite dev server proxies `/api` requests to `http://127.0.0.1:9119` (the FastAPI backend).
+The Vite dev server proxies `/api`, `/api/pty`, and `/dashboard-plugins` to `http://127.0.0.1:9119` (the FastAPI backend). It also fetches the backend's `index.html` on each dev page load so the ephemeral session token stays in sync.
+
+If the `hermes` entry point is not installed, use:
+
+```bash
+python -m hermes_cli.main dashboard --tui --no-open
+```
 
 ## Build
 
@@ -28,7 +45,7 @@ The Vite dev server proxies `/api` requests to `http://127.0.0.1:9119` (the Fast
 npm run build
 ```
 
-This outputs to `../hermes_cli/web_dist/`, which the FastAPI server serves as a static SPA. The built assets are included in the Python package via `pyproject.toml` package-data.
+This outputs to `../../hermes_cli/web_dist/`, which the FastAPI server serves as a static SPA. The built assets are included in the Python package via `pyproject.toml` package-data.
 
 ## Structure
 

apps/dashboard/package.json

@@ -1,10 +1,10 @@
 {
-  "name": "web",
+  "name": "dashboard",
   "private": true,
   "version": "0.0.0",
   "type": "module",
   "scripts": {
-    "sync-assets": "rm -rf public/fonts public/ds-assets && cp -r node_modules/@nous-research/ui/dist/fonts public/fonts && cp -r node_modules/@nous-research/ui/dist/assets public/ds-assets",
+    "sync-assets": "node scripts/sync-assets.cjs",
     "predev": "npm run sync-assets",
     "prebuild": "npm run sync-assets",
     "dev": "vite",
@@ -13,10 +13,16 @@
     "preview": "vite preview"
   },
   "dependencies": {
-    "@nous-research/ui": "^0.3.0",
+    "@hermes/shared": "file:../shared",
+    "@nous-research/ui": "^0.10.0",
     "@observablehq/plot": "^0.6.17",
     "@react-three/fiber": "^9.6.0",
     "@tailwindcss/vite": "^4.2.1",
+    "@xterm/addon-fit": "^0.11.0",
+    "@xterm/addon-unicode11": "^0.9.0",
+    "@xterm/addon-web-links": "^0.12.0",
+    "@xterm/addon-webgl": "^0.19.0",
+    "@xterm/xterm": "^6.0.0",
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
     "gsap": "^3.15.0",
@@ -26,7 +32,8 @@
     "react-dom": "^19.2.4",
     "react-router-dom": "^7.14.1",
     "tailwind-merge": "^3.5.0",
-    "tailwindcss": "^4.2.1"
+    "tailwindcss": "^4.2.1",
+    "unicode-animations": "^1.0.3"
   },
   "devDependencies": {
     "@eslint/js": "^9.39.4",

apps/dashboard/scripts/sync-assets.cjs

@@ -0,0 +1,46 @@
+#!/usr/bin/env node
+/**
+ * Copy font and asset folders from @nous-research/ui into public/ for Vite.
+ *
+ * Locates @nous-research/ui by walking up from this script looking for
+ * node_modules/@nous-research/ui — works whether the dep is co-located
+ * (non-workspace layout) or hoisted to the repo root (npm workspaces).
+ */
+const fs = require('node:fs')
+const path = require('node:path')
+
+const DASHBOARD_ROOT = path.resolve(__dirname, '..')
+
+function locateUiPackage() {
+  let dir = DASHBOARD_ROOT
+  const { root } = path.parse(dir)
+  while (true) {
+    const candidate = path.join(dir, 'node_modules', '@nous-research', 'ui')
+    if (fs.existsSync(path.join(candidate, 'package.json'))) {
+      return candidate
+    }
+    if (dir === root) break
+    dir = path.dirname(dir)
+  }
+  throw new Error(
+    '@nous-research/ui not found. Run `npm install` from the repo root.'
+  )
+}
+
+const uiRoot = locateUiPackage()
+const distRoot = path.join(uiRoot, 'dist')
+
+const mappings = [
+  ['fonts', path.join(DASHBOARD_ROOT, 'public', 'fonts')],
+  ['assets', path.join(DASHBOARD_ROOT, 'public', 'ds-assets')],
+]
+
+for (const [srcName, destPath] of mappings) {
+  const srcPath = path.join(distRoot, srcName)
+  if (!fs.existsSync(srcPath)) {
+    throw new Error(`Missing ${srcPath} in @nous-research/ui — rebuild that package.`)
+  }
+  fs.rmSync(destPath, { recursive: true, force: true })
+  fs.cpSync(srcPath, destPath, { recursive: true })
+  console.log(`synced ${path.relative(DASHBOARD_ROOT, destPath)}`)
+}

apps/dashboard/src/App.tsx

@@ -0,0 +1,819 @@
+import {
+  useCallback,
+  useEffect,
+  useMemo,
+  useState,
+  type ComponentType,
+  type ReactNode,
+} from "react";
+import {
+  Routes,
+  Route,
+  NavLink,
+  Navigate,
+  useLocation,
+  useNavigate,
+} from "react-router-dom";
+import {
+  Activity,
+  BarChart3,
+  BookOpen,
+  Clock,
+  Code,
+  Cpu,
+  Database,
+  Download,
+  Eye,
+  FileText,
+  Globe,
+  Heart,
+  KeyRound,
+  Menu,
+  MessageSquare,
+  Package,
+  Puzzle,
+  RotateCw,
+  Settings,
+  Shield,
+  Sparkles,
+  Star,
+  Terminal,
+  Users,
+  Wrench,
+  X,
+  Zap,
+} from "lucide-react";
+import { Button } from "@nous-research/ui/ui/components/button";
+import { ListItem } from "@nous-research/ui/ui/components/list-item";
+import { SelectionSwitcher } from "@nous-research/ui/ui/components/selection-switcher";
