docs: move Discord behavior guidance to top

fix(update): drop autostash by stash selector

agent/anthropic_adapter.py

@@ -102,30 +102,15 @@ def build_anthropic_client(api_key: str, base_url: str = None):
 
 
 def read_claude_code_credentials() -> Optional[Dict[str, Any]]:
-    """Read credentials from Claude Code's config files.
+    """Read refreshable Claude Code OAuth credentials from ~/.claude/.credentials.json.
 
-    Checks two locations (in order):
-      1. ~/.claude.json — top-level primaryApiKey (native binary, v2.x)
-      2. ~/.claude/.credentials.json — claudeAiOauth block (npm/legacy installs)
+    This intentionally excludes ~/.claude.json primaryApiKey. Opencode's
+    subscription flow is OAuth/setup-token based with refreshable credentials,
+    and native direct Anthropic provider usage should follow that path rather
+    than auto-detecting Claude's first-party managed key.
 
     Returns dict with {accessToken, refreshToken?, expiresAt?} or None.
     """
-    # 1. Native binary (v2.x): ~/.claude.json with top-level primaryApiKey
-    claude_json = Path.home() / ".claude.json"
-    if claude_json.exists():
-        try:
-            data = json.loads(claude_json.read_text(encoding="utf-8"))
-            primary_key = data.get("primaryApiKey", "")
-            if primary_key:
-                return {
-                    "accessToken": primary_key,
-                    "refreshToken": "",
-                    "expiresAt": 0,  # Managed keys don't have a user-visible expiry
-                }
-        except (json.JSONDecodeError, OSError, IOError) as e:
-            logger.debug("Failed to read ~/.claude.json: %s", e)
-
-    # 2. Legacy/npm installs: ~/.claude/.credentials.json
     cred_path = Path.home() / ".claude" / ".credentials.json"
     if cred_path.exists():
         try:
@@ -138,6 +123,7 @@ def read_claude_code_credentials() -> Optional[Dict[str, Any]]:
                         "accessToken": access_token,
                         "refreshToken": oauth_data.get("refreshToken", ""),
                         "expiresAt": oauth_data.get("expiresAt", 0),
+                        "source": "claude_code_credentials_file",
                     }
         except (json.JSONDecodeError, OSError, IOError) as e:
             logger.debug("Failed to read ~/.claude/.credentials.json: %s", e)
@@ -145,6 +131,20 @@ def read_claude_code_credentials() -> Optional[Dict[str, Any]]:
     return None
 
 
+def read_claude_managed_key() -> Optional[str]:
+    """Read Claude's native managed key from ~/.claude.json for diagnostics only."""
+    claude_json = Path.home() / ".claude.json"
+    if claude_json.exists():
+        try:
+            data = json.loads(claude_json.read_text(encoding="utf-8"))
+            primary_key = data.get("primaryApiKey", "")
+            if isinstance(primary_key, str) and primary_key.strip():
+                return primary_key.strip()
+        except (json.JSONDecodeError, OSError, IOError) as e:
+            logger.debug("Failed to read ~/.claude.json: %s", e)
+    return None
+
+
 def is_claude_code_token_valid(creds: Dict[str, Any]) -> bool:
     """Check if Claude Code credentials have a non-expired access token."""
     import time
@@ -273,6 +273,35 @@ def _prefer_refreshable_claude_code_token(env_token: str, creds: Optional[Dict[s
     return None
 
 
+def get_anthropic_token_source(token: Optional[str] = None) -> str:
+    """Best-effort source classification for an Anthropic credential token."""
+    token = (token or "").strip()
+    if not token:
+        return "none"
+
+    env_token = os.getenv("ANTHROPIC_TOKEN", "").strip()
+    if env_token and env_token == token:
+        return "anthropic_token_env"
+
+    cc_env_token = os.getenv("CLAUDE_CODE_OAUTH_TOKEN", "").strip()
+    if cc_env_token and cc_env_token == token:
+        return "claude_code_oauth_token_env"
+
+    creds = read_claude_code_credentials()
+    if creds and creds.get("accessToken") == token:
+        return str(creds.get("source") or "claude_code_credentials")
+
+    managed_key = read_claude_managed_key()
+    if managed_key and managed_key == token:
+        return "claude_json_primary_api_key"
+
+    api_key = os.getenv("ANTHROPIC_API_KEY", "").strip()
+    if api_key and api_key == token:
+        return "anthropic_api_key_env"
+
+    return "unknown"
+
+
 def resolve_anthropic_token() -> Optional[str]:
     """Resolve an Anthropic token from all available sources.
 
@@ -391,6 +420,68 @@ def _sanitize_tool_id(tool_id: str) -> str:
     return sanitized or "tool_0"
 
 
+def _convert_openai_image_part_to_anthropic(part: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+    """Convert an OpenAI-style image block to Anthropic's image source format."""
+    image_data = part.get("image_url", {})
+    url = image_data.get("url", "") if isinstance(image_data, dict) else str(image_data)
+    if not isinstance(url, str) or not url.strip():
+        return None
+    url = url.strip()
+
+    if url.startswith("data:"):
+        header, sep, data = url.partition(",")
+        if sep and ";base64" in header:
+            media_type = header[5:].split(";", 1)[0] or "image/png"
+            return {
+                "type": "image",
+                "source": {
+                    "type": "base64",
+                    "media_type": media_type,
+                    "data": data,
+                },
+            }
+
+    if url.startswith("http://") or url.startswith("https://"):
+        return {
+            "type": "image",
+            "source": {
+                "type": "url",
+                "url": url,
+            },
+        }
+
+    return None
+
+
+def _convert_user_content_part_to_anthropic(part: Any) -> Optional[Dict[str, Any]]:
+    if isinstance(part, dict):
+        ptype = part.get("type")
+        if ptype == "text":
+            block = {"type": "text", "text": part.get("text", "")}
+            if isinstance(part.get("cache_control"), dict):
+                block["cache_control"] = dict(part["cache_control"])
+            return block
+        if ptype == "image_url":
+            return _convert_openai_image_part_to_anthropic(part)
+        if ptype == "image" and part.get("source"):
+            return dict(part)
+        if ptype == "image" and part.get("data"):
+            media_type = part.get("mimeType") or part.get("media_type") or "image/png"
+            return {
+                "type": "image",
+                "source": {
+                    "type": "base64",
+                    "media_type": media_type,
+                    "data": part.get("data", ""),
+                },
+            }
+        if ptype == "tool_result":
+            return dict(part)
+    elif part is not None:
+        return {"type": "text", "text": str(part)}
+    return None
+
+
 def convert_tools_to_anthropic(tools: List[Dict]) -> List[Dict]:
     """Convert OpenAI tool definitions to Anthropic format."""
     if not tools:
@@ -495,7 +586,15 @@ def convert_messages_to_anthropic(
             continue
 
         # Regular user message
-        result.append({"role": "user", "content": content})
+        if isinstance(content, list):
+            converted_blocks = []
+            for part in content:
+                converted = _convert_user_content_part_to_anthropic(part)
+                if converted is not None:
+                    converted_blocks.append(converted)
+            result.append({"role": "user", "content": converted_blocks or [{"type": "text", "text": ""}]})
+        else:
+            result.append({"role": "user", "content": content})
 
     # Strip orphaned tool_use blocks (no matching tool_result follows)
     tool_result_ids = set()

agent/auxiliary_client.py

@@ -1,4 +1,4 @@
-"""Shared auxiliary OpenAI client for cheap/fast side tasks.
+"""Shared auxiliary client router for side tasks.
 
 Provides a single resolution chain so every consumer (context compression,
 session search, web extraction, vision analysis, browser vision) picks up
@@ -10,21 +10,21 @@ Resolution order for text tasks (auto mode):
   3. Custom endpoint (OPENAI_BASE_URL + OPENAI_API_KEY)
   4. Codex OAuth (Responses API via chatgpt.com with gpt-5.3-codex,
      wrapped to look like a chat.completions client)
-  5. Direct API-key providers (z.ai/GLM, Kimi/Moonshot, MiniMax, MiniMax-CN)
-     — checked via PROVIDER_REGISTRY entries with auth_type='api_key'
-  6. None
+  5. Native Anthropic
+  6. Direct API-key providers (z.ai/GLM, Kimi/Moonshot, MiniMax, MiniMax-CN)
+  7. None
 
 Resolution order for vision/multimodal tasks (auto mode):
-  1. OpenRouter
-  2. Nous Portal
-  3. Codex OAuth (gpt-5.3-codex supports vision via Responses API)
-  4. Custom endpoint (for local vision models: Qwen-VL, LLaVA, Pixtral, etc.)
-  5. None  (API-key providers like z.ai/Kimi/MiniMax are skipped —
-     they may not support multimodal)
+  1. Selected main provider, if it is one of the supported vision backends below
+  2. OpenRouter
+  3. Nous Portal
+  4. Codex OAuth (gpt-5.3-codex supports vision via Responses API)
+  5. Native Anthropic
+  6. Custom endpoint (for local vision models: Qwen-VL, LLaVA, Pixtral, etc.)
+  7. None
 
 Per-task provider overrides (e.g. AUXILIARY_VISION_PROVIDER,
-CONTEXT_COMPRESSION_PROVIDER) can force a specific provider for each task:
-"openrouter", "nous", "codex", or "main" (= steps 3-5).
+CONTEXT_COMPRESSION_PROVIDER) can force a specific provider for each task.
 Default "auto" follows the chains above.
 
