fix(tui): restore resumed transcript lineage

hermes_state.py

@@ -1132,20 +1132,29 @@ class SessionDB:
                 current = child_id
         return session_id
 
-    def get_messages_as_conversation(self, session_id: str) -> List[Dict[str, Any]]:
+    def get_messages_as_conversation(
+        self, session_id: str, include_ancestors: bool = False
+    ) -> List[Dict[str, Any]]:
         """
         Load messages in the OpenAI conversation format (role + content dicts).
         Used by the gateway to restore conversation history.
         """
+        session_ids = [session_id]
+        if include_ancestors:
+            session_ids = self._session_lineage_root_to_tip(session_id)
+
         with self._lock:
-            cursor = self._conn.execute(
-                "SELECT role, content, tool_call_id, tool_calls, tool_name, "
-                "reasoning, reasoning_content, reasoning_details, codex_reasoning_items, "
-                "codex_message_items "
-                "FROM messages WHERE session_id = ? ORDER BY timestamp, id",
-                (session_id,),
-            )
-            rows = cursor.fetchall()
+            rows = []
+            for sid in session_ids:
+                cursor = self._conn.execute(
+                    "SELECT role, content, tool_call_id, tool_calls, tool_name, "
+                    "reasoning, reasoning_content, reasoning_details, codex_reasoning_items, "
+                    "codex_message_items "
+                    "FROM messages WHERE session_id = ? ORDER BY timestamp, id",
+                    (sid,),
+                )
+                rows.extend(cursor.fetchall())
+
         messages = []
         for row in rows:
             msg = {"role": row["role"], "content": row["content"]}
@@ -1185,9 +1194,47 @@ class SessionDB:
                     except (json.JSONDecodeError, TypeError):
                         logger.warning("Failed to deserialize codex_message_items, falling back to None")
                         msg["codex_message_items"] = None
+            if include_ancestors and self._is_duplicate_replayed_user_message(messages, msg):
+                continue
             messages.append(msg)
         return messages
 
+    def _session_lineage_root_to_tip(self, session_id: str) -> List[str]:
+        if not session_id:
+            return [session_id]
+
+        chain = []
+        current = session_id
+        seen = set()
+        with self._lock:
+            for _ in range(100):
+                if not current or current in seen:
+                    break
+                seen.add(current)
+                chain.append(current)
+                row = self._conn.execute(
+                    "SELECT parent_session_id FROM sessions WHERE id = ?",
+                    (current,),
+                ).fetchone()
+                if row is None:
+                    break
+                current = row["parent_session_id"] if hasattr(row, "keys") else row[0]
+        return list(reversed(chain)) or [session_id]
+
+    @staticmethod
+    def _is_duplicate_replayed_user_message(messages: List[Dict[str, Any]], msg: Dict[str, Any]) -> bool:
+        if msg.get("role") != "user":
+            return False
+        content = msg.get("content")
+        if not isinstance(content, str) or not content:
+            return False
+        for prev in reversed(messages):
+            if prev.get("role") == "user" and prev.get("content") == content:
+                return True
+            if prev.get("role") == "assistant" and (prev.get("content") or prev.get("tool_calls")):
+                return False
+        return False
+
     # =========================================================================
     # Search
     # =========================================================================

tests/test_hermes_state.py

@@ -222,6 +222,35 @@ class TestMessageStorage:
         assert conv[0] == {"role": "user", "content": "Hello"}
         assert conv[1] == {"role": "assistant", "content": "Hi!"}
 
+    def test_get_messages_as_conversation_includes_ancestor_chain(self, db):
+        db.create_session("root", "tui")
+        db.append_message("root", role="user", content="first prompt")
+        db.append_message("root", role="assistant", content="first answer")
+        db.create_session("child", "tui", parent_session_id="root")
+        db.append_message("child", role="user", content="second prompt")
+        db.append_message("child", role="assistant", content="second answer")
+
+        conv = db.get_messages_as_conversation("child", include_ancestors=True)
+
+        assert [m["content"] for m in conv] == [
+            "first prompt",
+            "first answer",
+            "second prompt",
+            "second answer",
+        ]
+
+    def test_get_messages_as_conversation_avoids_repeated_resume_prompts_from_ancestors(self, db):
+        db.create_session("root", "tui")
+        db.append_message("root", role="user", content="same prompt")
+        db.append_message("root", role="user", content="same prompt")
+        db.append_message("root", role="assistant", content="answer")
+        db.create_session("child", "tui", parent_session_id="root")
+        db.append_message("child", role="user", content="next prompt")
+
+        conv = db.get_messages_as_conversation("child", include_ancestors=True)
+
+        assert [m["content"] for m in conv if m["role"] == "user"] == ["same prompt", "next prompt"]
+
     def test_finish_reason_stored(self, db):
         db.create_session(session_id="s1", source="cli")
         db.append_message("s1", role="assistant", content="Done", finish_reason="stop")

tests/test_tui_gateway_server.py

@@ -59,6 +59,69 @@ def test_write_json_returns_false_on_broken_pipe(monkeypatch):
     assert server.write_json({"ok": True}) is False
 