+import { Spinner } from "@nous-research/ui/ui/components/spinner";
+import { Typography } from "@/components/NouiTypography";
+import { cn } from "@/lib/utils";
+import { Backdrop } from "@/components/Backdrop";
+import { SidebarFooter } from "@/components/SidebarFooter";
+import { SidebarStatusStrip } from "@/components/SidebarStatusStrip";
+import { PageHeaderProvider } from "@/contexts/PageHeaderProvider";
+import { useSystemActions } from "@/contexts/useSystemActions";
+import type { SystemAction } from "@/contexts/system-actions-context";
+import ConfigPage from "@/pages/ConfigPage";
+import DocsPage from "@/pages/DocsPage";
+import EnvPage from "@/pages/EnvPage";
+import SessionsPage from "@/pages/SessionsPage";
+import LogsPage from "@/pages/LogsPage";
+import AnalyticsPage from "@/pages/AnalyticsPage";
+import ModelsPage from "@/pages/ModelsPage";
+import CronPage from "@/pages/CronPage";
+import ProfilesPage from "@/pages/ProfilesPage";
+import SkillsPage from "@/pages/SkillsPage";
+import PluginsPage from "@/pages/PluginsPage";
+import ChatPage from "@/pages/ChatPage";
+import { LanguageSwitcher } from "@/components/LanguageSwitcher";
+import { ThemeSwitcher } from "@/components/ThemeSwitcher";
+import { useI18n } from "@/i18n";
+import type { Translations } from "@/i18n/types";
+import { PluginPage, PluginSlot, usePlugins } from "@/plugins";
+import type { PluginManifest } from "@/plugins";
+import { useTheme } from "@/themes";
+import { isDashboardEmbeddedChatEnabled } from "@/lib/dashboard-flags";
+
+function RootRedirect() {
+  return <Navigate to="/sessions" replace />;
+}
+
+function UnknownRouteFallback({ pluginsLoading }: { pluginsLoading: boolean }) {
+  if (pluginsLoading) {
+    // Render nothing during the plugin-load window — a spinner here would just flash.
+    return null;
+  }
+  return <Navigate to="/sessions" replace />;
+}
+
+const CHAT_NAV_ITEM: NavItem = {
+  path: "/chat",
+  labelKey: "chat",
+  label: "Chat",
+  icon: Terminal,
+};
+
+/**
+ * Built-in routes except /chat.  Chat is rendered persistently (outside
+ * <Routes>) when embedded — see the persistent chat host block rendered
+ * inline near the bottom of this file — so the PTY child, WebSocket,
+ * and xterm instance survive when the user visits another tab and comes
+ * back.  A `display:none` toggle hides the terminal without unmounting.
+ * Routing still owns the URL so /chat deep-links, browser back/forward,
+ * and nav highlight keep working.
+ */
+const BUILTIN_ROUTES_CORE: Record<string, ComponentType> = {
+  "/": RootRedirect,
+  "/sessions": SessionsPage,
+  "/analytics": AnalyticsPage,
+  "/models": ModelsPage,
+  "/logs": LogsPage,
+  "/cron": CronPage,
+  "/skills": SkillsPage,
+  "/plugins": PluginsPage,
+  "/profiles": ProfilesPage,
+  "/config": ConfigPage,
+  "/env": EnvPage,
+  "/docs": DocsPage,
+};
+
+// Route placeholder for /chat.  The persistent ChatPage host (rendered
+// outside <Routes> when embedded chat is on) paints on top; this empty
+// element just claims the path so the `*` catch-all redirect doesn't
+// fire when the user navigates to /chat.
+function ChatRouteSink() {
+  return null;
+}
+
+const BUILTIN_NAV_REST: NavItem[] = [
+  {
+    path: "/sessions",
+    labelKey: "sessions",
+    label: "Sessions",
+    icon: MessageSquare,
+  },
+  {
+    path: "/analytics",
+    labelKey: "analytics",
+    label: "Analytics",
+    icon: BarChart3,
+  },
+  {
+    path: "/models",
+    labelKey: "models",
+    label: "Models",
+    icon: Cpu,
+  },
+  { path: "/logs", labelKey: "logs", label: "Logs", icon: FileText },
+  { path: "/cron", labelKey: "cron", label: "Cron", icon: Clock },
+  { path: "/skills", labelKey: "skills", label: "Skills", icon: Package },
+  { path: "/plugins", labelKey: "plugins", label: "Plugins", icon: Puzzle },
+  { path: "/profiles", labelKey: "profiles", label: "Profiles", icon: Users },
+  { path: "/config", labelKey: "config", label: "Config", icon: Settings },
+  { path: "/env", labelKey: "keys", label: "Keys", icon: KeyRound },
+  {
+    path: "/docs",
+    labelKey: "documentation",
+    label: "Documentation",
+    icon: BookOpen,
+  },
+];
+
+const ICON_MAP: Record<string, ComponentType<{ className?: string }>> = {
+  Activity,
+  BarChart3,
+  Clock,
+  Cpu,
+  FileText,
+  KeyRound,
+  MessageSquare,
+  Package,
+  Settings,
+  Puzzle,
+  Sparkles,
+  Terminal,
+  Globe,
+  Database,
+  Shield,
+  Users,
+  Wrench,
+  Zap,
+  Heart,
+  Star,
+  Code,
+  Eye,
+};
+
+function resolveIcon(name: string): ComponentType<{ className?: string }> {
+  return ICON_MAP[name] ?? Puzzle;
+}
+
+function buildNavItems(
+  builtIn: NavItem[],
+  manifests: PluginManifest[],