 Per-task model overrides (e.g. AUXILIARY_VISION_MODEL,
@@ -78,6 +78,7 @@ auxiliary_is_nous: bool = False
 _OPENROUTER_MODEL = "google/gemini-3-flash-preview"
 _NOUS_MODEL = "gemini-3-flash"
 _NOUS_DEFAULT_BASE_URL = "https://inference-api.nousresearch.com/v1"
+_ANTHROPIC_DEFAULT_BASE_URL = "https://api.anthropic.com"
 _AUTH_JSON_PATH = get_hermes_home() / "auth.json"
 
 # Codex fallback: uses the Responses API (the only endpoint the Codex
@@ -313,6 +314,114 @@ class AsyncCodexAuxiliaryClient:
         self.base_url = sync_wrapper.base_url
 
 
+class _AnthropicCompletionsAdapter:
+    """OpenAI-client-compatible adapter for Anthropic Messages API."""
+
+    def __init__(self, real_client: Any, model: str):
+        self._client = real_client
+        self._model = model
+
+    def create(self, **kwargs) -> Any:
+        from agent.anthropic_adapter import build_anthropic_kwargs, normalize_anthropic_response
+
+        messages = kwargs.get("messages", [])
+        model = kwargs.get("model", self._model)
+        tools = kwargs.get("tools")
+        tool_choice = kwargs.get("tool_choice")
+        max_tokens = kwargs.get("max_tokens") or kwargs.get("max_completion_tokens") or 2000
+        temperature = kwargs.get("temperature")
+
+        normalized_tool_choice = None
+        if isinstance(tool_choice, str):
+            normalized_tool_choice = tool_choice
+        elif isinstance(tool_choice, dict):
+            choice_type = str(tool_choice.get("type", "")).lower()
+            if choice_type == "function":
+                normalized_tool_choice = tool_choice.get("function", {}).get("name")
+            elif choice_type in {"auto", "required", "none"}:
+                normalized_tool_choice = choice_type
+
+        anthropic_kwargs = build_anthropic_kwargs(
+            model=model,
+            messages=messages,
+            tools=tools,
+            max_tokens=max_tokens,
+            reasoning_config=None,
+            tool_choice=normalized_tool_choice,
+        )
+        if temperature is not None:
+            anthropic_kwargs["temperature"] = temperature
+
+        response = self._client.messages.create(**anthropic_kwargs)
+        assistant_message, finish_reason = normalize_anthropic_response(response)
+
+        usage = None
+        if hasattr(response, "usage") and response.usage:
+            prompt_tokens = getattr(response.usage, "input_tokens", 0) or 0
+            completion_tokens = getattr(response.usage, "output_tokens", 0) or 0
+            total_tokens = getattr(response.usage, "total_tokens", 0) or (prompt_tokens + completion_tokens)
+            usage = SimpleNamespace(
+                prompt_tokens=prompt_tokens,
+                completion_tokens=completion_tokens,
+                total_tokens=total_tokens,
+            )
+
+        choice = SimpleNamespace(
+            index=0,
+            message=assistant_message,
+            finish_reason=finish_reason,
+        )
+        return SimpleNamespace(
+            choices=[choice],
+            model=model,
+            usage=usage,
+        )
+
+
+class _AnthropicChatShim:
+    def __init__(self, adapter: _AnthropicCompletionsAdapter):
+        self.completions = adapter
+
+
+class AnthropicAuxiliaryClient:
+    """OpenAI-client-compatible wrapper over a native Anthropic client."""
+
+    def __init__(self, real_client: Any, model: str, api_key: str, base_url: str):
+        self._real_client = real_client
+        adapter = _AnthropicCompletionsAdapter(real_client, model)
+        self.chat = _AnthropicChatShim(adapter)
+        self.api_key = api_key
+        self.base_url = base_url
+
+    def close(self):
+        close_fn = getattr(self._real_client, "close", None)
+        if callable(close_fn):
+            close_fn()
+
+
+class _AsyncAnthropicCompletionsAdapter:
+    def __init__(self, sync_adapter: _AnthropicCompletionsAdapter):
+        self._sync = sync_adapter
+
+    async def create(self, **kwargs) -> Any:
+        import asyncio
+        return await asyncio.to_thread(self._sync.create, **kwargs)
+
+
+class _AsyncAnthropicChatShim:
+    def __init__(self, adapter: _AsyncAnthropicCompletionsAdapter):
+        self.completions = adapter
+
+
+class AsyncAnthropicAuxiliaryClient:
+    def __init__(self, sync_wrapper: "AnthropicAuxiliaryClient"):
+        sync_adapter = sync_wrapper.chat.completions
+        async_adapter = _AsyncAnthropicCompletionsAdapter(sync_adapter)
+        self.chat = _AsyncAnthropicChatShim(async_adapter)
+        self.api_key = sync_wrapper.api_key
+        self.base_url = sync_wrapper.base_url
+
+
 def _read_nous_auth() -> Optional[dict]:
     """Read and validate ~/.hermes/auth.json for an active Nous provider.
 
@@ -384,6 +493,9 @@ def _resolve_api_key_provider() -> Tuple[Optional[OpenAI], Optional[str]]:
                 break
         if not api_key:
             continue
+        if provider_id == "anthropic":
+            return _try_anthropic()
+
         # Resolve base URL (with optional env-var override)
         # Kimi Code keys (sk-kimi-) need api.kimi.com/coding/v1
         env_url = ""
@@ -534,6 +646,22 @@ def _try_codex() -> Tuple[Optional[Any], Optional[str]]:
     return CodexAuxiliaryClient(real_client, _CODEX_AUX_MODEL), _CODEX_AUX_MODEL
 
 
+def _try_anthropic() -> Tuple[Optional[Any], Optional[str]]:
+    try:
+        from agent.anthropic_adapter import build_anthropic_client, resolve_anthropic_token
+    except ImportError:
+        return None, None
+
+    token = resolve_anthropic_token()
+    if not token:
+        return None, None
+
+    model = _API_KEY_PROVIDER_AUX_MODELS.get("anthropic", "claude-haiku-4-5-20251001")
+    logger.debug("Auxiliary client: Anthropic native (%s)", model)
+    real_client = build_anthropic_client(token, _ANTHROPIC_DEFAULT_BASE_URL)
+    return AnthropicAuxiliaryClient(real_client, model, token, _ANTHROPIC_DEFAULT_BASE_URL), model
+
+
 def _resolve_forced_provider(forced: str) -> Tuple[Optional[OpenAI], Optional[str]]:
     """Resolve a specific forced provider.  Returns (None, None) if creds missing."""
     if forced == "openrouter":
@@ -596,6 +724,8 @@ def _to_async_client(sync_client, model: str):
 
     if isinstance(sync_client, CodexAuxiliaryClient):
         return AsyncCodexAuxiliaryClient(sync_client), model
+    if isinstance(sync_client, AnthropicAuxiliaryClient):
+        return AsyncAnthropicAuxiliaryClient(sync_client), model
 
     async_kwargs = {
         "api_key": sync_client.api_key,
@@ -756,6 +886,14 @@ def resolve_provider_client(
         return None, None
 
     if pconfig.auth_type == "api_key":
+        if provider == "anthropic":
+            client, default_model = _try_anthropic()
+            if client is None:
+                logger.warning("resolve_provider_client: anthropic requested but no Anthropic credentials found")
+                return None, None
+            final_model = model or default_model
+            return (_to_async_client(client, final_model) if async_mode else (client, final_model))
+
         # Find the first configured API key
         api_key = ""
         for env_var in pconfig.api_key_env_vars:
@@ -849,6 +987,7 @@ _VISION_AUTO_PROVIDER_ORDER = (
     "openrouter",
     "nous",
     "openai-codex",
+    "anthropic",
     "custom",
 )
 
@@ -870,6 +1009,8 @@ def _resolve_strict_vision_backend(provider: str) -> Tuple[Optional[Any], Option
         return _try_nous()
     if provider == "openai-codex":
         return _try_codex()
+    if provider == "anthropic":
+        return _try_anthropic()
     if provider == "custom":
         return _try_custom_endpoint()
     return None, None
@@ -879,19 +1020,36 @@ def _strict_vision_backend_available(provider: str) -> bool:
     return _resolve_strict_vision_backend(provider)[0] is not None
 
 
+def _preferred_main_vision_provider() -> Optional[str]:
+    """Return the selected main provider when it is also a supported vision backend."""
+    try:
+        from hermes_cli.config import load_config
+
+        config = load_config()
+        model_cfg = config.get("model", {})
+        if isinstance(model_cfg, dict):
+            provider = _normalize_vision_provider(model_cfg.get("provider", ""))
+            if provider in _VISION_AUTO_PROVIDER_ORDER:
+                return provider
+    except Exception:
+        pass
+    return None
+
+
 def get_available_vision_backends() -> List[str]:
     """Return the currently available vision backends in auto-selection order.
 
     This is the single source of truth for setup, tool gating, and runtime
-    auto-routing of vision tasks. Phase 1 keeps the auto list conservative:
-    OpenRouter, Nous Portal, Codex OAuth, then custom OpenAI-compatible
-    endpoints. Explicit provider overrides can still route elsewhere.
+    auto-routing of vision tasks. The selected main provider is preferred when
+    it is also a known-good vision backend; otherwise Hermes falls back through
+    the standard conservative order.
     """
-    return [
-        provider
-        for provider in _VISION_AUTO_PROVIDER_ORDER
-        if _strict_vision_backend_available(provider)
-    ]
+    ordered = list(_VISION_AUTO_PROVIDER_ORDER)
+    preferred = _preferred_main_vision_provider()
+    if preferred in ordered:
+        ordered.remove(preferred)
+        ordered.insert(0, preferred)
+    return [provider for provider in ordered if _strict_vision_backend_available(provider)]
 
 
 def resolve_vision_provider_client(

cron/jobs.py

@@ -292,6 +292,9 @@ def create_job(
     origin: Optional[Dict[str, Any]] = None,
     skill: Optional[str] = None,
     skills: Optional[List[str]] = None,
+    model: Optional[str] = None,
+    provider: Optional[str] = None,
+    base_url: Optional[str] = None,
 ) -> Dict[str, Any]:
     """
     Create a new cron job.
@@ -305,6 +308,9 @@ def create_job(
         origin: Source info where job was created (for "origin" delivery)
         skill: Optional legacy single skill name to load before running the prompt
         skills: Optional ordered list of skills to load before running the prompt
+        model: Optional per-job model override
+        provider: Optional per-job provider override
+        base_url: Optional per-job base URL override
 
     Returns:
         The created job dict
@@ -323,6 +329,13 @@ def create_job(
     now = _hermes_now().isoformat()
 
     normalized_skills = _normalize_skill_list(skill, skills)
+    normalized_model = str(model).strip() if isinstance(model, str) else None
+    normalized_provider = str(provider).strip() if isinstance(provider, str) else None
+    normalized_base_url = str(base_url).strip().rstrip("/") if isinstance(base_url, str) else None
+    normalized_model = normalized_model or None
+    normalized_provider = normalized_provider or None
+    normalized_base_url = normalized_base_url or None
+
     label_source = (prompt or (normalized_skills[0] if normalized_skills else None)) or "cron job"
     job = {
         "id": job_id,
@@ -330,6 +343,9 @@ def create_job(
         "prompt": prompt,
         "skills": normalized_skills,
         "skill": normalized_skills[0] if normalized_skills else None,
+        "model": normalized_model,
+        "provider": normalized_provider,
+        "base_url": normalized_base_url,
         "schedule": parsed_schedule,
         "schedule_display": parsed_schedule.get("display", schedule),
         "repeat": {

cron/scheduler.py

@@ -261,7 +261,7 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
             if delivery_target.get("thread_id") is not None:
                 os.environ["HERMES_CRON_AUTO_DELIVER_THREAD_ID"] = str(delivery_target["thread_id"])
 