 
+def test_history_to_messages_preserves_tool_calls_for_resume_display():
+    history = [
+        {"role": "user", "content": "first prompt"},
+        {
+            "role": "assistant",
+            "content": "",
+            "tool_calls": [
+                {
+                    "id": "call_1",
+                    "function": {
+                        "name": "search_files",
+                        "arguments": json.dumps({"pattern": "resume"}),
+                    },
+                }
+            ],
+        },
+        {"role": "tool", "content": "{}", "tool_call_id": "call_1"},
+        {"role": "assistant", "content": "first answer"},
+        {"role": "user", "content": "second prompt"},
+    ]
+
+    assert server._history_to_messages(history) == [
+        {"role": "user", "text": "first prompt"},
+        {"context": "resume", "name": "search_files", "role": "tool"},
+        {"role": "assistant", "text": "first answer"},
+        {"role": "user", "text": "second prompt"},
+    ]
+
+
+def test_session_resume_uses_parent_lineage_for_display(monkeypatch):
+    captured = {}
+
+    class FakeDB:
+        def get_session(self, target):
+            return {"id": target}
+
+        def reopen_session(self, target):
+            captured["reopened"] = target
+
+        def get_messages_as_conversation(self, target, include_ancestors=False):
+            captured.setdefault("history_calls", []).append((target, include_ancestors))
+            return [
+                {"role": "user", "content": "root prompt"},
+                {"role": "assistant", "content": "root answer"},
+            ] if include_ancestors else [{"role": "user", "content": "tip prompt"}]
+
+    monkeypatch.setattr(server, "_get_db", lambda: FakeDB())
+    monkeypatch.setattr(server, "_enable_gateway_prompts", lambda: None)
+    monkeypatch.setattr(server, "_set_session_context", lambda target: [])
+    monkeypatch.setattr(server, "_clear_session_context", lambda tokens: None)
+    monkeypatch.setattr(server, "_make_agent", lambda *args, **kwargs: types.SimpleNamespace(model="test"))
+    monkeypatch.setattr(server, "_session_info", lambda agent: {"model": "test", "tools": {}, "skills": {}})
+    monkeypatch.setattr(server, "_init_session", lambda sid, key, agent, history, cols=80: None)
+
+    resp = server.handle_request({"id": "1", "method": "session.resume", "params": {"session_id": "tip"}})
+
+    assert resp["result"]["messages"] == [
+        {"role": "user", "text": "root prompt"},
+        {"role": "assistant", "text": "root answer"},
+    ]
+    assert captured["history_calls"] == [("tip", False), ("tip", True)]
+
+
 def test_status_callback_emits_kind_and_text():
     with patch("tui_gateway.server._emit") as emit:
         cb = server._agent_cbs("sid")["status_callback"]

tui_gateway/server.py

@@ -913,8 +913,16 @@ def _probe_config_health(cfg: dict) -> str:
 
 
 def _session_info(agent) -> dict:
+    reasoning_config = getattr(agent, "reasoning_config", None)
+    reasoning_effort = ""
+    if isinstance(reasoning_config, dict) and reasoning_config.get("enabled") is not False:
+        reasoning_effort = str(reasoning_config.get("effort", "") or "")
+    service_tier = getattr(agent, "service_tier", None) or ""
     info: dict = {
         "model": getattr(agent, "model", ""),
+        "reasoning_effort": reasoning_effort,
+        "service_tier": service_tier,
+        "fast": service_tier == "priority",
         "tools": {},
         "skills": {},
         "cwd": os.getcwd(),
@@ -1013,7 +1021,7 @@ def _tool_summary(name: str, result: str, duration_s: float | None) -> str | Non
         if n is not None:
             text = f"Extracted {n} {'page' if n == 1 else 'pages'}"
 
-    return f"{text or 'Completed'}{suffix}" if (text or dur) else None
+    return f"{text}{suffix}" if text else None
 
 
 def _on_tool_start(sid: str, tool_call_id: str, name: str, args: dict):
@@ -1029,10 +1037,13 @@ def _on_tool_start(sid: str, tool_call_id: str, name: str, args: dict):
             pass
         session.setdefault("tool_started_at", {})[tool_call_id] = time.time()
     if _tool_progress_enabled(sid):
+        payload = {"tool_id": tool_call_id, "name": name, "context": _tool_ctx(name, args)}
+        if name == "todo" and isinstance(args, dict) and isinstance(args.get("todos"), list):
+            payload["todos"] = args.get("todos")
         _emit(
             "tool.start",
             sid,
-            {"tool_id": tool_call_id, "name": name, "context": _tool_ctx(name, args)},
+            payload,
         )
 
 
@@ -1050,6 +1061,13 @@ def _on_tool_complete(sid: str, tool_call_id: str, name: str, args: dict, result
     summary = _tool_summary(name, result, duration_s)
     if summary:
         payload["summary"] = summary
+    if name == "todo":
+        try:
+            data = json.loads(result)
+            if isinstance(data, dict) and isinstance(data.get("todos"), list):
+                payload["todos"] = data.get("todos")
+        except Exception:
+            pass
     try:
         from agent.display import render_edit_diff_with_delta
 