+): NavItem[] {
+  const items = [...builtIn];
+
+  for (const manifest of manifests) {
+    if (manifest.tab.override) continue;
+    if (manifest.tab.hidden) continue;
+
+    const pluginItem: NavItem = {
+      path: manifest.tab.path,
+      label: manifest.label,
+      icon: resolveIcon(manifest.icon),
+    };
+
+    const pos = manifest.tab.position ?? "end";
+    if (pos === "end") {
+      items.push(pluginItem);
+    } else if (pos.startsWith("after:")) {
+      const target = "/" + pos.slice(6);
+      const idx = items.findIndex((i) => i.path === target);
+      items.splice(idx >= 0 ? idx + 1 : items.length, 0, pluginItem);
+    } else if (pos.startsWith("before:")) {
+      const target = "/" + pos.slice(7);
+      const idx = items.findIndex((i) => i.path === target);
+      items.splice(idx >= 0 ? idx : items.length, 0, pluginItem);
+    } else {
+      items.push(pluginItem);
+    }
+  }
+
+  return items;
+}
+
+/** Split merged nav into built-in sidebar entries vs plugin tabs, preserving plugin order hints. */
+function partitionSidebarNav(
+  builtIn: NavItem[],
+  manifests: PluginManifest[],
+): { coreItems: NavItem[]; pluginItems: NavItem[] } {
+  const merged = buildNavItems(builtIn, manifests);
+  const builtinPaths = new Set(builtIn.map((i) => i.path));
+  const coreItems: NavItem[] = [];
+  const pluginItems: NavItem[] = [];
+  for (const item of merged) {
+    if (builtinPaths.has(item.path)) coreItems.push(item);
+    else pluginItems.push(item);
+  }
+  return { coreItems, pluginItems };
+}
+
+function buildRoutes(
+  builtinRoutes: Record<string, ComponentType>,
+  manifests: PluginManifest[],
+): Array<{
+  key: string;
+  path: string;
+  element: ReactNode;
+}> {
+  const byOverride = new Map<string, PluginManifest>();
+  const addons: PluginManifest[] = [];
+
+  for (const m of manifests) {
+    if (m.tab.override) {
+      byOverride.set(m.tab.override, m);
+    } else {
+      addons.push(m);
+    }
+  }
+
+  const routes: Array<{
+    key: string;
+    path: string;
+    element: ReactNode;
+  }> = [];
+
+  for (const [path, Component] of Object.entries(builtinRoutes)) {
+    const om = byOverride.get(path);
+    if (om) {
+      routes.push({
+        key: `override:${om.name}`,
+        path,
+        element: <PluginPage name={om.name} />,
+      });
+    } else {
+      routes.push({ key: `builtin:${path}`, path, element: <Component /> });
+    }
+  }
+
+  for (const m of addons) {
+    if (m.tab.hidden) continue;
+    if (m.tab.path === "/plugins") continue;
+    if (builtinRoutes[m.tab.path]) continue;
+    routes.push({
+      key: `plugin:${m.name}`,
+      path: m.tab.path,
+      element: <PluginPage name={m.name} />,
+    });
+  }
+
+  for (const m of manifests) {
+    if (!m.tab.hidden) continue;
+    if (m.tab.path === "/plugins") continue;
+    if (builtinRoutes[m.tab.path] || m.tab.override) continue;
+    routes.push({
+      key: `plugin:hidden:${m.name}`,
+      path: m.tab.path,
+      element: <PluginPage name={m.name} />,
+    });
+  }
+
+  return routes;
+}
+
+export default function App() {
+  const { t } = useI18n();
+  const { pathname } = useLocation();
+  const { manifests, loading: pluginsLoading } = usePlugins();
+  const { theme } = useTheme();
+  const [mobileOpen, setMobileOpen] = useState(false);
+  const closeMobile = useCallback(() => setMobileOpen(false), []);
+  const isDocsRoute = pathname === "/docs" || pathname === "/docs/";
+  const normalizedPath = pathname.replace(/\/$/, "") || "/";
+  const isChatRoute = normalizedPath === "/chat";
+  const embeddedChat = isDashboardEmbeddedChatEnabled();
+
+  // A plugin can replace the built-in /chat page via `tab.override: "/chat"`
+  // in its manifest.  When one does, `buildRoutes` already swaps the route
+  // element for <PluginPage /> — but we also have to suppress the
+  // persistent ChatPage host below, or the plugin's page and the built-in
+  // terminal would paint on top of each other.  The override is niche
+  // (nothing ships overriding /chat today) but it's an advertised
+  // extension point, so preserve the pre-persistence contract: when a
+  // plugin owns /chat, the built-in chat UI is entirely absent.
+  //
+  // Waiting on `pluginsLoading` is load-bearing: manifests arrive
+  // asynchronously from /api/dashboard/plugins, so on initial render
+  // `chatOverriddenByPlugin` is always false.  Without the loading
+  // gate, the persistent host would mount, spawn a PTY, and THEN get
+  // yanked out from under the user when the plugin's manifest resolves
+  // — killing the session mid-paint.  Delaying host mount by the
+  // plugin-load window (typically <50ms, worst case 2s safety timeout)
+  // is the cheaper trade-off.