-        model = os.getenv("HERMES_MODEL") or "anthropic/claude-opus-4.6"
+        model = job.get("model") or os.getenv("HERMES_MODEL") or "anthropic/claude-opus-4.6"
 
         # Load config.yaml for model, reasoning, prefill, toolsets, provider routing
         _cfg = {}
@@ -272,10 +272,11 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
                 with open(_cfg_path) as _f:
                     _cfg = yaml.safe_load(_f) or {}
                 _model_cfg = _cfg.get("model", {})
-                if isinstance(_model_cfg, str):
-                    model = _model_cfg
-                elif isinstance(_model_cfg, dict):
-                    model = _model_cfg.get("default", model)
+                if not job.get("model"):
+                    if isinstance(_model_cfg, str):
+                        model = _model_cfg
+                    elif isinstance(_model_cfg, dict):
+                        model = _model_cfg.get("default", model)
         except Exception as e:
             logger.warning("Job '%s': failed to load config.yaml, using defaults: %s", job_id, e)
 
@@ -320,9 +321,12 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
             format_runtime_provider_error,
         )
         try:
-            runtime = resolve_runtime_provider(
-                requested=os.getenv("HERMES_INFERENCE_PROVIDER"),
-            )
+            runtime_kwargs = {
+                "requested": job.get("provider") or os.getenv("HERMES_INFERENCE_PROVIDER"),
+            }
+            if job.get("base_url"):
+                runtime_kwargs["explicit_base_url"] = job.get("base_url")
+            runtime = resolve_runtime_provider(**runtime_kwargs)
         except Exception as exc:
             message = format_runtime_provider_error(exc)
             raise RuntimeError(message) from exc

environments/tool_call_parsers/deepseek_v3_parser.py

@@ -10,12 +10,13 @@ Format uses special unicode tokens:
     <｜tool▁call▁end｜>
     <｜tool▁calls▁end｜>
 
-Based on VLLM's DeepSeekV3ToolParser.extract_tool_calls()
+Fixes Issue #989: Support for multiple simultaneous tool calls.
 """
 
 import re
 import uuid
-from typing import List, Optional
+import logging
+from typing import List, Optional, Tuple
 
 from openai.types.chat.chat_completion_message_tool_call import (
     ChatCompletionMessageToolCall,
@@ -24,6 +25,7 @@ from openai.types.chat.chat_completion_message_tool_call import (
 
 from environments.tool_call_parsers import ParseResult, ToolCallParser, register_parser
 
+logger = logging.getLogger(__name__)
 
 @register_parser("deepseek_v3")
 class DeepSeekV3ToolCallParser(ToolCallParser):
@@ -32,45 +34,56 @@ class DeepSeekV3ToolCallParser(ToolCallParser):
 
     Uses special unicode tokens with fullwidth angle brackets and block elements.
     Extracts type, function name, and JSON arguments from the structured format.
+    Ensures all tool calls are captured when the model executes multiple actions.
     """
 
     START_TOKEN = "<｜tool▁calls▁begin｜>"
 
-    # Regex captures: type, function_name, function_arguments
+    # Updated PATTERN: Using \s* instead of literal \n for increased robustness
+    # against variations in model formatting (Issue #989).
     PATTERN = re.compile(
-        r"<｜tool▁call▁begin｜>(?P<type>.*?)<｜tool▁sep｜>(?P<function_name>.*?)\n```json\n(?P<function_arguments>.*?)\n```<｜tool▁call▁end｜>",
+        r"<｜tool▁call▁begin｜>(?P<type>.*?)<｜tool▁sep｜>(?P<function_name>.*?)\s*```json\s*(?P<function_arguments>.*?)\s*```\s*<｜tool▁call▁end｜>",
         re.DOTALL,
     )
 
     def parse(self, text: str) -> ParseResult:
+        """
+        Parses the input text and extracts all available tool calls.
+        """
         if self.START_TOKEN not in text:
             return text, None
 
         try:
-            matches = self.PATTERN.findall(text)
+            # Using finditer to capture ALL tool calls in the sequence
+            matches = list(self.PATTERN.finditer(text))
             if not matches:
                 return text, None
 
             tool_calls: List[ChatCompletionMessageToolCall] = []
+            
             for match in matches:
-                tc_type, func_name, func_args = match
+                func_name = match.group("function_name").strip()
+                func_args = match.group("function_arguments").strip()
+                
                 tool_calls.append(
                     ChatCompletionMessageToolCall(
                         id=f"call_{uuid.uuid4().hex[:8]}",
                         type="function",
                         function=Function(
-                            name=func_name.strip(),
-                            arguments=func_args.strip(),
+                            name=func_name,
+                            arguments=func_args,
                         ),
                     )
                 )
 
-            if not tool_calls:
-                return text, None
+            if tool_calls:
+                # Content is text before the first tool call block
+                content_index = text.find(self.START_TOKEN)
+                content = text[:content_index].strip()
+                return content if content else None, tool_calls
 
-            # Content is everything before the tool calls section
-            content = text[: text.find(self.START_TOKEN)].strip()
-            return content if content else None, tool_calls
-
-        except Exception:
+            return text, None
+
+        except Exception as e:
+            logger.error(f"Error parsing DeepSeek V3 tool calls: {e}")
             return text, None

gateway/config.py

@@ -21,6 +21,17 @@ from hermes_cli.config import get_hermes_home
 logger = logging.getLogger(__name__)
 
 
+def _coerce_bool(value: Any, default: bool = True) -> bool:
+    """Coerce bool-ish config values, preserving a caller-provided default."""
+    if value is None:
+        return default
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, str):
+        return value.strip().lower() in ("true", "1", "yes", "on")
+    return bool(value)
+
+
 class Platform(Enum):
     """Supported messaging platforms."""
     LOCAL = "local"
@@ -160,6 +171,9 @@ class GatewayConfig:
     
     # Delivery settings
     always_log_local: bool = True  # Always save cron outputs to local files
+
+    # STT settings
+    stt_enabled: bool = True  # Whether to auto-transcribe inbound voice messages
     
     def get_connected_platforms(self) -> List[Platform]:
         """Return list of platforms that are enabled and configured."""
@@ -224,6 +238,7 @@ class GatewayConfig:
             "quick_commands": self.quick_commands,
             "sessions_dir": str(self.sessions_dir),
             "always_log_local": self.always_log_local,
+            "stt_enabled": self.stt_enabled,
         }
     
     @classmethod
@@ -260,6 +275,10 @@ class GatewayConfig:
         if not isinstance(quick_commands, dict):
             quick_commands = {}
 
+        stt_enabled = data.get("stt_enabled")
+        if stt_enabled is None:
+            stt_enabled = data.get("stt", {}).get("enabled") if isinstance(data.get("stt"), dict) else None
+
         return cls(
             platforms=platforms,
             default_reset_policy=default_policy,
@@ -269,6 +288,7 @@ class GatewayConfig:
             quick_commands=quick_commands,
             sessions_dir=sessions_dir,
             always_log_local=data.get("always_log_local", True),
+            stt_enabled=_coerce_bool(stt_enabled, True),
         )
 
 
@@ -318,6 +338,12 @@ def load_gateway_config() -> GatewayConfig:
                 else:
                     logger.warning("Ignoring invalid quick_commands in config.yaml (expected mapping, got %s)", type(qc).__name__)
 
+            # Bridge STT enable/disable from config.yaml into gateway runtime.
+            # This keeps the gateway aligned with the user-facing config source.
+            stt_cfg = yaml_cfg.get("stt")
+            if isinstance(stt_cfg, dict) and "enabled" in stt_cfg:
+                config.stt_enabled = _coerce_bool(stt_cfg.get("enabled"), True)
+
             # Bridge discord settings from config.yaml to env vars
             # (env vars take precedence — only set if not already defined)
             discord_cfg = yaml_cfg.get("discord", {})

gateway/run.py

@@ -3550,7 +3550,7 @@ class GatewayRunner:
         audio_paths: List[str],
     ) -> str:
         """
-        Auto-transcribe user voice/audio messages using OpenAI Whisper API
+        Auto-transcribe user voice/audio messages using the configured STT provider
         and prepend the transcript to the message text.
 
         Args:
@@ -3560,6 +3560,12 @@ class GatewayRunner:
         Returns:
             The enriched message string with transcriptions prepended.
         """
+        if not getattr(self.config, "stt_enabled", True):
+            disabled_note = "[The user sent voice message(s), but transcription is disabled in config.]"
+            if user_text:
+                return f"{disabled_note}\n\n{user_text}"
+            return disabled_note
+
         from tools.transcription_tools import transcribe_audio, get_stt_model_from_config
         import asyncio
 

hermes_cli/config.py

@@ -219,7 +219,8 @@ DEFAULT_CONFIG = {
     },
     
     "stt": {
-        "provider": "local",  # "local" (free, faster-whisper) | "openai" (Whisper API)
+        "enabled": True,
+        "provider": "local",  # "local" (free, faster-whisper) | "groq" | "openai" (Whisper API)
         "local": {
             "model": "base",  # tiny, base, small, medium, large-v3
         },
@@ -300,7 +301,7 @@ DEFAULT_CONFIG = {
     },
 
     # Config schema version - bump this when adding new required fields
-    "_config_version": 7,
+    "_config_version": 8,
 }
 
 # =============================================================================

hermes_cli/main.py

@@ -2016,6 +2016,22 @@ def _stash_local_changes_if_needed(git_cmd: list[str], cwd: Path) -> Optional[st
 
 
 
+def _resolve_stash_selector(git_cmd: list[str], cwd: Path, stash_ref: str) -> Optional[str]:
+    stash_list = subprocess.run(
+        git_cmd + ["stash", "list", "--format=%gd %H"],
+        cwd=cwd,
+        capture_output=True,
+        text=True,
+        check=True,
+    )
+    for line in stash_list.stdout.splitlines():
+        selector, _, commit = line.partition(" ")
+        if commit.strip() == stash_ref:
+            return selector.strip()
+    return None
+
+
+
 def _restore_stashed_changes(
     git_cmd: list[str],
     cwd: Path,
@@ -2052,7 +2068,27 @@ def _restore_stashed_changes(
         print(f"Resolve manually with: git stash apply {stash_ref}")
         sys.exit(1)
 