@@ -1698,7 +1716,8 @@ def _(rid, params: dict) -> dict:
     try:
         db.reopen_session(target)
         history = db.get_messages_as_conversation(target)
-        messages = _history_to_messages(history)
+        display_history = db.get_messages_as_conversation(target, include_ancestors=True)
+        messages = _history_to_messages(display_history)
         tokens = _set_session_context(target)
         try:
             agent = _make_agent(sid, target, session_id=target)
@@ -1746,11 +1765,20 @@ def _(rid, params: dict) -> dict:
 @method("session.history")
 def _(rid, params: dict) -> dict:
     session, err = _sess(params, rid)
-    return err or _ok(
+    if err:
+        return err
+    history = list(session.get("history", []))
+    db = _get_db()
+    if db is not None and session.get("session_key"):
+        try:
+            history = db.get_messages_as_conversation(session["session_key"], include_ancestors=True)
+        except Exception:
+            pass
+    return _ok(
         rid,
         {
             "count": len(session.get("history", [])),
-            "messages": _history_to_messages(list(session.get("history", []))),
+            "messages": _history_to_messages(history),
         },
     )
 

ui-tui/src/__tests__/messages.test.ts

@@ -1,7 +1,26 @@
 import { describe, expect, it } from 'vitest'
 
+import { toTranscriptMessages } from '../domain/messages.js'
 import { upsert } from '../lib/messages.js'
 
+describe('toTranscriptMessages', () => {
+  it('preserves assistant tool-call rows so resume does not drop prior turns', () => {
+    const rows = [
+      { role: 'user', text: 'first prompt' },
+      { role: 'tool', context: 'repo', name: 'search_files', text: 'ignored raw result' },
+      { role: 'assistant', text: 'first answer' },
+      { role: 'user', text: 'second prompt' }
+    ]
+
+    expect(toTranscriptMessages(rows).map(msg => [msg.role, msg.text])).toEqual([
+      ['user', 'first prompt'],
+      ['assistant', 'first answer'],
+      ['user', 'second prompt']
+    ])
+    expect(toTranscriptMessages(rows)[1]?.tools?.[0]).toContain('Search Files')
+  })
+})
+
 describe('upsert', () => {
   it('appends when last role differs', () => {
     expect(upsert([{ role: 'user', text: 'hi' }], 'assistant', 'hello')).toHaveLength(2)

ui-tui/src/__tests__/text.test.ts

@@ -1,14 +1,18 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  boundedLiveRenderText,
+  buildToolTrailLine,
   edgePreview,
   estimateRows,
   estimateTokensRough,
   fmtK,
   isToolTrailResultLine,
   lastCotTrailIndex,
+  parseToolTrailResultLine,
   pasteTokenLabel,
-  sameToolTrailGroup
+  sameToolTrailGroup,
+  splitToolDuration
 } from '../lib/text.js'
 
 describe('isToolTrailResultLine', () => {
@@ -19,6 +23,16 @@ describe('isToolTrailResultLine', () => {
   })
 })
 
+describe('buildToolTrailLine', () => {
+  it('puts completion duration inline before the result marker', () => {
+    const line = buildToolTrailLine('read_file', 'x', false, '', 0.94)
+
+    expect(line).toBe('Read File("x") (0.9s) ✓')
+    expect(parseToolTrailResultLine(line)).toEqual({ call: 'Read File("x") (0.9s)', detail: '', mark: '✓' })
+    expect(splitToolDuration('Read File("x") (0.9s)')).toEqual({ label: 'Read File("x")', duration: ' (0.9s)' })
+  })
+})
+
 describe('lastCotTrailIndex', () => {
   it('finds last non-result line', () => {
     expect(lastCotTrailIndex(['a ✓', 'thinking…'])).toBe(1)
@@ -68,6 +82,28 @@ describe('estimateTokensRough', () => {
   })
 })
 
+describe('boundedLiveRenderText', () => {
+  it('preserves short live text verbatim', () => {
+    expect(boundedLiveRenderText('one\ntwo', { maxChars: 100, maxLines: 10 })).toBe('one\ntwo')
+  })
+
+  it('keeps the live tail by character budget', () => {
+    const out = boundedLiveRenderText('abcdefghij', { maxChars: 4, maxLines: 10 })
+
+    expect(out).toContain('ghij')
+    expect(out).toContain('omitted')
+    expect(out).not.toContain('abcdef')
+  })
+
+  it('keeps the live tail by line budget', () => {
+    const out = boundedLiveRenderText(['a', 'b', 'c', 'd'].join('\n'), { maxChars: 100, maxLines: 2 })
+
+    expect(out).toContain('c\nd')
+    expect(out).toContain('omitted 2 lines')
+    expect(out).not.toContain('a\nb')
+  })
+})
+
 describe('edgePreview', () => {
   it('keeps both ends for long text', () => {
     expect(edgePreview('Vampire Bondage ropes slipped from her neck, still stained with blood', 8, 18)).toBe(