+  const chatOverriddenByPlugin = useMemo(
+    () => manifests.some((m) => m.tab.override === "/chat"),
+    [manifests],
+  );
+
+  const builtinRoutes = useMemo(
+    () => ({
+      ...BUILTIN_ROUTES_CORE,
+      ...(embeddedChat ? { "/chat": ChatRouteSink } : {}),
+    }),
+    [embeddedChat],
+  );
+
+  const builtinNav = useMemo(
+    () =>
+      embeddedChat ? [CHAT_NAV_ITEM, ...BUILTIN_NAV_REST] : BUILTIN_NAV_REST,
+    [embeddedChat],
+  );
+
+  const sidebarNav = useMemo(
+    () => partitionSidebarNav(builtinNav, manifests),
+    [builtinNav, manifests],
+  );
+  const routes = useMemo(
+    () => buildRoutes(builtinRoutes, manifests),
+    [builtinRoutes, manifests],
+  );
+  const pluginTabMeta = useMemo(
+    () =>
+      manifests
+        .filter((m) => !m.tab.hidden)
+        .map((m) => ({
+          path: m.tab.override ?? m.tab.path,
+          label: m.label,
+        })),
+    [manifests],
+  );
+
+  const layoutVariant = theme.layoutVariant ?? "standard";
+
+  useEffect(() => {
+    if (!mobileOpen) return;
+    const onKey = (e: KeyboardEvent) => {
+      if (e.key === "Escape") setMobileOpen(false);
+    };
+    document.addEventListener("keydown", onKey);
+    const prevOverflow = document.body.style.overflow;
+    document.body.style.overflow = "hidden";
+    return () => {
+      document.removeEventListener("keydown", onKey);
+      document.body.style.overflow = prevOverflow;
+    };
+  }, [mobileOpen]);
+
+  useEffect(() => {
+    const mql = window.matchMedia("(min-width: 1024px)");
+    const onChange = (e: MediaQueryListEvent) => {
+      if (e.matches) setMobileOpen(false);
+    };
+    mql.addEventListener("change", onChange);
+    return () => mql.removeEventListener("change", onChange);
+  }, []);
+
+  return (
+    <div
+      data-layout-variant={layoutVariant}
+      className="font-mondwest flex h-dvh max-h-dvh min-h-0 flex-col overflow-hidden bg-black uppercase text-midground antialiased"
+    >
+      <SelectionSwitcher />
+      <Backdrop />
+      <PluginSlot name="backdrop" />
+
+      <header
+        className={cn(
+          "lg:hidden fixed top-0 left-0 right-0 z-40 h-12",
+          "flex items-center gap-2 px-3",
+          "border-b border-current/20",
+          "bg-background-base/90 backdrop-blur-sm",
+        )}
+        style={{
+          background: "var(--component-header-background)",
+          borderImage: "var(--component-header-border-image)",
+          clipPath: "var(--component-header-clip-path)",
+        }}
+      >
+        <Button
+          ghost
+          size="icon"
+          onClick={() => setMobileOpen(true)}
+          aria-label={t.app.openNavigation}
+          aria-expanded={mobileOpen}
+          aria-controls="app-sidebar"
+          className="text-midground/70 hover:text-midground"
+        >
+          <Menu />
+        </Button>
+
+        <Typography
+          className="font-bold text-[0.95rem] leading-[0.95] tracking-[0.05em] text-midground"
+          style={{ mixBlendMode: "plus-lighter" }}
+        >
+          {t.app.brand}
+        </Typography>
+      </header>
+
+      {mobileOpen && (
+        <Button
+          ghost
+          aria-label={t.app.closeNavigation}
+          onClick={closeMobile}
+          className={cn(
+            "lg:hidden fixed inset-0 z-40 p-0 block",
+            "bg-black/60 backdrop-blur-sm",
+          )}
+        />
+      )}
+
+      <PluginSlot name="header-banner" />
+
+      <div className="flex min-h-0 min-w-0 flex-1 flex-col overflow-hidden pt-12 lg:pt-0">
+        <div className="flex min-h-0 min-w-0 flex-1">
+          <aside
+            id="app-sidebar"
+            aria-label={t.app.navigation}
+            className={cn(
+              "fixed top-0 left-0 z-50 flex h-dvh max-h-dvh w-64 min-h-0 flex-col",
+              "border-r border-current/20",
+              "bg-background-base/95 backdrop-blur-sm",
+              "transition-transform duration-200 ease-out",
+              mobileOpen ? "translate-x-0" : "-translate-x-full",
+              "lg:sticky lg:top-0 lg:translate-x-0 lg:shrink-0",
+            )}
+            style={{
+              background: "var(--component-sidebar-background)",
+              clipPath: "var(--component-sidebar-clip-path)",
+              borderImage: "var(--component-sidebar-border-image)",
+            }}
+          >
+            <div
+              className={cn(
+                "flex h-14 shrink-0 items-center justify-between gap-2",
+                "border-b border-current/20",
+              )}
+            >
+              <div className="flex items-center gap-2">
+                <PluginSlot name="header-left" />
+
+                <Typography
+                  className="font-bold text-[1.125rem] leading-[0.95] tracking-[0.0525rem] text-midground"
+                  style={{ mixBlendMode: "plus-lighter" }}
+                >
+                  Hermes
+                  <br />
+                  Agent
+                </Typography>
+              </div>
+
+              <Button
+                ghost
+                size="icon"
+                onClick={closeMobile}
+                aria-label={t.app.closeNavigation}
+                className="lg:hidden text-midground/70 hover:text-midground"
+              >
+                <X />
+              </Button>
+            </div>
+
+            <nav
+              className="min-h-0 w-full flex-1 overflow-y-auto overflow-x-hidden border-t border-current/10 py-2"
+              aria-label={t.app.navigation}
+            >
+              <ul className="flex flex-col">
+                {sidebarNav.coreItems.map((item) => (
+                  <SidebarNavLink
+                    closeMobile={closeMobile}
+                    item={item}
+                    key={item.path}
+                    t={t}
+                  />
+                ))}
+              </ul>
+
+              {sidebarNav.pluginItems.length > 0 && (
+                <div
+                  aria-labelledby="hermes-sidebar-plugin-nav-heading"
+                  className="flex flex-col border-t border-current/10 pb-2"
+                  role="group"
+                >
+                  <span
+                    className={cn(