-    subprocess.run(git_cmd + ["stash", "drop", stash_ref], cwd=cwd, check=True)
+    stash_selector = _resolve_stash_selector(git_cmd, cwd, stash_ref)
+    if stash_selector is None:
+        print("⚠ Local changes were restored, but Hermes couldn't find the stash entry to drop.")
+        print("  The stash was left in place. You can remove it manually after checking the result.")
+        print(f"  Look for commit {stash_ref} in `git stash list --format='%gd %H'` and drop that selector.")
+    else:
+        drop = subprocess.run(
+            git_cmd + ["stash", "drop", stash_selector],
+            cwd=cwd,
+            capture_output=True,
+            text=True,
+        )
+        if drop.returncode != 0:
+            print("⚠ Local changes were restored, but Hermes couldn't drop the saved stash entry.")
+            if drop.stdout.strip():
+                print(drop.stdout.strip())
+            if drop.stderr.strip():
+                print(drop.stderr.strip())
+            print("  The stash was left in place. You can remove it manually after checking the result.")
+            print(f"  If needed: git stash drop {stash_selector}")
+
     print("⚠ Local changes were restored on top of the updated codebase.")
     print("  Review `git diff` / `git status` if Hermes behaves unexpectedly.")
     return True

hermes_cli/setup.py

@@ -1268,11 +1268,9 @@ def setup_model_provider(config: dict):
 
     _vision_needs_setup = not bool(_vision_backends)
 
-    if selected_provider in {"openrouter", "nous", "openai-codex"}:
-        # If the user just selected one of our known-good vision backends during
-        # setup, treat vision as covered. Auth/setup failure returns earlier.
-        _vision_needs_setup = False
-    elif selected_provider == "custom" and "custom" in _vision_backends:
+    if selected_provider in _vision_backends:
+        # If the user just selected a backend Hermes can already use for
+        # vision, treat it as covered. Auth/setup failure returns earlier.
         _vision_needs_setup = False
 
     if _vision_needs_setup:

run_agent.py

@@ -2407,7 +2407,7 @@ class AIAgent:
                 fn_name = getattr(item, "name", "") or ""
                 arguments = getattr(item, "arguments", "{}")
                 if not isinstance(arguments, str):
-                    arguments = str(arguments)
+                    arguments = json.dumps(arguments, ensure_ascii=False)
                 raw_call_id = getattr(item, "call_id", None)
                 raw_item_id = getattr(item, "id", None)
                 embedded_call_id, _ = self._split_responses_tool_id(raw_item_id)
@@ -2428,7 +2428,7 @@ class AIAgent:
                 fn_name = getattr(item, "name", "") or ""
                 arguments = getattr(item, "input", "{}")
                 if not isinstance(arguments, str):
-                    arguments = str(arguments)
+                    arguments = json.dumps(arguments, ensure_ascii=False)
                 raw_call_id = getattr(item, "call_id", None)
                 raw_item_id = getattr(item, "id", None)
                 embedded_call_id, _ = self._split_responses_tool_id(raw_item_id)

skills/productivity/google-workspace/SKILL.md

@@ -102,7 +102,9 @@ This prints a URL. **Send the URL to the user** and tell them:
 ### Step 4: Exchange the code
 