ui-tui/src/app/turnController.ts

@@ -2,18 +2,20 @@ import {
   REASONING_PULSE_MS,
   STREAM_BATCH_MS,
   STREAM_IDLE_BATCH_MS,
+  STREAM_SCROLL_BATCH_MS,
   STREAM_TYPING_BATCH_MS
 } from '../config/timing.js'
 import type { SessionInterruptResponse, SubagentEventPayload } from '../gatewayTypes.js'
 import { hasReasoningTag, splitReasoning } from '../lib/reasoning.js'
 import {
+  boundedLiveRenderText,
   buildToolTrailLine,
   estimateTokensRough,
   isTransientTrailLine,
   sameToolTrailGroup,
   toolTrailLabel
 } from '../lib/text.js'
-import type { ActiveTool, ActivityItem, Msg, SubagentProgress } from '../types.js'
+import type { ActiveTool, ActivityItem, Msg, SubagentProgress, TodoItem } from '../types.js'
 
 import { resetFlowOverlays } from './overlayStore.js'
 import { pushSnapshot } from './spawnHistoryStore.js'
@@ -40,7 +42,52 @@ const diffSegmentBody = (msg: Msg): null | string => {
 
 const hasDetails = (msg: Msg): boolean => Boolean(msg.thinking || msg.tools?.length || msg.toolTokens)
 
-const textSegments = (segments: Msg[]) => segments.filter(msg => msg.role === 'assistant' && msg.kind !== 'diff').map(msg => msg.text)
+const isToolOnly = (msg: Msg | undefined) =>
+  Boolean(msg && msg.kind === 'trail' && !msg.thinking?.trim() && !msg.text && msg.tools?.length)
+
+const mergeSequentialToolOnly = (segments: Msg[]) =>
+  segments.reduce<Msg[]>((acc, msg) => {
+    if (isToolOnly(msg) && isToolOnly(acc.at(-1))) {
+      const prev = acc.at(-1)!
+
+      return [...acc.slice(0, -1), { ...prev, tools: [...(prev.tools ?? []), ...(msg.tools ?? [])] }]
+    }
+
+    return [...acc, msg]
+  }, [])
+
+const isTodoStatus = (status: unknown): status is TodoItem['status'] =>
+  status === 'pending' || status === 'in_progress' || status === 'completed' || status === 'cancelled'
+
+const parseTodos = (value: unknown): null | TodoItem[] => {
+  if (!Array.isArray(value)) {
+    return null
+  }
+
+  return value
+    .map(item => {
+      if (!item || typeof item !== 'object') {
+        return null
+      }
+
+      const row = item as Record<string, unknown>
+      const status = row.status
+
+      if (!isTodoStatus(status)) {
+        return null
+      }
+
+      return {
+        content: String(row.content ?? '').trim(),
+        id: String(row.id ?? '').trim(),
+        status
+      }
+    })
+    .filter((item): item is TodoItem => Boolean(item?.id && item.content))
+}
+
+const textSegments = (segments: Msg[]) =>
+  segments.filter(msg => msg.role === 'assistant' && msg.kind !== 'diff').map(msg => msg.text)
 
 const finalTail = (finalText: string, segments: Msg[]) => {
   let tail = finalText
@@ -88,6 +135,7 @@ class TurnController {
   turnTools: string[] = []
 
   private activeTools: ActiveTool[] = []
+  private activeReasoningText = ''
   private reasoningSegmentIndex: null | number = null
   private activityId = 0
   private reasoningStreamingTimer: Timer = null
@@ -100,12 +148,18 @@ class TurnController {
     this.streamDelay = STREAM_TYPING_BATCH_MS
   }
 
+  boostStreamingForScroll() {
+    this.streamDelay = Math.max(this.streamDelay, STREAM_SCROLL_BATCH_MS)
+  }
+
   relaxStreaming() {
     this.streamDelay = STREAM_IDLE_BATCH_MS
   }
 
   clearReasoning() {
     this.reasoningTimer = clear(this.reasoningTimer)
+    this.activeReasoningText = ''
+    this.reasoningSegmentIndex = null
     this.reasoningText = ''
     this.toolTokenAcc = 0
     patchTurnState({ reasoning: '', reasoningTokens: 0, toolTokens: 0 })
@@ -144,6 +198,8 @@ class TurnController {
     this.interrupted = true
     gw.request<SessionInterruptResponse>('session.interrupt', { session_id: sid }).catch(() => {})
 
+    this.closeReasoningSegment()
+
     const segments = this.segmentMessages
     const partial = this.bufRef.trimStart()
     const tools = this.pendingSegmentTools
@@ -193,7 +249,7 @@ class TurnController {
   }
 
   private syncReasoningSegment() {
-    const thinking = this.reasoningText.trim()
+    const thinking = this.activeReasoningText.trim()
 
     if (!thinking) {
       return
@@ -205,8 +261,7 @@ class TurnController {
       text: '',
       thinking,
       thinkingTokens: estimateTokensRough(thinking),
-      toolTokens: this.toolTokenAcc || undefined,
-      ...(this.pendingSegmentTools.length && { tools: this.pendingSegmentTools })
+      toolTokens: this.toolTokenAcc || undefined
     }
 
     if (this.reasoningSegmentIndex === null) {
@@ -219,13 +274,40 @@ class TurnController {
     patchTurnState({ streamSegments: this.segmentMessages })
   }
 