+                      "px-5 pt-2.5 pb-1",
+                      "font-mondwest text-[0.6rem] tracking-[0.15em] uppercase opacity-30",
+                    )}
+                    id="hermes-sidebar-plugin-nav-heading"
+                  >
+                    {t.app.pluginNavSection}
+                  </span>
+
+                  <ul className="flex flex-col">
+                    {sidebarNav.pluginItems.map((item) => (
+                      <SidebarNavLink
+                        closeMobile={closeMobile}
+                        item={item}
+                        key={item.path}
+                        t={t}
+                      />
+                    ))}
+                  </ul>
+                </div>
+              )}
+            </nav>
+
+            <SidebarSystemActions onNavigate={closeMobile} />
+
+            <div
+              className={cn(
+                "flex shrink-0 items-center justify-between gap-2",
+                "px-3 py-2",
+                "border-t border-current/20",
+              )}
+            >
+              <div className="flex min-w-0 items-center gap-2">
+                <PluginSlot name="header-right" />
+                <ThemeSwitcher dropUp />
+                <LanguageSwitcher />
+              </div>
+            </div>
+
+            <SidebarFooter />
+          </aside>
+
+          <PageHeaderProvider pluginTabs={pluginTabMeta}>
+            <div
+              className={cn(
+                "relative z-2 flex min-w-0 min-h-0 flex-1 flex-col",
+                "px-3 sm:px-6",
+                isChatRoute
+                  ? "pb-3 pt-1 sm:pb-4 sm:pt-2 lg:pt-4"
+                  : "pt-2 sm:pt-4 lg:pt-6 pb-4 sm:pb-8",
+                isDocsRoute && "min-h-0 flex-1",
+              )}
+            >
+              <PluginSlot name="pre-main" />
+              <div
+                className={cn(
+                  "w-full min-w-0",
+                  (isDocsRoute || isChatRoute) &&
+                    "min-h-0 flex flex-1 flex-col",
+                )}
+              >
+                <Routes>
+                  {routes.map(({ key, path, element }) => (
+                    <Route key={key} path={path} element={element} />
+                  ))}
+                  <Route
+                    path="*"
+                    element={
+                      <UnknownRouteFallback pluginsLoading={pluginsLoading} />
+                    }
+                  />
+                </Routes>
+
+                {embeddedChat &&
+                  !chatOverriddenByPlugin &&
+                  (pluginsLoading ? (
+                    isChatRoute ? (
+                      <div
+                        className="flex min-h-0 min-w-0 flex-1 items-center justify-center"
+                        aria-busy="true"
+                        aria-live="polite"
+                      >
+                        <div className="flex items-center gap-2 text-sm text-muted-foreground">
+                          <Spinner />
+                          <span>Loading chat…</span>
+                        </div>
+                      </div>
+                    ) : null
+                  ) : (
+                    <div
+                      data-chat-active={isChatRoute ? "true" : "false"}
+                      className={cn(
+                        "min-h-0 min-w-0",
+                        isChatRoute ? "flex flex-1 flex-col" : "hidden",
+                      )}
+                      aria-hidden={!isChatRoute}
+                    >
+                      <ChatPage isActive={isChatRoute} />
+                    </div>
+                  ))}
+              </div>
+              <PluginSlot name="post-main" />
+            </div>
+          </PageHeaderProvider>
+        </div>
+      </div>
+
+      <PluginSlot name="overlay" />
+    </div>
+  );
+}
+
+function SidebarNavLink({ closeMobile, item, t }: SidebarNavLinkProps) {
+  const { path, label, labelKey, icon: Icon } = item;
+
+  const navLabel = labelKey
+    ? ((t.app.nav as Record<string, string>)[labelKey] ?? label)
+    : label;
+
+  return (
+    <li>
+      <NavLink
+        to={path}
+        end={path === "/sessions"}
+        onClick={closeMobile}
+        className={({ isActive }) =>
+          cn(
+            "group relative flex items-center gap-3",
+            "px-5 py-2.5",
+            "font-mondwest text-[0.8rem] tracking-[0.12em]",
+            "whitespace-nowrap transition-colors cursor-pointer",
+            "focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-midground",
+            isActive ? "text-midground" : "opacity-60 hover:opacity-100",
+          )
+        }
+        style={{
+          clipPath: "var(--component-tab-clip-path)",
+        }}
+      >
+        {({ isActive }) => (
+          <>
+            <Icon className="h-3.5 w-3.5 shrink-0" />
+            <span className="truncate">{navLabel}</span>
+
+            <span
+              aria-hidden
+              className="absolute inset-y-0.5 left-1.5 right-1.5 bg-midground opacity-0 pointer-events-none transition-opacity duration-200 group-hover:opacity-5"
+            />
+
+            {isActive && (
+              <span
+                aria-hidden
+                className="absolute left-0 top-0 bottom-0 w-px bg-midground"
+                style={{ mixBlendMode: "plus-lighter" }}
+              />
+            )}
+          </>
+        )}
+      </NavLink>
+    </li>
+  );
+}
+
+function SidebarSystemActions({ onNavigate }: { onNavigate: () => void }) {
+  const { t } = useI18n();
+  const navigate = useNavigate();
+  const { activeAction, isBusy, isRunning, pendingAction, runAction } =
+    useSystemActions();
+
+  const items: SystemActionItem[] = [
+    {
+      action: "restart",
+      icon: RotateCw,
+      label: t.status.restartGateway,
+      runningLabel: t.status.restartingGateway,
+      spin: true,
+    },
+    {
+      action: "update",
+      icon: Download,
+      label: t.status.updateHermes,
+      runningLabel: t.status.updatingHermes,
+      spin: false,
+    },
+  ];
+
+  const handleClick = (action: SystemAction) => {
+    if (isBusy) return;
+    void runAction(action);
+    navigate("/sessions");
+    onNavigate();
+  };
+