 The user will paste back either a URL like `http://localhost:1/?code=4/0A...&scope=...`
-or just the code string. Either works:
+or just the code string. Either works. The `--auth-url` step stores a temporary
+pending OAuth session locally so `--auth-code` can complete the PKCE exchange
+later, even on headless systems:
 
 ```bash
 $GSETUP --auth-code "THE_URL_OR_CODE_THE_USER_PASTED"
@@ -119,6 +121,7 @@ Should print `AUTHENTICATED`. Setup is complete — token refreshes automaticall
 ### Notes
 
 - Token is stored at `~/.hermes/google_token.json` and auto-refreshes.
+- Pending OAuth session state/verifier are stored temporarily at `~/.hermes/google_oauth_pending.json` until exchange completes.
 - To revoke: `$GSETUP --revoke`
 
 ## Usage

skills/productivity/google-workspace/scripts/setup.py

@@ -31,6 +31,7 @@ from pathlib import Path
 HERMES_HOME = Path(os.getenv("HERMES_HOME", Path.home() / ".hermes"))
 TOKEN_PATH = HERMES_HOME / "google_token.json"
 CLIENT_SECRET_PATH = HERMES_HOME / "google_client_secret.json"
+PENDING_AUTH_PATH = HERMES_HOME / "google_oauth_pending.json"
 
 SCOPES = [
     "https://www.googleapis.com/auth/gmail.readonly",
@@ -141,6 +142,58 @@ def store_client_secret(path: str):
     print(f"OK: Client secret saved to {CLIENT_SECRET_PATH}")
 
 
+def _save_pending_auth(*, state: str, code_verifier: str):
+    """Persist the OAuth session bits needed for a later token exchange."""
+    PENDING_AUTH_PATH.write_text(
+        json.dumps(
+            {
+                "state": state,
+                "code_verifier": code_verifier,
+                "redirect_uri": REDIRECT_URI,
+            },
+            indent=2,
+        )
+    )
+
+
+def _load_pending_auth() -> dict:
+    """Load the pending OAuth session created by get_auth_url()."""
+    if not PENDING_AUTH_PATH.exists():
+        print("ERROR: No pending OAuth session found. Run --auth-url first.")
+        sys.exit(1)
+
+    try:
+        data = json.loads(PENDING_AUTH_PATH.read_text())
+    except Exception as e:
+        print(f"ERROR: Could not read pending OAuth session: {e}")
+        print("Run --auth-url again to start a fresh OAuth session.")
+        sys.exit(1)
+
+    if not data.get("state") or not data.get("code_verifier"):
+        print("ERROR: Pending OAuth session is missing PKCE data.")
+        print("Run --auth-url again to start a fresh OAuth session.")
+        sys.exit(1)
+
+    return data
+
+
+def _extract_code_and_state(code_or_url: str) -> tuple[str, str | None]:
+    """Accept either a raw auth code or the full redirect URL pasted by the user."""
+    if not code_or_url.startswith("http"):
+        return code_or_url, None
+
+    from urllib.parse import parse_qs, urlparse
+
+    parsed = urlparse(code_or_url)
+    params = parse_qs(parsed.query)
+    if "code" not in params:
+        print("ERROR: No 'code' parameter found in URL.")
+        sys.exit(1)
+
+    state = params.get("state", [None])[0]
+    return params["code"][0], state
+
+
 def get_auth_url():
     """Print the OAuth authorization URL. User visits this in a browser."""
     if not CLIENT_SECRET_PATH.exists():
@@ -154,11 +207,13 @@ def get_auth_url():
         str(CLIENT_SECRET_PATH),
         scopes=SCOPES,
         redirect_uri=REDIRECT_URI,
+        autogenerate_code_verifier=True,
     )
-    auth_url, _ = flow.authorization_url(
+    auth_url, state = flow.authorization_url(
         access_type="offline",
         prompt="consent",
     )
+    _save_pending_auth(state=state, code_verifier=flow.code_verifier)
     # Print just the URL so the agent can extract it cleanly
     print(auth_url)
 
@@ -169,26 +224,23 @@ def exchange_auth_code(code: str):
         print("ERROR: No client secret stored. Run --client-secret first.")
         sys.exit(1)
 
+    pending_auth = _load_pending_auth()
+    code, returned_state = _extract_code_and_state(code)
+    if returned_state and returned_state != pending_auth["state"]:
+        print("ERROR: OAuth state mismatch. Run --auth-url again to start a fresh session.")
+        sys.exit(1)
+
     _ensure_deps()
     from google_auth_oauthlib.flow import Flow
 
     flow = Flow.from_client_secrets_file(
         str(CLIENT_SECRET_PATH),
         scopes=SCOPES,
-        redirect_uri=REDIRECT_URI,
+        redirect_uri=pending_auth.get("redirect_uri", REDIRECT_URI),
+        state=pending_auth["state"],
+        code_verifier=pending_auth["code_verifier"],
     )
 
-    # The code might come as a full redirect URL or just the code itself
-    if code.startswith("http"):
-        # Extract code from redirect URL: http://localhost:1/?code=CODE&scope=...
-        from urllib.parse import urlparse, parse_qs
-        parsed = urlparse(code)
-        params = parse_qs(parsed.query)
-        if "code" not in params:
-            print("ERROR: No 'code' parameter found in URL.")
-            sys.exit(1)
-        code = params["code"][0]
-
     try:
         flow.fetch_token(code=code)
     except Exception as e:
@@ -198,6 +250,7 @@ def exchange_auth_code(code: str):
 
     creds = flow.credentials
     TOKEN_PATH.write_text(creds.to_json())
+    PENDING_AUTH_PATH.unlink(missing_ok=True)
     print(f"OK: Authenticated. Token saved to {TOKEN_PATH}")
 
 
@@ -229,6 +282,7 @@ def revoke():
         print(f"Remote revocation failed (token may already be invalid): {e}")
 
     TOKEN_PATH.unlink(missing_ok=True)
+    PENDING_AUTH_PATH.unlink(missing_ok=True)
     print(f"Deleted {TOKEN_PATH}")
 
 

tests/agent/test_auxiliary_client.py

@@ -10,6 +10,8 @@ import pytest
 from agent.auxiliary_client import (
     get_text_auxiliary_client,
     get_vision_auxiliary_client,
+    get_available_vision_backends,
+    resolve_provider_client,
     auxiliary_max_tokens_param,
     _read_codex_access_token,
     _get_auxiliary_provider,
@@ -24,6 +26,7 @@ def _clean_env(monkeypatch):
     for key in (
         "OPENROUTER_API_KEY", "OPENAI_BASE_URL", "OPENAI_API_KEY",
         "OPENAI_MODEL", "LLM_MODEL", "NOUS_INFERENCE_BASE_URL",
+        "ANTHROPIC_API_KEY", "ANTHROPIC_TOKEN", "CLAUDE_CODE_OAUTH_TOKEN",
         # Per-task provider/model/direct-endpoint overrides
         "AUXILIARY_VISION_PROVIDER", "AUXILIARY_VISION_MODEL",
         "AUXILIARY_VISION_BASE_URL", "AUXILIARY_VISION_API_KEY",
@@ -210,14 +213,74 @@ class TestGetTextAuxiliaryClient:
 
 
 class TestVisionClientFallback:
-    """Vision client auto mode only tries OpenRouter + Nous (multimodal-capable)."""
+    """Vision client auto mode resolves known-good multimodal backends."""
 
     def test_vision_returns_none_without_any_credentials(self):
-        with patch("agent.auxiliary_client._read_nous_auth", return_value=None):
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("agent.auxiliary_client._try_anthropic", return_value=(None, None)),
+        ):
             client, model = get_vision_auxiliary_client()
         assert client is None
         assert model is None
 
+    def test_vision_auto_includes_anthropic_when_configured(self, monkeypatch):
+        monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
+            patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
+        ):
+            backends = get_available_vision_backends()
+
+        assert "anthropic" in backends
+
+    def test_resolve_provider_client_returns_native_anthropic_wrapper(self, monkeypatch):
+        monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
+            patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
+        ):
+            client, model = resolve_provider_client("anthropic")
+
+        assert client is not None
+        assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
+        assert model == "claude-haiku-4-5-20251001"
+
+    def test_vision_auto_uses_anthropic_when_no_higher_priority_backend(self, monkeypatch):
+        monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
+            patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
+        ):
+            client, model = get_vision_auxiliary_client()
+
+        assert client is not None
+        assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
+        assert model == "claude-haiku-4-5-20251001"
+
+    def test_selected_anthropic_provider_is_preferred_for_vision_auto(self, monkeypatch):
+        monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
+        monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
+
+        def fake_load_config():
+            return {"model": {"provider": "anthropic", "default": "claude-sonnet-4-6"}}
+
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
+            patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
+            patch("agent.auxiliary_client.OpenAI") as mock_openai,
+            patch("hermes_cli.config.load_config", fake_load_config),
+        ):
+            client, model = get_vision_auxiliary_client()
+
+        assert client is not None
+        assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
+        assert model == "claude-haiku-4-5-20251001"
+
     def test_vision_auto_includes_codex(self, codex_auth_dir):
         """Codex supports vision (gpt-5.3-codex), so auto mode should use it."""
         with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \

tests/cron/test_scheduler.py

@@ -309,6 +309,57 @@ class TestRunJobConfigLogging:
             f"Expected 'failed to parse prefill messages' warning in logs, got: {[r.message for r in caplog.records]}"
 
 
+class TestRunJobPerJobOverrides:
+    def test_job_level_model_provider_and_base_url_overrides_are_used(self, tmp_path):
+        config_yaml = tmp_path / "config.yaml"
+        config_yaml.write_text(
+            "model:\n"
+            "  default: gpt-5.4\n"
+            "  provider: openai-codex\n"
+            "  base_url: https://chatgpt.com/backend-api/codex\n"
+        )
+
+        job = {
+            "id": "briefing-job",
+            "name": "briefing",
+            "prompt": "hello",
+            "model": "perplexity/sonar-pro",
+            "provider": "custom",
+            "base_url": "http://127.0.0.1:4000/v1",
+        }
+
+        fake_db = MagicMock()
+        fake_runtime = {
+            "provider": "openrouter",
+            "api_mode": "chat_completions",
+            "base_url": "http://127.0.0.1:4000/v1",
+            "api_key": "***",
+        }
+
+        with patch("cron.scheduler._hermes_home", tmp_path), \
+             patch("cron.scheduler._resolve_origin", return_value=None), \
+             patch("dotenv.load_dotenv"), \
+             patch("hermes_state.SessionDB", return_value=fake_db), \
+             patch("hermes_cli.runtime_provider.resolve_runtime_provider", return_value=fake_runtime) as runtime_mock, \
+             patch("run_agent.AIAgent") as mock_agent_cls:
+            mock_agent = MagicMock()
+            mock_agent.run_conversation.return_value = {"final_response": "ok"}
+            mock_agent_cls.return_value = mock_agent
+
+            success, output, final_response, error = run_job(job)
+
+        assert success is True
+        assert error is None
+        assert final_response == "ok"
+        assert "ok" in output
+        runtime_mock.assert_called_once_with(
+            requested="custom",
+            explicit_base_url="http://127.0.0.1:4000/v1",
+        )
+        assert mock_agent_cls.call_args.kwargs["model"] == "perplexity/sonar-pro"
+        fake_db.close.assert_called_once()
+
+
 class TestRunJobSkillBacked:
     def test_run_job_loads_skill_and_disables_recursive_cron_tools(self, tmp_path):
         job = {

tests/gateway/test_stt_config.py

@@ -0,0 +1,53 @@
+"""Gateway STT config tests — honor stt.enabled: false from config.yaml."""
+
+from pathlib import Path
+from unittest.mock import AsyncMock, patch
+
+import pytest
+import yaml
+
+from gateway.config import GatewayConfig, load_gateway_config
+
+