+  private closeReasoningSegment() {
+    this.syncReasoningSegment()
+    this.activeReasoningText = ''
+    this.reasoningSegmentIndex = null
+  }
+
+  private pushSegment(msg: Msg) {
+    if (isToolOnly(msg) && isToolOnly(this.segmentMessages.at(-1)!)) {
+      const prev = this.segmentMessages.at(-1)!
+      this.segmentMessages = [
+        ...this.segmentMessages.slice(0, -1),
+        { ...prev, tools: [...(prev.tools ?? []), ...(msg.tools ?? [])] }
+      ]
+
+      return
+    }
+
+    this.segmentMessages = [...this.segmentMessages, msg]
+  }
+
   flushStreamingSegment() {
     const raw = this.bufRef.trimStart()
-    const split = raw ? (hasReasoningTag(raw) ? splitReasoning(raw) : { reasoning: '', text: raw }) : { reasoning: '', text: '' }
+
+    const split = raw
+      ? hasReasoningTag(raw)
+        ? splitReasoning(raw)
+        : { reasoning: '', text: raw }
+      : { reasoning: '', text: '' }
 
     if (split.reasoning && !this.reasoningText.trim()) {
       this.reasoningText = split.reasoning
+      this.activeReasoningText = split.reasoning
       patchTurnState({ reasoning: this.reasoningText, reasoningTokens: estimateTokensRough(this.reasoningText) })
+      this.syncReasoningSegment()
     }
 
     const msg: Msg = {
@@ -238,7 +320,7 @@ class TurnController {
     this.streamTimer = clear(this.streamTimer)
 
     if (split.text || hasDetails(msg)) {
-      this.segmentMessages = [...this.segmentMessages, msg]
+      this.pushSegment(msg)
     }
 
     this.pendingSegmentTools = []
@@ -256,6 +338,31 @@ class TurnController {
     }, REASONING_PULSE_MS)
   }
 
+  recordTodos(value: unknown) {
+    const todos = parseTodos(value)
+
+    if (todos !== null) {
+      patchTurnState({ todos })
+    }
+  }
+
+  private flushPendingToolsIntoLastSegment() {
+    const last = this.segmentMessages[this.segmentMessages.length - 1]
+
+    if (!this.pendingSegmentTools.length || !isToolOnly(last)) {
+      return false
+    }
+
+    this.segmentMessages = [
+      ...this.segmentMessages.slice(0, -1),
+      { ...last, tools: [...(last.tools ?? []), ...this.pendingSegmentTools] }
+    ]
+    this.pendingSegmentTools = []
+    patchTurnState({ streamPendingTools: [], streamSegments: this.segmentMessages })
+
+    return true
+  }
+
   pushInlineDiffSegment(diffText: string, tools: string[] = []) {
     // Strip CLI chrome the gateway emits before the unified diff (e.g. a
     // leading "┊ review diff" header written by `_emit_inline_diff` for the
@@ -283,7 +390,10 @@ class TurnController {
       return
     }
 
-    this.segmentMessages = [...this.segmentMessages, { kind: 'diff', role: 'assistant', text: block, ...(tools.length && { tools }) }]
+    this.segmentMessages = [
+      ...this.segmentMessages,
+      { kind: 'diff', role: 'assistant', text: block, ...(tools.length && { tools }) }
+    ]
     patchTurnState({ streamSegments: this.segmentMessages })
   }
 
@@ -328,13 +438,25 @@ class TurnController {
   }
 
   recordMessageComplete(payload: { rendered?: string; reasoning?: string; text?: string }) {
+    this.closeReasoningSegment()
+
     const rawText = (payload.rendered ?? payload.text ?? this.bufRef).trimStart()
     const split = splitReasoning(rawText)
     const finalText = finalTail(split.text, this.segmentMessages)
     const existingReasoning = this.reasoningText.trim() || String(payload.reasoning ?? '').trim()
     const savedReasoning = [existingReasoning, existingReasoning ? '' : split.reasoning].filter(Boolean).join('\n\n')
     const savedToolTokens = this.toolTokenAcc
-    const tools = this.pendingSegmentTools
+    let tools = this.pendingSegmentTools
+    const last = this.segmentMessages[this.segmentMessages.length - 1]
+
+    if (tools.length && isToolOnly(last)) {
+      this.segmentMessages = [
+        ...this.segmentMessages.slice(0, -1),
+        { ...last, tools: [...(last.tools ?? []), ...tools] }
+      ]
+      this.pendingSegmentTools = []
+      tools = []
+    }
 
     // Drop diff-only segments the agent is about to narrate in the final
     // reply. Without this, a closing "here's the diff …" message would
@@ -343,13 +465,19 @@ class TurnController {
     // assistant narration stays put.
     const finalHasOwnDiffFence = /```(?:diff|patch)\b/i.test(finalText)
 
-    const segments = this.segmentMessages.filter(msg => {
-      const body = diffSegmentBody(msg)
+    const segments = mergeSequentialToolOnly(
+      this.segmentMessages.filter(msg => {
+        const body = diffSegmentBody(msg)
 