+  return (
+    <div
+      className={cn(
+        "shrink-0 flex flex-col",
+        "border-t border-current/10",
+        "py-1",
+      )}
+    >
+      <span
+        className={cn(
+          "px-5 pt-0.5 pb-0.5",
+          "font-mondwest text-[0.6rem] tracking-[0.15em] uppercase opacity-30",
+        )}
+      >
+        {t.app.system}
+      </span>
+
+      <SidebarStatusStrip />
+
+      <ul className="flex flex-col">
+        {items.map(({ action, icon: Icon, label, runningLabel, spin }) => {
+          const isPending = pendingAction === action;
+          const isActionRunning =
+            activeAction === action && isRunning && !isPending;
+          const busy = isPending || isActionRunning;
+          const displayLabel = isActionRunning ? runningLabel : label;
+          const disabled = isBusy && !busy;
+
+          return (
+            <li key={action}>
+              <ListItem
+                onClick={() => handleClick(action)}
+                disabled={disabled}
+                aria-busy={busy}
+                active={busy}
+                className={cn(
+                  "gap-3 px-5 py-1.5 whitespace-nowrap",
+                  "font-mondwest text-[0.75rem] tracking-[0.1em]",
+                  "transition-opacity",
+                  busy
+                    ? "text-midground opacity-100"
+                    : "opacity-60 hover:opacity-100",
+                  "disabled:opacity-30",
+                )}
+              >
+                {isPending ? (
+                  <Spinner className="shrink-0 text-[0.875rem]" />
+                ) : isActionRunning && spin ? (
+                  <Spinner className="shrink-0 text-[0.875rem]" />
+                ) : (
+                  <Icon
+                    className={cn(
+                      "h-3.5 w-3.5 shrink-0",
+                      isActionRunning && !spin && "animate-pulse",
+                    )}
+                  />
+                )}
+
+                <span className="truncate">{displayLabel}</span>
+
+                <span
+                  aria-hidden
+                  className="absolute inset-y-0.5 left-1.5 right-1.5 bg-midground opacity-0 pointer-events-none transition-opacity duration-200 group-hover:opacity-5"
+                />
+
+                {busy && (
+                  <span
+                    aria-hidden
+                    className="absolute left-0 top-0 bottom-0 w-px bg-midground"
+                    style={{ mixBlendMode: "plus-lighter" }}
+                  />
+                )}
+              </ListItem>
+            </li>
+          );
+        })}
+      </ul>
+    </div>
+  );
+}
+
+interface NavItem {
+  icon: ComponentType<{ className?: string }>;
+  label: string;
+  labelKey?: string;
+  path: string;
+}
+
+interface SidebarNavLinkProps {
+  closeMobile: () => void;
+  item: NavItem;
+  t: Translations;
+}
+
+interface SystemActionItem {
+  action: SystemAction;
+  icon: ComponentType<{ className?: string }>;
+  label: string;
+  runningLabel: string;
+  spin: boolean;
+}

apps/dashboard/src/components/AutoField.tsx

@@ -1,7 +1,7 @@
+import { Select, SelectOption } from "@nous-research/ui/ui/components/select";
+import { Switch } from "@nous-research/ui/ui/components/switch";
 import { Input } from "@/components/ui/input";
 import { Label } from "@/components/ui/label";
-import { Select, SelectOption } from "@/components/ui/select";
-import { Switch } from "@/components/ui/switch";
 
 function FieldHint({ schema, schemaKey }: { schema: Record<string, unknown>; schemaKey: string }) {
   const keyPath = schemaKey.includes(".") ? schemaKey : "";

apps/dashboard/src/components/Backdrop.tsx

@@ -38,11 +38,25 @@ export function Backdrop() {
       <div
         aria-hidden
         className="pointer-events-none fixed inset-0 z-[2]"
-        style={{ mixBlendMode: "difference", opacity: 0.033 }}
+        style={
+          {
+            // Themes can override the filler background by setting
+            // `assets.bg` — the <img> hides itself when a CSS bg is set
+            // so the two don't double-darken. CSS var fallbacks keep the
+            // default behaviour unchanged when no theme customises these.
+            mixBlendMode:
+              "var(--component-backdrop-filler-blend-mode, difference)",
+            opacity: "var(--component-backdrop-filler-opacity, 0.033)",
+            backgroundImage: "var(--theme-asset-bg)",
+            backgroundSize: "var(--component-backdrop-background-size, cover)",
+            backgroundPosition:
+              "var(--component-backdrop-background-position, center)",
+          } as unknown as React.CSSProperties
+        }
       >
         <img
           alt=""
-          className="h-[150dvh] w-auto min-w-[100dvw] object-cover object-top-left invert"
+          className="h-[150dvh] w-auto min-w-[100dvw] object-cover object-top-left invert theme-default-filler"
           fetchPriority="low"
           src="/ds-assets/filler-bg0.jpg"
         />

apps/dashboard/src/components/ChatSidebar.tsx

@@ -0,0 +1,384 @@
+/**
+ * ChatSidebar — structured-events panel that sits next to the xterm.js
+ * terminal in the dashboard Chat tab.
+ *
+ * Two WebSockets, one per concern:
+ *
+ *   1. **JSON-RPC sidecar** (`GatewayClient` → /api/ws) — drives the
+ *      sidebar's own slot of the dashboard's in-process gateway.  Owns
+ *      the model badge / picker / connection state / error banner.
+ *      Independent of the PTY pane's session by design — those are the
+ *      pieces the sidebar needs to be able to drive directly (model
+ *      switch via slash.exec, etc.).
+ *
+ *   2. **Event subscriber** (/api/events?channel=…) — passive, receives
+ *      every dispatcher emit from the PTY-side `tui_gateway.entry` that
+ *      the dashboard fanned out.  This is how `tool.start/progress/
+ *      complete` from the agent loop reach the sidebar even though the
+ *      PTY child runs three processes deep from us.  The `channel` id
+ *      ties this listener to the same chat tab's PTY child — see
+ *      `ChatPage.tsx` for where the id is generated.
+ *
+ * Best-effort throughout: WS failures show in the badge / banner, the
+ * terminal pane keeps working unimpaired.