+def test_gateway_config_stt_disabled_from_dict_nested():
+    config = GatewayConfig.from_dict({"stt": {"enabled": False}})
+    assert config.stt_enabled is False
+
+
+def test_load_gateway_config_bridges_stt_enabled_from_config_yaml(tmp_path, monkeypatch):
+    hermes_home = tmp_path / ".hermes"
+    hermes_home.mkdir()
+    (hermes_home / "config.yaml").write_text(
+        yaml.dump({"stt": {"enabled": False}}),
+        encoding="utf-8",
+    )
+
+    monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+    monkeypatch.setattr(Path, "home", lambda: tmp_path)
+
+    config = load_gateway_config()
+
+    assert config.stt_enabled is False
+
+
+@pytest.mark.asyncio
+async def test_enrich_message_with_transcription_skips_when_stt_disabled():
+    from gateway.run import GatewayRunner
+
+    runner = GatewayRunner.__new__(GatewayRunner)
+    runner.config = GatewayConfig(stt_enabled=False)
+
+    with patch(
+        "tools.transcription_tools.transcribe_audio",
+        side_effect=AssertionError("transcribe_audio should not be called when STT is disabled"),
+    ), patch(
+        "tools.transcription_tools.get_stt_model_from_config",
+        return_value=None,
+    ):
+        result = await runner._enrich_message_with_transcription(
+            "caption",
+            ["/tmp/voice.ogg"],
+        )
+
+    assert "transcription is disabled" in result.lower()
+    assert "caption" in result

tests/hermes_cli/test_setup.py

@@ -25,7 +25,11 @@ def test_nous_oauth_setup_keeps_current_model_when_syncing_disk_provider(
 
     config = load_config()
 
-    prompt_choices = iter([0, 2])
+    # Provider selection always comes first. Depending on available vision
+    # backends, setup may either skip the optional vision step or prompt for
+    # it before the default-model choice. Provide enough selections for both
+    # paths while still ending on "keep current model".
+    prompt_choices = iter([0, 2, 2])
     monkeypatch.setattr(
         "hermes_cli.setup.prompt_choice",
         lambda *args, **kwargs: next(prompt_choices),

tests/hermes_cli/test_setup_model_provider.py

@@ -111,6 +111,7 @@ def test_setup_keep_current_config_provider_uses_provider_specific_model_menu(tm
     monkeypatch.setattr("hermes_cli.auth.get_active_provider", lambda: None)
     monkeypatch.setattr("hermes_cli.auth.detect_external_credentials", lambda: [])
     monkeypatch.setattr("hermes_cli.models.provider_model_ids", lambda provider: [])
+    monkeypatch.setattr("agent.auxiliary_client.get_available_vision_backends", lambda: [])
 
     setup_model_provider(config)
     save_config(config)
@@ -149,6 +150,7 @@ def test_setup_keep_current_anthropic_can_configure_openai_vision_default(tmp_pa
     monkeypatch.setattr("hermes_cli.auth.get_active_provider", lambda: None)
     monkeypatch.setattr("hermes_cli.auth.detect_external_credentials", lambda: [])
     monkeypatch.setattr("hermes_cli.models.provider_model_ids", lambda provider: [])
+    monkeypatch.setattr("agent.auxiliary_client.get_available_vision_backends", lambda: [])
 
     setup_model_provider(config)
     env = _read_env(tmp_path)
@@ -224,3 +226,17 @@ def test_setup_summary_marks_codex_auth_as_vision_available(tmp_path, monkeypatc
     assert "missing run 'hermes setup' to configure" not in output
     assert "Mixture of Agents" in output
     assert "missing OPENROUTER_API_KEY" in output
+
+
+def test_setup_summary_marks_anthropic_auth_as_vision_available(tmp_path, monkeypatch, capsys):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    _clear_provider_env(monkeypatch)
+    monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
+    monkeypatch.setattr("shutil.which", lambda _name: None)
+    monkeypatch.setattr("agent.auxiliary_client.get_available_vision_backends", lambda: ["anthropic"])
+
+    _print_setup_summary(load_config(), tmp_path)
+    output = capsys.readouterr().out
+
+    assert "Vision (image analysis)" in output
+    assert "missing run 'hermes setup' to configure" not in output

tests/hermes_cli/test_update_autostash.py

@@ -46,6 +46,20 @@ def test_stash_local_changes_if_needed_returns_specific_stash_commit(monkeypatch
     assert calls[2][0][-3:] == ["rev-parse", "--verify", "refs/stash"]
 
 
+def test_resolve_stash_selector_returns_matching_entry(monkeypatch, tmp_path):
+    def fake_run(cmd, **kwargs):
+        assert cmd == ["git", "stash", "list", "--format=%gd %H"]
+        return SimpleNamespace(
+            stdout="stash@{0} def456\nstash@{1} abc123\n",
+            returncode=0,
+        )
+
+    monkeypatch.setattr(hermes_main.subprocess, "run", fake_run)
+
+    assert hermes_main._resolve_stash_selector(["git"], tmp_path, "abc123") == "stash@{1}"
+
+
+
 def test_restore_stashed_changes_prompts_before_applying(monkeypatch, tmp_path, capsys):
     calls = []
 
@@ -53,6 +67,8 @@ def test_restore_stashed_changes_prompts_before_applying(monkeypatch, tmp_path,
         calls.append((cmd, kwargs))
         if cmd[1:3] == ["stash", "apply"]:
             return SimpleNamespace(stdout="applied\n", stderr="", returncode=0)
+        if cmd[1:3] == ["stash", "list"]:
+            return SimpleNamespace(stdout="stash@{1} abc123\n", stderr="", returncode=0)
         if cmd[1:3] == ["stash", "drop"]:
             return SimpleNamespace(stdout="dropped\n", stderr="", returncode=0)
         raise AssertionError(f"unexpected command: {cmd}")
@@ -64,7 +80,8 @@ def test_restore_stashed_changes_prompts_before_applying(monkeypatch, tmp_path,
 
     assert restored is True
     assert calls[0][0] == ["git", "stash", "apply", "abc123"]
-    assert calls[1][0] == ["git", "stash", "drop", "abc123"]
+    assert calls[1][0] == ["git", "stash", "list", "--format=%gd %H"]
+    assert calls[2][0] == ["git", "stash", "drop", "stash@{1}"]
     out = capsys.readouterr().out
     assert "Restore local changes now? [Y/n]" in out
     assert "restored on top of the updated codebase" in out
@@ -99,6 +116,8 @@ def test_restore_stashed_changes_applies_without_prompt_when_disabled(monkeypatc
         calls.append((cmd, kwargs))
         if cmd[1:3] == ["stash", "apply"]:
             return SimpleNamespace(stdout="applied\n", stderr="", returncode=0)
+        if cmd[1:3] == ["stash", "list"]:
+            return SimpleNamespace(stdout="stash@{0} abc123\n", stderr="", returncode=0)
         if cmd[1:3] == ["stash", "drop"]:
             return SimpleNamespace(stdout="dropped\n", stderr="", returncode=0)
         raise AssertionError(f"unexpected command: {cmd}")
@@ -109,9 +128,64 @@ def test_restore_stashed_changes_applies_without_prompt_when_disabled(monkeypatc
 
     assert restored is True
     assert calls[0][0] == ["git", "stash", "apply", "abc123"]
+    assert calls[1][0] == ["git", "stash", "list", "--format=%gd %H"]
+    assert calls[2][0] == ["git", "stash", "drop", "stash@{0}"]
     assert "Restore local changes now?" not in capsys.readouterr().out
 
 
+
+def test_restore_stashed_changes_keeps_going_when_stash_entry_cannot_be_resolved(monkeypatch, tmp_path, capsys):
+    calls = []
+
+    def fake_run(cmd, **kwargs):
+        calls.append((cmd, kwargs))
+        if cmd[1:3] == ["stash", "apply"]:
+            return SimpleNamespace(stdout="applied\n", stderr="", returncode=0)
+        if cmd[1:3] == ["stash", "list"]:
+            return SimpleNamespace(stdout="stash@{0} def456\n", stderr="", returncode=0)
+        raise AssertionError(f"unexpected command: {cmd}")
+
+    monkeypatch.setattr(hermes_main.subprocess, "run", fake_run)
+
+    restored = hermes_main._restore_stashed_changes(["git"], tmp_path, "abc123", prompt_user=False)
+
+    assert restored is True
+    assert calls == [
+        (["git", "stash", "apply", "abc123"], {"cwd": tmp_path, "capture_output": True, "text": True}),
+        (["git", "stash", "list", "--format=%gd %H"], {"cwd": tmp_path, "capture_output": True, "text": True, "check": True}),
+    ]
+    out = capsys.readouterr().out
+    assert "couldn't find the stash entry to drop" in out
+    assert "stash was left in place" in out
+    assert "Look for commit abc123" in out
+
+
+
+def test_restore_stashed_changes_keeps_going_when_drop_fails(monkeypatch, tmp_path, capsys):
+    calls = []
+
+    def fake_run(cmd, **kwargs):
+        calls.append((cmd, kwargs))
+        if cmd[1:3] == ["stash", "apply"]:
+            return SimpleNamespace(stdout="applied\n", stderr="", returncode=0)
+        if cmd[1:3] == ["stash", "list"]:
+            return SimpleNamespace(stdout="stash@{0} abc123\n", stderr="", returncode=0)
+        if cmd[1:3] == ["stash", "drop"]:
+            return SimpleNamespace(stdout="", stderr="drop failed\n", returncode=1)
+        raise AssertionError(f"unexpected command: {cmd}")
+
+    monkeypatch.setattr(hermes_main.subprocess, "run", fake_run)
+
+    restored = hermes_main._restore_stashed_changes(["git"], tmp_path, "abc123", prompt_user=False)
+
+    assert restored is True
+    assert calls[2][0] == ["git", "stash", "drop", "stash@{0}"]
+    out = capsys.readouterr().out
+    assert "couldn't drop the saved stash entry" in out
+    assert "drop failed" in out
+    assert "git stash drop stash@{0}" in out
+
+
 def test_restore_stashed_changes_exits_cleanly_when_apply_fails(monkeypatch, tmp_path, capsys):
     calls = []
 

tests/skills/test_google_oauth_setup.py

@@ -0,0 +1,203 @@
+"""Regression tests for Google Workspace OAuth setup.
+
+These tests cover the headless/manual auth-code flow where the browser step and
+code exchange happen in separate process invocations.
+"""
+
+import importlib.util
+import json
+import sys
+import types
+from pathlib import Path
+
+import pytest
+
+
+SCRIPT_PATH = (
+    Path(__file__).resolve().parents[2]
+    / "skills/productivity/google-workspace/scripts/setup.py"
+)
+
+
+class FakeCredentials:
+    def __init__(self, payload=None):
+        self._payload = payload or {
+            "token": "access-token",
+            "refresh_token": "refresh-token",
+            "token_uri": "https://oauth2.googleapis.com/token",
+            "client_id": "client-id",
+            "client_secret": "client-secret",
+            "scopes": ["scope-a"],
+        }
+
+    def to_json(self):
+        return json.dumps(self._payload)
+
+
+class FakeFlow:
+    created = []
+    default_state = "generated-state"
+    default_verifier = "generated-code-verifier"
+    credentials_payload = None
+    fetch_error = None
+
+    def __init__(
+        self,
+        client_secrets_file,
+        scopes,
+        *,
+        redirect_uri=None,
+        state=None,
+        code_verifier=None,
+        autogenerate_code_verifier=False,
+    ):
+        self.client_secrets_file = client_secrets_file
+        self.scopes = scopes
+        self.redirect_uri = redirect_uri
+        self.state = state
+        self.code_verifier = code_verifier
+        self.autogenerate_code_verifier = autogenerate_code_verifier
+        self.authorization_kwargs = None
+        self.fetch_token_calls = []
+        self.credentials = FakeCredentials(self.credentials_payload)
+
+        if autogenerate_code_verifier and not self.code_verifier:
+            self.code_verifier = self.default_verifier
+        if not self.state:
+            self.state = self.default_state
+
+    @classmethod
+    def reset(cls):
+        cls.created = []
+        cls.default_state = "generated-state"
+        cls.default_verifier = "generated-code-verifier"
+        cls.credentials_payload = None
+        cls.fetch_error = None
+
+    @classmethod
+    def from_client_secrets_file(cls, client_secrets_file, scopes, **kwargs):
+        inst = cls(client_secrets_file, scopes, **kwargs)
+        cls.created.append(inst)
+        return inst
+
+    def authorization_url(self, **kwargs):
+        self.authorization_kwargs = kwargs
+        return f"https://auth.example/authorize?state={self.state}", self.state
+
+    def fetch_token(self, **kwargs):