-      return body === null || (!finalHasOwnDiffFence && !finalText.includes(body))
-    })
+        return body === null || (!finalHasOwnDiffFence && !finalText.includes(body))
+      })
+    )
+
+    const hasReasoningSegment =
+      this.reasoningSegmentIndex !== null || segments.some(msg => Boolean(msg.thinking?.trim()))
+
+    const finalThinking = hasReasoningSegment ? '' : savedReasoning.trim()
 
-    const finalThinking = savedReasoning.trim()
     const finalDetails: Msg = {
       kind: 'trail',
       role: 'system',
@@ -359,8 +487,8 @@ class TurnController {
       toolTokens: savedToolTokens || undefined,
       ...(tools.length && { tools })
     }
-    const hasReasoningSegment = this.reasoningSegmentIndex !== null
-    const finalMessages = hasDetails(finalDetails) && !hasReasoningSegment ? [...segments, finalDetails] : [...segments]
+
+    const finalMessages = hasDetails(finalDetails) ? [...segments, finalDetails] : [...segments]
 
     if (finalText) {
       finalMessages.push({ role: 'assistant', text: finalText })
@@ -387,6 +515,7 @@ class TurnController {
     this.turnTools = []
     this.persistedToolLabels.clear()
     this.bufRef = ''
+    this.interrupted = false
     patchTurnState({ activity: [], outcome: '' })
 
     return { finalMessages, finalText, wasInterrupted }
@@ -419,6 +548,7 @@ class TurnController {
     }
 
     this.reasoningText = incoming
+    this.activeReasoningText = incoming
     this.scheduleReasoning()
     this.syncReasoningSegment()
     this.pulseReasoningStreaming()
@@ -429,30 +559,63 @@ class TurnController {
       return
     }
 
+    if (!this.activeReasoningText.trim() && this.pendingSegmentTools.length) {
+      this.flushStreamingSegment()
+    }
+
     this.reasoningText += text
+    this.activeReasoningText += text
+
+    if (this.reasoningText.length > 80_000) {
+      this.reasoningText = this.reasoningText.slice(-60_000)
+    }
+
     this.scheduleReasoning()
     this.syncReasoningSegment()
     this.pulseReasoningStreaming()
   }
 
-  recordToolComplete(toolId: string, fallbackName?: string, error?: string, summary?: string) {
-    const line = this.completeTool(toolId, fallbackName, error, summary)
+  recordToolComplete(
+    toolId: string,
+    fallbackName?: string,
+    error?: string,
+    summary?: string,
+    duration?: number,
+    todos?: unknown
+  ) {
+    this.recordTodos(todos)
+    const line = this.completeTool(toolId, fallbackName, error, summary, duration)
 
     this.pendingSegmentTools = [...this.pendingSegmentTools, line]
+    this.flushPendingToolsIntoLastSegment()
     this.publishToolState()
   }
 
-  recordInlineDiffToolComplete(diffText: string, toolId: string, fallbackName?: string, error?: string) {
+  recordInlineDiffToolComplete(
+    diffText: string,
+    toolId: string,
+    fallbackName?: string,
+    error?: string,
+    duration?: number
+  ) {
     this.flushStreamingSegment()
-    this.pushInlineDiffSegment(diffText, [this.completeTool(toolId, fallbackName, error, '')])
+    this.pushInlineDiffSegment(diffText, [this.completeTool(toolId, fallbackName, error, '', duration)])
     this.publishToolState()
   }
 
-  private completeTool(toolId: string, fallbackName?: string, error?: string, summary?: string) {
+  private completeTool(toolId: string, fallbackName?: string, error?: string, summary?: string, duration?: number) {
     const done = this.activeTools.find(tool => tool.id === toolId)
     const name = done?.name ?? fallbackName ?? 'tool'
     const label = toolTrailLabel(name)
-    const line = buildToolTrailLine(name, done?.context || '', Boolean(error), error || summary || '')
+    const fallbackDuration = done?.startedAt ? (Date.now() - done.startedAt) / 1000 : undefined
+
+    const line = buildToolTrailLine(
+      name,
+      done?.context || '',
+      Boolean(error),
+      error || summary || '',
+      duration ?? fallbackDuration
+    )
 
     this.activeTools = this.activeTools.filter(tool => tool.id !== toolId)
 
@@ -496,6 +659,7 @@ class TurnController {
 
   recordToolStart(toolId: string, name: string, context: string) {
     this.flushStreamingSegment()
+    this.closeReasoningSegment()
     this.pruneTransient()
     this.endReasoningPhase()
 
@@ -514,6 +678,7 @@ class TurnController {
     this.bufRef = ''
     this.interrupted = false
     this.lastStatusNote = ''
+    this.activeReasoningText = ''
     this.pendingSegmentTools = []
     this.protocolWarned = false
     this.reasoningSegmentIndex = null
@@ -552,7 +717,7 @@ class TurnController {
       this.streamTimer = null
       const raw = this.bufRef.trimStart()
       const visible = hasReasoningTag(raw) ? splitReasoning(raw).text : raw
-      patchTurnState({ streaming: visible })
+      patchTurnState({ streaming: boundedLiveRenderText(visible) })
     }, this.streamDelay)
   }
 