+ */
+
+import { Button } from "@nous-research/ui/ui/components/button";
+import { Badge } from "@nous-research/ui/ui/components/badge";
+import { Card } from "@/components/ui/card";
+
+import { ModelPickerDialog } from "@/components/ModelPickerDialog";
+import { ToolCall, type ToolEntry } from "@/components/ToolCall";
+import { GatewayClient, type ConnectionState } from "@/lib/gatewayClient";
+
+import { cn } from "@/lib/utils";
+import { AlertCircle, ChevronDown, RefreshCw } from "lucide-react";
+import { useCallback, useEffect, useMemo, useState } from "react";
+
+interface SessionInfo {
+  cwd?: string;
+  model?: string;
+  provider?: string;
+  credential_warning?: string;
+}
+
+interface RpcEnvelope {
+  method?: string;
+  params?: { type?: string; payload?: unknown };
+}
+
+const TOOL_LIMIT = 20;
+
+const STATE_LABEL: Record<ConnectionState, string> = {
+  idle: "idle",
+  connecting: "connecting",
+  open: "live",
+  closed: "closed",
+  error: "error",
+};
+
+const STATE_TONE: Record<
+  ConnectionState,
+  "secondary" | "warning" | "success" | "destructive"
+> = {
+  idle: "secondary",
+  connecting: "warning",
+  open: "success",
+  closed: "secondary",
+  error: "destructive",
+};
+
+interface ChatSidebarProps {
+  channel: string;
+  className?: string;
+}
+
+export function ChatSidebar({ channel, className }: ChatSidebarProps) {
+  // `version` bumps on reconnect; gw is derived so we never call setState
+  // for it inside an effect (React 19's set-state-in-effect rule). The
+  // counter is the dependency on purpose — it's not read in the memo body,
+  // it's the signal that says "rebuild the client".
+  const [version, setVersion] = useState(0);
+  // eslint-disable-next-line react-hooks/exhaustive-deps
+  const gw = useMemo(() => new GatewayClient(), [version]);
+
+  const [state, setState] = useState<ConnectionState>("idle");
+  const [sessionId, setSessionId] = useState<string | null>(null);
+  const [info, setInfo] = useState<SessionInfo>({});
+  const [tools, setTools] = useState<ToolEntry[]>([]);
+  const [modelOpen, setModelOpen] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+
+  useEffect(() => {
+    let cancelled = false;
+    const offState = gw.onState(setState);
+
+    const offSessionInfo = gw.on<SessionInfo>("session.info", (ev) => {
+      if (ev.session_id) {
+        setSessionId(ev.session_id);
+      }
+
+      if (ev.payload) {
+        setInfo((prev) => ({ ...prev, ...ev.payload }));
+      }
+    });
+
+    const offError = gw.on<{ message?: string }>("error", (ev) => {
+      const message = ev.payload?.message;
+
+      if (message) {
+        setError(message);
+      }
+    });
+
+    // Adopt whichever session the gateway hands us. session.create on the
+    // sidecar is independent of the PTY pane's session by design — we
+    // only need a sid to drive the model picker's slash.exec calls.
+    gw.connect()
+      .then(() => {
+        if (cancelled) {
+          return;
+        }
+        return gw.request<{ session_id: string }>("session.create", {});
+      })
+      .then((created) => {
+        if (cancelled || !created?.session_id) {
+          return;
+        }
+        setSessionId(created.session_id);
+      })
+      .catch((e: Error) => {
+        if (!cancelled) {
+          setError(e.message);
+        }
+      });
+
+    return () => {
+      cancelled = true;
+      offState();
+      offSessionInfo();
+      offError();
+      gw.close();
+    };
+  }, [gw]);
+
+  // Event subscriber WebSocket — receives the rebroadcast of every
+  // dispatcher emit from the PTY child's gateway.  See /api/pub +
+  // /api/events in hermes_cli/web_server.py for the broadcast hop.
+  //
+  // Failures (auth/loopback rejection, server too old to expose the
+  // endpoint, transient drops) surface in the same banner as the
+  // JSON-RPC sidecar so the sidebar matches its documented best-effort
+  // UX and the user always has a reconnect affordance.
+  useEffect(() => {
+    const token = window.__HERMES_SESSION_TOKEN__;
+
+    if (!token || !channel) {
+      return;
+    }
+
+    const proto = window.location.protocol === "https:" ? "wss:" : "ws:";
+    const qs = new URLSearchParams({ token, channel });
+    const ws = new WebSocket(
+      `${proto}//${window.location.host}/api/events?${qs.toString()}`,
+    );
+
+    // `unmounting` suppresses the banner during cleanup — `ws.close()`
+    // from the effect's return fires a close event with code 1005 that
+    // would otherwise look like an unexpected drop.