+        self.fetch_token_calls.append(kwargs)
+        if self.fetch_error:
+            raise self.fetch_error
+
+
+@pytest.fixture
+def setup_module(monkeypatch, tmp_path):
+    FakeFlow.reset()
+
+    google_auth_module = types.ModuleType("google_auth_oauthlib")
+    flow_module = types.ModuleType("google_auth_oauthlib.flow")
+    flow_module.Flow = FakeFlow
+    google_auth_module.flow = flow_module
+    monkeypatch.setitem(sys.modules, "google_auth_oauthlib", google_auth_module)
+    monkeypatch.setitem(sys.modules, "google_auth_oauthlib.flow", flow_module)
+
+    spec = importlib.util.spec_from_file_location("google_workspace_setup_test", SCRIPT_PATH)
+    module = importlib.util.module_from_spec(spec)
+    assert spec.loader is not None
+    spec.loader.exec_module(module)
+
+    monkeypatch.setattr(module, "_ensure_deps", lambda: None)
+    monkeypatch.setattr(module, "CLIENT_SECRET_PATH", tmp_path / "google_client_secret.json")
+    monkeypatch.setattr(module, "TOKEN_PATH", tmp_path / "google_token.json")
+    monkeypatch.setattr(module, "PENDING_AUTH_PATH", tmp_path / "google_oauth_pending.json", raising=False)
+
+    client_secret = {
+        "installed": {
+            "client_id": "client-id",
+            "client_secret": "client-secret",
+            "auth_uri": "https://accounts.google.com/o/oauth2/auth",
+            "token_uri": "https://oauth2.googleapis.com/token",
+        }
+    }
+    module.CLIENT_SECRET_PATH.write_text(json.dumps(client_secret))
+    return module
+
+
+class TestGetAuthUrl:
+    def test_persists_state_and_code_verifier_for_later_exchange(self, setup_module, capsys):
+        setup_module.get_auth_url()
+
+        out = capsys.readouterr().out.strip()
+        assert out == "https://auth.example/authorize?state=generated-state"
+
+        saved = json.loads(setup_module.PENDING_AUTH_PATH.read_text())
+        assert saved["state"] == "generated-state"
+        assert saved["code_verifier"] == "generated-code-verifier"
+
+        flow = FakeFlow.created[-1]
+        assert flow.autogenerate_code_verifier is True
+        assert flow.authorization_kwargs == {"access_type": "offline", "prompt": "consent"}
+
+
+class TestExchangeAuthCode:
+    def test_reuses_saved_pkce_material_for_plain_code(self, setup_module):
+        setup_module.PENDING_AUTH_PATH.write_text(
+            json.dumps({"state": "saved-state", "code_verifier": "saved-verifier"})
+        )
+
+        setup_module.exchange_auth_code("4/test-auth-code")
+
+        flow = FakeFlow.created[-1]
+        assert flow.state == "saved-state"
+        assert flow.code_verifier == "saved-verifier"
+        assert flow.fetch_token_calls == [{"code": "4/test-auth-code"}]
+        assert json.loads(setup_module.TOKEN_PATH.read_text())["token"] == "access-token"
+        assert not setup_module.PENDING_AUTH_PATH.exists()
+
+    def test_extracts_code_from_redirect_url_and_checks_state(self, setup_module):
+        setup_module.PENDING_AUTH_PATH.write_text(
+            json.dumps({"state": "saved-state", "code_verifier": "saved-verifier"})
+        )
+
+        setup_module.exchange_auth_code(
+            "http://localhost:1/?code=4/extracted-code&state=saved-state&scope=gmail"
+        )
+
+        flow = FakeFlow.created[-1]
+        assert flow.fetch_token_calls == [{"code": "4/extracted-code"}]
+
+    def test_rejects_state_mismatch(self, setup_module, capsys):
+        setup_module.PENDING_AUTH_PATH.write_text(
+            json.dumps({"state": "saved-state", "code_verifier": "saved-verifier"})
+        )
+
+        with pytest.raises(SystemExit):
+            setup_module.exchange_auth_code(
+                "http://localhost:1/?code=4/extracted-code&state=wrong-state"
+            )
+
+        out = capsys.readouterr().out
+        assert "state mismatch" in out.lower()
+        assert not setup_module.TOKEN_PATH.exists()
+
+    def test_requires_pending_auth_session(self, setup_module, capsys):
+        with pytest.raises(SystemExit):
+            setup_module.exchange_auth_code("4/test-auth-code")
+
+        out = capsys.readouterr().out
+        assert "run --auth-url first" in out.lower()
+        assert not setup_module.TOKEN_PATH.exists()
+
+    def test_keeps_pending_auth_session_when_exchange_fails(self, setup_module, capsys):
+        setup_module.PENDING_AUTH_PATH.write_text(
+            json.dumps({"state": "saved-state", "code_verifier": "saved-verifier"})
+        )
+        FakeFlow.fetch_error = Exception("invalid_grant: Missing code verifier")
+
+        with pytest.raises(SystemExit):
+            setup_module.exchange_auth_code("4/test-auth-code")
+
+        out = capsys.readouterr().out
+        assert "token exchange failed" in out.lower()
+        assert setup_module.PENDING_AUTH_PATH.exists()
+        assert not setup_module.TOKEN_PATH.exists()

tests/test_anthropic_adapter.py

@@ -16,6 +16,7 @@ from agent.anthropic_adapter import (
     build_anthropic_kwargs,
     convert_messages_to_anthropic,
     convert_tools_to_anthropic,
+    get_anthropic_token_source,
     is_claude_code_token_valid,
     normalize_anthropic_response,
     normalize_model_name,
@@ -87,16 +88,25 @@ class TestReadClaudeCodeCredentials:
         cred_file.parent.mkdir(parents=True)
         cred_file.write_text(json.dumps({
             "claudeAiOauth": {
-                "accessToken": "sk-ant-oat01-test-token",
-                "refreshToken": "sk-ant-ort01-refresh",
+                "accessToken": "sk-ant-oat01-token",
+                "refreshToken": "sk-ant-oat01-refresh",
                 "expiresAt": int(time.time() * 1000) + 3600_000,
             }
         }))
         monkeypatch.setattr("agent.anthropic_adapter.Path.home", lambda: tmp_path)
         creds = read_claude_code_credentials()
         assert creds is not None
-        assert creds["accessToken"] == "sk-ant-oat01-test-token"
-        assert creds["refreshToken"] == "sk-ant-ort01-refresh"
+        assert creds["accessToken"] == "sk-ant-oat01-token"
+        assert creds["refreshToken"] == "sk-ant-oat01-refresh"
+        assert creds["source"] == "claude_code_credentials_file"
+
+    def test_ignores_primary_api_key_for_native_anthropic_resolution(self, tmp_path, monkeypatch):
+        claude_json = tmp_path / ".claude.json"
+        claude_json.write_text(json.dumps({"primaryApiKey": "sk-ant-api03-primary"}))
+        monkeypatch.setattr("agent.anthropic_adapter.Path.home", lambda: tmp_path)
+
+        creds = read_claude_code_credentials()
+        assert creds is None
 
     def test_returns_none_for_missing_file(self, tmp_path, monkeypatch):
         monkeypatch.setattr("agent.anthropic_adapter.Path.home", lambda: tmp_path)
@@ -139,6 +149,24 @@ class TestResolveAnthropicToken:
         monkeypatch.setenv("ANTHROPIC_TOKEN", "sk-ant-oat01-mytoken")
         assert resolve_anthropic_token() == "sk-ant-oat01-mytoken"
 
+    def test_reports_claude_json_primary_key_source(self, monkeypatch, tmp_path):
+        monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+        monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
+        monkeypatch.delenv("CLAUDE_CODE_OAUTH_TOKEN", raising=False)
+        (tmp_path / ".claude.json").write_text(json.dumps({"primaryApiKey": "sk-ant-api03-primary"}))
+        monkeypatch.setattr("agent.anthropic_adapter.Path.home", lambda: tmp_path)
+
+        assert get_anthropic_token_source("sk-ant-api03-primary") == "claude_json_primary_api_key"
+
+    def test_does_not_resolve_primary_api_key_as_native_anthropic_token(self, monkeypatch, tmp_path):
+        monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+        monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
+        monkeypatch.delenv("CLAUDE_CODE_OAUTH_TOKEN", raising=False)
+        (tmp_path / ".claude.json").write_text(json.dumps({"primaryApiKey": "sk-ant-api03-primary"}))
+        monkeypatch.setattr("agent.anthropic_adapter.Path.home", lambda: tmp_path)
+
+        assert resolve_anthropic_token() is None
+
     def test_falls_back_to_api_key_when_no_oauth_sources_exist(self, monkeypatch, tmp_path):
         monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-mykey")
         monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
@@ -567,6 +595,56 @@ class TestConvertMessages:
         assert tool_block["content"] == "result"
         assert tool_block["cache_control"] == {"type": "ephemeral"}
 
+    def test_converts_data_url_image_to_anthropic_image_block(self):
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "Describe this image"},
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": "data:image/png;base64,ZmFrZQ=="},
+                    },
+                ],
+            }
+        ]
+
+        _, result = convert_messages_to_anthropic(messages)
+        blocks = result[0]["content"]
+        assert blocks[0] == {"type": "text", "text": "Describe this image"}
+        assert blocks[1] == {
+            "type": "image",
+            "source": {
+                "type": "base64",
+                "media_type": "image/png",
+                "data": "ZmFrZQ==",
+            },
+        }
+
+    def test_converts_remote_image_url_to_anthropic_image_block(self):
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "Describe this image"},
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": "https://example.com/cat.png"},
+                    },
+                ],
+            }
+        ]
+
+        _, result = convert_messages_to_anthropic(messages)
+        blocks = result[0]["content"]
+        assert blocks[1] == {
+            "type": "image",
+            "source": {
+                "type": "url",
+                "url": "https://example.com/cat.png",
+            },
+        }
+
     def test_empty_cached_assistant_tool_turn_converts_without_empty_text_block(self):
         messages = apply_anthropic_cache_control([
             {"role": "system", "content": "System prompt"},

tests/test_run_agent.py

@@ -2533,3 +2533,56 @@ class TestVprintForceOnErrors:
             agent._vprint("debug")
             agent._vprint("error", force=True)
         assert len(printed) == 2
+
+
+class TestNormalizeCodexDictArguments:
+    """_normalize_codex_response must produce valid JSON strings for tool
+    call arguments, even when the Responses API returns them as dicts."""
+
+    def _make_codex_response(self, item_type, arguments, item_status="completed"):
+        """Build a minimal Responses API response with a single tool call."""
+        item = SimpleNamespace(
+            type=item_type,
+            status=item_status,
+        )
+        if item_type == "function_call":
+            item.name = "web_search"
+            item.arguments = arguments
+            item.call_id = "call_abc123"
+            item.id = "fc_abc123"
+        elif item_type == "custom_tool_call":
+            item.name = "web_search"
+            item.input = arguments
+            item.call_id = "call_abc123"
+            item.id = "fc_abc123"
+        return SimpleNamespace(
+            output=[item],
+            status="completed",
+        )
+
+    def test_function_call_dict_arguments_produce_valid_json(self, agent):
+        """dict arguments from function_call must be serialised with
+        json.dumps, not str(), so downstream json.loads() succeeds."""
+        args_dict = {"query": "weather in NYC", "units": "celsius"}
+        response = self._make_codex_response("function_call", args_dict)
+        msg, _ = agent._normalize_codex_response(response)
+        tc = msg.tool_calls[0]
+        parsed = json.loads(tc.function.arguments)
+        assert parsed == args_dict
+
+    def test_custom_tool_call_dict_arguments_produce_valid_json(self, agent):
+        """dict arguments from custom_tool_call must also use json.dumps."""
+        args_dict = {"path": "/tmp/test.txt", "content": "hello"}