@@ -560,6 +725,8 @@ class TurnController {
     this.endReasoningPhase()
     this.clearReasoning()
     this.activeTools = []
+    this.activeReasoningText = ''
+    this.reasoningSegmentIndex = null
     this.turnTools = []
     this.toolTokenAcc = 0
     this.persistedToolLabels.clear()

ui-tui/src/components/messageLine.tsx

@@ -5,9 +5,9 @@ import { LONG_MSG } from '../config/limits.js'
 import { sectionMode } from '../domain/details.js'
 import { userDisplay } from '../domain/messages.js'
 import { ROLE } from '../domain/roles.js'
-import { compactPreview, hasAnsi, isPasteBackedText, stripAnsi } from '../lib/text.js'
+import { boundedLiveRenderText, compactPreview, hasAnsi, isPasteBackedText, stripAnsi } from '../lib/text.js'
 import type { Theme } from '../theme.js'
-import type { DetailsMode, Msg, SectionVisibility } from '../types.js'
+import type { ActiveTool, DetailsMode, Msg, SectionVisibility } from '../types.js'
 
 import { Md } from './markdown.js'
 import { ToolTrail } from './thinking.js'
@@ -20,7 +20,8 @@ export const MessageLine = memo(function MessageLine({
   isStreaming = false,
   msg,
   sections,
-  t
+  t,
+  tools = []
 }: MessageLineProps) {
   // Per-section overrides win over the global mode, so resolve each section
   // we might consume here once and gate visibility on the *content-bearing*
@@ -34,7 +35,7 @@ export const MessageLine = memo(function MessageLine({
   const activityMode = sectionMode('activity', detailsMode, sections, detailsModeCommandOverride)
   const thinking = msg.thinking?.trim() ?? ''
 
-  if (msg.kind === 'trail' && (msg.tools?.length || thinking)) {
+  if (msg.kind === 'trail' && (msg.tools?.length || tools.length || thinking)) {
     return thinkingMode !== 'hidden' || toolsMode !== 'hidden' || activityMode !== 'hidden' ? (
       <Box flexDirection="column">
         <ToolTrail
@@ -44,6 +45,7 @@ export const MessageLine = memo(function MessageLine({
           reasoningTokens={msg.thinkingTokens}
           sections={sections}
           t={t}
+          tools={tools}
           toolTokens={msg.toolTokens}
           trail={msg.tools ?? []}
         />
@@ -86,7 +88,11 @@ export const MessageLine = memo(function MessageLine({
     }
 
     if (msg.role === 'assistant') {
-      return isStreaming ? <Text color={body}>{msg.text}</Text> : <Md compact={compact} t={t} text={msg.text} />
+      return isStreaming ? (
+        <Text color={body}>{boundedLiveRenderText(msg.text)}</Text>
+      ) : (
+        <Md compact={compact} t={t} text={msg.text} />
+      )
     }
 
     if (msg.role === 'user' && msg.text.length > LONG_MSG && isPasteBackedText(msg.text)) {
@@ -154,4 +160,5 @@ interface MessageLineProps {
   msg: Msg
   sections?: SectionVisibility
   t: Theme
+  tools?: ActiveTool[]
 }

ui-tui/src/components/thinking.tsx

@@ -16,12 +16,14 @@ import {
   widthByDepth
 } from '../lib/subagentTree.js'
 import {
+  boundedLiveRenderText,
   compactPreview,
   estimateTokensRough,
   fmtK,
   formatToolCall,
   parseToolTrailResultLine,
   pick,
+  splitToolDuration,
   thinkingPreview,
   toolTrailLabel
 } from '../lib/text.js'
@@ -633,7 +635,12 @@ export const Thinking = memo(function Thinking({
   streaming?: boolean
   t: Theme
 }) {
-  const preview = useMemo(() => thinkingPreview(reasoning, mode, THINKING_COT_MAX), [mode, reasoning])
+  const preview = useMemo(() => {
+    const raw = thinkingPreview(reasoning, mode, THINKING_COT_MAX)
+
+    return mode === 'full' ? boundedLiveRenderText(raw) : raw
+  }, [mode, reasoning])
+
   const lines = useMemo(() => preview.split('\n').map(line => line.replace(/\t/g, '  ')), [preview])
 
   if (!preview && !active) {
@@ -790,7 +797,7 @@ export const ToolTrail = memo(function ToolTrail({
     if (parsed) {
       groups.push({
         color: parsed.mark === '✗' ? t.color.error : t.color.cornsilk,
-        content: parsed.detail ? parsed.call : `${parsed.call} ${parsed.mark}`,
+        content: parsed.call,
         details: [],
         key: `tr-${i}`,
         label: parsed.call
@@ -886,6 +893,21 @@ export const ToolTrail = memo(function ToolTrail({
   const delegateGroups = groups.filter(g => g.label.startsWith('Delegate Task'))
   const inlineDelegateKey = hasSubagents && delegateGroups.length === 1 ? delegateGroups[0]!.key : null
 