+    const DISCONNECTED = "events feed disconnected — tool calls may not appear";
+    let unmounting = false;
+    const surface = (msg: string) => !unmounting && setError(msg);
+
+    ws.addEventListener("error", () => surface(DISCONNECTED));
+
+    ws.addEventListener("close", (ev) => {
+      if (ev.code === 4401 || ev.code === 4403) {
+        surface(`events feed rejected (${ev.code}) — reload the page`);
+      } else if (ev.code !== 1000) {
+        surface(DISCONNECTED);
+      }
+    });
+
+    ws.addEventListener("message", (ev) => {
+      let frame: RpcEnvelope;
+
+      try {
+        frame = JSON.parse(ev.data);
+      } catch {
+        return;
+      }
+
+      if (frame.method !== "event" || !frame.params) {
+        return;
+      }
+
+      const { type, payload } = frame.params;
+
+      if (type === "tool.start") {
+        const p = payload as
+          | { tool_id?: string; name?: string; context?: string }
+          | undefined;
+        const toolId = p?.tool_id;
+
+        if (!toolId) {
+          return;
+        }
+
+        setTools((prev) =>
+          [
+            ...prev,
+            {
+              kind: "tool" as const,
+              id: `tool-${toolId}-${prev.length}`,
+              tool_id: toolId,
+              name: p?.name ?? "tool",
+              context: p?.context,
+              status: "running" as const,
+              startedAt: Date.now(),
+            },
+          ].slice(-TOOL_LIMIT),
+        );
+      } else if (type === "tool.progress") {
+        const p = payload as
+          | { name?: string; preview?: string }
+          | undefined;
+
+        if (!p?.name || !p.preview) {
+          return;
+        }
+
+        setTools((prev) =>
+          prev.map((t) =>
+            t.status === "running" && t.name === p.name
+              ? { ...t, preview: p.preview }
+              : t,
+          ),
+        );
+      } else if (type === "tool.complete") {
+        const p = payload as
+          | {
+              tool_id?: string;
+              summary?: string;
+              error?: string;
+              inline_diff?: string;
+            }
+          | undefined;
+
+        if (!p?.tool_id) {
+          return;
+        }
+
+        setTools((prev) =>
+          prev.map((t) =>
+            t.tool_id === p.tool_id
+              ? {
+                  ...t,
+                  status: p.error ? "error" : "done",
+                  summary: p.summary,
+                  error: p.error,
+                  inline_diff: p.inline_diff,
+                  completedAt: Date.now(),
+                }
+              : t,
+          ),
+        );
+      }
+    });
+
+    return () => {
+      unmounting = true;
+      ws.close();
+    };
+  }, [channel, version]);
+
+  const reconnect = useCallback(() => {
+    setError(null);
+    setTools([]);
+    setVersion((v) => v + 1);
+  }, []);
+
+  // Picker hands us a fully-formed slash command (e.g. "/model anthropic/...").
+  // Fire-and-forget through `slash.exec`; the TUI pane will render the result
+  // via PTY, so the sidebar doesn't need to surface output of its own.
+  const onModelSubmit = useCallback(
+    (slashCommand: string) => {
+      if (!sessionId) {
+        return;
+      }
+
+      void gw.request("slash.exec", {
+        session_id: sessionId,
+        command: slashCommand,
+      });
+      setModelOpen(false);
+    },
+    [gw, sessionId],
+  );
+
+  const canPickModel = state === "open" && !!sessionId;
+  const modelLabel = (info.model ?? "—").split("/").slice(-1)[0] ?? "—";
+  const banner = error ?? info.credential_warning ?? null;
+
+  return (
+    <aside
+      className={cn(
+        "flex h-full w-full min-w-0 shrink-0 flex-col gap-3 normal-case lg:w-80",
+        className,
+      )}
+    >
+      <Card className="flex items-center justify-between gap-2 px-3 py-2">
+        <div className="min-w-0">
+          <div className="text-xs uppercase tracking-wider text-muted-foreground">
+            model
+          </div>
+
+          <Button
+            ghost
+            size="sm"
+            disabled={!canPickModel}
+            onClick={() => setModelOpen(true)}
+            suffix={
+              canPickModel ? (
+                <ChevronDown className="opacity-60" />
+              ) : undefined
+            }
+            className="self-start min-w-0 px-0 py-0 normal-case tracking-normal text-sm font-medium hover:underline disabled:no-underline"
+            title={info.model ?? "switch model"}
+          >
+            <span className="truncate">{modelLabel}</span>
+          </Button>
+        </div>
+
+        <Badge tone={STATE_TONE[state]}>{STATE_LABEL[state]}</Badge>
+      </Card>
+
+      {banner && (
+        <Card className="flex items-start gap-2 border-destructive/40 bg-destructive/5 px-3 py-2 text-xs">
+          <AlertCircle className="mt-0.5 h-3.5 w-3.5 shrink-0 text-destructive" />
+
+          <div className="min-w-0 flex-1">
+            <div className="wrap-break-word text-destructive">{banner}</div>
+
+            {error && (
+              <Button
+                size="sm"
+                outlined
+                className="mt-1"
+                onClick={reconnect}
+                prefix={<RefreshCw />}
+              >
+                reconnect
+              </Button>
+            )}
+          </div>
+        </Card>
+      )}
+
+      <Card className="flex min-h-0 flex-1 flex-col px-2 py-2">
+        <div className="px-1 pb-2 text-xs uppercase tracking-wider text-muted-foreground">
+          tools
+        </div>
+
+        <div className="flex min-h-0 flex-1 flex-col gap-1.5 overflow-y-auto pr-1">
+          {tools.length === 0 ? (
+            <div className="px-2 py-4 text-center text-xs text-muted-foreground">
+              no tool calls yet
+            </div>
+          ) : (
+            tools.map((t) => <ToolCall key={t.id} tool={t} />)
+          )}
+        </div>
+      </Card>
+
+      {modelOpen && canPickModel && sessionId && (
+        <ModelPickerDialog
+          gw={gw}
+          sessionId={sessionId}
+          onClose={() => setModelOpen(false)}
+          onSubmit={onModelSubmit}
+        />
+      )}
+    </aside>
+  );
+}

apps/dashboard/src/components/DeleteConfirmDialog.tsx

@@ -0,0 +1,40 @@
+import { ConfirmDialog } from "@/components/ui/confirm-dialog";
+import { useI18n } from "@/i18n";
+
+export function DeleteConfirmDialog({
+  cancelLabel,
+  confirmLabel,
+  description,
+  loading,
+  onCancel,
+  onConfirm,
+  open,
+  title,
+}: DeleteConfirmDialogProps) {
+  const { t } = useI18n();
+
+  return (
+    <ConfirmDialog
+      open={open}
+      onCancel={onCancel}
+      onConfirm={onConfirm}
+      title={title}
+      description={description}
+      loading={loading}
+      destructive
+      confirmLabel={confirmLabel ?? t.common.delete}
+      cancelLabel={cancelLabel ?? t.common.cancel}
+    />
+  );
+}
+
+interface DeleteConfirmDialogProps {
+  cancelLabel?: string;
+  confirmLabel?: string;
+  description?: string;
+  loading: boolean;
+  onCancel: () => void;
+  onConfirm: () => void;
+  open: boolean;
+  title: string;
+}