+  const toolLabel = (group: Group) => {
+    const { duration, label } = splitToolDuration(String(group.content))
+
+    return duration ? (
+      <>
+        {label}
+        <Text color={t.color.dim} dim>
+          {duration}
+        </Text>
+      </>
+    ) : (
+      group.content
+    )
+  }
+
   // ── Backstop: floating alerts when every panel is hidden ─────────
   //
   // Per-section overrides win over the global details_mode (they're computed
@@ -1051,7 +1073,7 @@ export const ToolTrail = memo(function ToolTrail({
                   content={
                     <>
                       <Text color={t.color.amber}>● </Text>
-                      {group.content}
+                      {toolLabel(group)}
                     </>
                   }
                   rails={rails}

ui-tui/src/config/limits.ts

@@ -1,4 +1,6 @@
 export const LARGE_PASTE = { chars: 8000, lines: 80 }
+export const LIVE_RENDER_MAX_CHARS = 16_000
+export const LIVE_RENDER_MAX_LINES = 240
 export const LONG_MSG = 300
 export const MAX_HISTORY = 800
 export const THINKING_COT_MAX = 160

ui-tui/src/lib/text.ts

@@ -1,4 +1,4 @@
-import { THINKING_COT_MAX } from '../config/limits.js'
+import { LIVE_RENDER_MAX_CHARS, LIVE_RENDER_MAX_LINES, THINKING_COT_MAX } from '../config/limits.js'
 import { VERBS } from '../content/verbs.js'
 import type { ThinkingMode } from '../types.js'
 
@@ -88,6 +88,61 @@ export const thinkingPreview = (reasoning: string, mode: ThinkingMode, max: numb
   return !raw || mode === 'collapsed' ? '' : mode === 'full' ? raw : compactPreview(raw.replace(WS_RE, ' '), max)
 }
 
+export const boundedLiveRenderText = (
+  text: string,
+  { maxChars = LIVE_RENDER_MAX_CHARS, maxLines = LIVE_RENDER_MAX_LINES } = {}
+) => {
+  if (text.length <= maxChars && text.split('\n', maxLines + 1).length <= maxLines) {
+    return text
+  }
+
+  let start = 0
+  let idx = text.length
+
+  for (let seen = 0; seen < maxLines && idx > 0; seen++) {
+    idx = text.lastIndexOf('\n', idx - 1)
+    start = idx < 0 ? 0 : idx + 1
+
+    if (idx < 0) {
+      break
+    }
+  }
+
+  const lineStart = start
+  start = Math.max(lineStart, text.length - maxChars)
+
+  if (start > lineStart) {
+    const nextBreak = text.indexOf('\n', start)
+
+    if (nextBreak >= 0 && nextBreak < text.length - 1) {
+      start = nextBreak + 1
+    }
+  }
+
+  const tail = text.slice(start).trimStart()
+  const omittedLines = countNewlines(text, start)
+  const omittedChars = Math.max(0, text.length - tail.length)
+
+  const label =
+    omittedLines > 0
+      ? `[showing live tail; omitted ${fmtK(omittedLines)} lines / ${fmtK(omittedChars)} chars]\n`
+      : `[showing live tail; omitted ${fmtK(omittedChars)} chars]\n`
+
+  return `${label}${tail}`
+}
+
+const countNewlines = (text: string, end: number) => {
+  let count = 0
+
+  for (let i = 0; i < end; i++) {
+    if (text.charCodeAt(i) === 10) {
+      count++
+    }
+  }
+
+  return count
+}
+
 export const stripTrailingPasteNewlines = (text: string) => (/[^\n]/.test(text) ? text.replace(/\n+$/, '') : text)
 
 export const toolTrailLabel = (name: string) =>
@@ -104,10 +159,17 @@ export const formatToolCall = (name: string, context = '') => {
   return preview ? `${label}("${preview}")` : label
 }
 
-export const buildToolTrailLine = (name: string, context: string, error?: boolean, note?: string) => {
+export const buildToolTrailLine = (
+  name: string,
+  context: string,
+  error?: boolean,
+  note?: string,
+  duration?: number
+) => {
   const detail = compactPreview(note ?? '', 72)
+  const took = duration !== undefined ? ` (${duration.toFixed(1)}s)` : ''
 
-  return `${formatToolCall(name, context)}${detail ? ` :: ${detail}` : ''} ${error ? ' ✗' : ' ✓'}`
+  return `${formatToolCall(name, context)}${took}${detail ? ` :: ${detail}` : ''} ${error ? '✗' : '✓'}`
 }
 
 export const isToolTrailResultLine = (line: string) => line.endsWith(' ✓') || line.endsWith(' ✗')
@@ -134,6 +196,12 @@ export const parseToolTrailResultLine = (line: string) => {
   return { call: body, detail: '', mark }
 }
 
+export const splitToolDuration = (call: string) => {
+  const match = call.match(/^(.*?)( \(\d+(?:\.\d)?s\))$/)
+
+  return match ? { label: match[1]!, duration: match[2]! } : { label: call, duration: '' }
+}
+
 export const isTransientTrailLine = (line: string) => line.startsWith('drafting ') || line === 'analyzing tool output…'
 
 export const sameToolTrailGroup = (label: string, entry: string) =>