+        response = self._make_codex_response("custom_tool_call", args_dict)
+        msg, _ = agent._normalize_codex_response(response)
+        tc = msg.tool_calls[0]
+        parsed = json.loads(tc.function.arguments)
+        assert parsed == args_dict
+
+    def test_string_arguments_unchanged(self, agent):
+        """String arguments must pass through without modification."""
+        args_str = '{"query": "test"}'
+        response = self._make_codex_response("function_call", args_str)
+        msg, _ = agent._normalize_codex_response(response)
+        tc = msg.tool_calls[0]
+        assert tc.function.arguments == args_str

tests/tools/test_approval.py

@@ -456,3 +456,20 @@ class TestViewFullCommand:
         # After first 'v', is_truncated becomes False, so second 'v' -> deny
         assert result == "deny"
 
+
+class TestForkBombDetection:
+    """The fork bomb regex must match the classic :(){ :|:& };: pattern."""
+
+    def test_classic_fork_bomb(self):
+        dangerous, key, desc = detect_dangerous_command(":(){ :|:& };:")
+        assert dangerous is True, "classic fork bomb not detected"
+        assert "fork bomb" in desc.lower()
+
+    def test_fork_bomb_with_spaces(self):
+        dangerous, key, desc = detect_dangerous_command(":()  {  : | :&  } ; :")
+        assert dangerous is True, "fork bomb with extra spaces not detected"
+
+    def test_colon_in_safe_command_not_flagged(self):
+        dangerous, key, desc = detect_dangerous_command("echo hello:world")
+        assert dangerous is False
+

tests/tools/test_cronjob_tools.py

@@ -137,6 +137,22 @@ class TestScheduleCronjob:
         ))
         assert result["repeat"] == "5 times"
 
+    def test_schedule_persists_runtime_overrides(self):
+        result = json.loads(schedule_cronjob(
+            prompt="Pinned job",
+            schedule="every 1h",
+            model="anthropic/claude-sonnet-4",
+            provider="custom",
+            base_url="http://127.0.0.1:4000/v1/",
+        ))
+        assert result["success"] is True
+
+        listing = json.loads(list_cronjobs())
+        job = listing["jobs"][0]
+        assert job["model"] == "anthropic/claude-sonnet-4"
+        assert job["provider"] == "custom"
+        assert job["base_url"] == "http://127.0.0.1:4000/v1"
+
 
 # =========================================================================
 # list_cronjobs
@@ -249,6 +265,33 @@ class TestUnifiedCronjobTool:
         assert updated["job"]["name"] == "New Name"
         assert updated["job"]["schedule"] == "every 120m"
 
+    def test_update_runtime_overrides_can_set_and_clear(self):
+        created = json.loads(
+            cronjob(
+                action="create",
+                prompt="Check",
+                schedule="every 1h",
+                model="anthropic/claude-sonnet-4",
+                provider="custom",
+                base_url="http://127.0.0.1:4000/v1",
+            )
+        )
+        job_id = created["job_id"]
+
+        updated = json.loads(
+            cronjob(
+                action="update",
+                job_id=job_id,
+                model="openai/gpt-4.1",
+                provider="openrouter",
+                base_url="",
+            )
+        )
+        assert updated["success"] is True
+        assert updated["job"]["model"] == "openai/gpt-4.1"
+        assert updated["job"]["provider"] == "openrouter"
+        assert updated["job"]["base_url"] is None
+
     def test_create_skill_backed_job(self):
         result = json.loads(
             cronjob(

tests/tools/test_local_env_blocklist.py

@@ -91,6 +91,25 @@ class TestProviderEnvBlocklist:
         for var in registry_vars:
             assert var not in result_env, f"{var} leaked into subprocess env"
 
+    def test_non_registry_provider_vars_are_stripped(self):
+        """Extra provider vars not in PROVIDER_REGISTRY must also be blocked."""
+        extra_provider_vars = {
+            "GOOGLE_API_KEY": "google-key",
+            "DEEPSEEK_API_KEY": "deepseek-key",
+            "MISTRAL_API_KEY": "mistral-key",
+            "GROQ_API_KEY": "groq-key",
+            "TOGETHER_API_KEY": "together-key",
+            "PERPLEXITY_API_KEY": "perplexity-key",
+            "COHERE_API_KEY": "cohere-key",
+            "FIREWORKS_API_KEY": "fireworks-key",
+            "XAI_API_KEY": "xai-key",
+            "HELICONE_API_KEY": "helicone-key",
+        }
+        result_env = _run_with_env(extra_os_env=extra_provider_vars)
+
+        for var in extra_provider_vars:
+            assert var not in result_env, f"{var} leaked into subprocess env"
+
     def test_safe_vars_are_preserved(self):
         """Standard env vars (PATH, HOME, USER) must still be passed through."""
         result_env = _run_with_env()
@@ -171,3 +190,18 @@ class TestBlocklistCoverage:
         must also be in the blocklist."""
         extras = {"ANTHROPIC_TOKEN", "CLAUDE_CODE_OAUTH_TOKEN"}
         assert extras.issubset(_HERMES_PROVIDER_ENV_BLOCKLIST)
+
+    def test_non_registry_provider_vars_are_in_blocklist(self):
+        extras = {
+            "GOOGLE_API_KEY",
+            "DEEPSEEK_API_KEY",
+            "MISTRAL_API_KEY",
+            "GROQ_API_KEY",
+            "TOGETHER_API_KEY",
+            "PERPLEXITY_API_KEY",
+            "COHERE_API_KEY",
+            "FIREWORKS_API_KEY",
+            "XAI_API_KEY",
+            "HELICONE_API_KEY",
+        }
+        assert extras.issubset(_HERMES_PROVIDER_ENV_BLOCKLIST)

tests/tools/test_transcription.py

@@ -59,6 +59,10 @@ class TestGetProvider:
             from tools.transcription_tools import _get_provider
             assert _get_provider({}) == "local"
 
+    def test_disabled_config_returns_none(self):
+        from tools.transcription_tools import _get_provider
+        assert _get_provider({"enabled": False, "provider": "openai"}) == "none"
+
 
 # ---------------------------------------------------------------------------
 # File validation
@@ -217,6 +221,18 @@ class TestTranscribeAudio:
         assert result["success"] is False
         assert "No STT provider" in result["error"]
 
+    def test_disabled_config_returns_disabled_error(self, tmp_path):
+        audio_file = tmp_path / "test.ogg"
+        audio_file.write_bytes(b"fake audio")
+
+        with patch("tools.transcription_tools._load_stt_config", return_value={"enabled": False}), \
+             patch("tools.transcription_tools._get_provider", return_value="none"):
+            from tools.transcription_tools import transcribe_audio
+            result = transcribe_audio(str(audio_file))
+
+        assert result["success"] is False
+        assert "disabled" in result["error"].lower()
+
     def test_invalid_file_returns_error(self):
         from tools.transcription_tools import transcribe_audio
         result = transcribe_audio("/nonexistent/file.ogg")

tools/approval.py

@@ -38,7 +38,7 @@ DANGEROUS_PATTERNS = [
     (r'\bsystemctl\s+(stop|disable|mask)\b', "stop/disable system service"),
     (r'\bkill\s+-9\s+-1\b', "kill all processes"),
     (r'\bpkill\s+-9\b', "force kill processes"),
-    (r':()\s*{\s*:\s*\|\s*:&\s*}\s*;:', "fork bomb"),
+    (r':\(\)\s*\{\s*:\s*\|\s*:\s*&\s*\}\s*;\s*:', "fork bomb"),
     (r'\b(bash|sh|zsh)\s+-c\s+', "shell command via -c flag"),
     (r'\b(python[23]?|perl|ruby|node)\s+-[ec]\s+', "script execution via -e/-c flag"),
     (r'\b(curl|wget)\b.*\|\s*(ba)?sh\b', "pipe remote content to shell"),

tools/cronjob_tools.py

@@ -103,6 +103,16 @@ def _canonical_skills(skill: Optional[str] = None, skills: Optional[Any] = None)
 
 
 
+def _normalize_optional_job_value(value: Optional[Any], *, strip_trailing_slash: bool = False) -> Optional[str]:
+    if value is None:
+        return None
+    text = str(value).strip()
+    if strip_trailing_slash:
+        text = text.rstrip("/")
+    return text or None
+
+
+
 def _format_job(job: Dict[str, Any]) -> Dict[str, Any]:
     prompt = job.get("prompt", "")
     skills = _canonical_skills(job.get("skill"), job.get("skills"))
@@ -112,6 +122,9 @@ def _format_job(job: Dict[str, Any]) -> Dict[str, Any]:
         "skill": skills[0] if skills else None,
         "skills": skills,
         "prompt_preview": prompt[:100] + "..." if len(prompt) > 100 else prompt,
+        "model": job.get("model"),
+        "provider": job.get("provider"),
+        "base_url": job.get("base_url"),
         "schedule": job.get("schedule_display"),
         "repeat": _repeat_display(job),
         "deliver": job.get("deliver", "local"),
@@ -136,6 +149,9 @@ def cronjob(
     include_disabled: bool = False,
     skill: Optional[str] = None,
     skills: Optional[List[str]] = None,
+    model: Optional[str] = None,
+    provider: Optional[str] = None,
+    base_url: Optional[str] = None,
     reason: Optional[str] = None,
     task_id: str = None,
 ) -> str:
@@ -164,6 +180,9 @@ def cronjob(
                 deliver=deliver,
                 origin=_origin_from_env(),
                 skills=canonical_skills,
+                model=_normalize_optional_job_value(model),
+                provider=_normalize_optional_job_value(provider),
+                base_url=_normalize_optional_job_value(base_url, strip_trailing_slash=True),
             )
             return json.dumps(
                 {
@@ -240,6 +259,12 @@ def cronjob(
                 canonical_skills = _canonical_skills(skill, skills)
                 updates["skills"] = canonical_skills
                 updates["skill"] = canonical_skills[0] if canonical_skills else None
+            if model is not None:
+                updates["model"] = _normalize_optional_job_value(model)
+            if provider is not None:
+                updates["provider"] = _normalize_optional_job_value(provider)
+            if base_url is not None:
+                updates["base_url"] = _normalize_optional_job_value(base_url, strip_trailing_slash=True)
             if repeat is not None:
                 repeat_state = dict(job.get("repeat") or {})
                 repeat_state["times"] = repeat
@@ -272,6 +297,9 @@ def schedule_cronjob(
     name: Optional[str] = None,
     repeat: Optional[int] = None,
     deliver: Optional[str] = None,
+    model: Optional[str] = None,
+    provider: Optional[str] = None,
+    base_url: Optional[str] = None,
     task_id: str = None,
 ) -> str:
     return cronjob(
@@ -281,6 +309,9 @@ def schedule_cronjob(
         name=name,
         repeat=repeat,
         deliver=deliver,
+        model=model,
+        provider=provider,
+        base_url=base_url,
         task_id=task_id,
     )
 
@@ -343,6 +374,18 @@ Important safety rule: cron-run sessions should not recursively schedule more cr
                 "type": "string",
                 "description": "Delivery target: origin, local, telegram, discord, signal, or platform:chat_id"
             },
+            "model": {
+                "type": "string",
+                "description": "Optional per-job model override used when the cron job runs"
+            },
+            "provider": {
+                "type": "string",
+                "description": "Optional per-job provider override used when resolving runtime credentials"
+            },
+            "base_url": {
+                "type": "string",
+                "description": "Optional per-job base URL override paired with provider/model routing"
+            },
             "include_disabled": {
                 "type": "boolean",
                 "description": "For list: include paused/completed jobs"
@@ -407,6 +450,9 @@ registry.register(
         include_disabled=args.get("include_disabled", False),
         skill=args.get("skill"),
         skills=args.get("skills"),
+        model=args.get("model"),
+        provider=args.get("provider"),
+        base_url=args.get("base_url"),
         reason=args.get("reason"),
         task_id=kw.get("task_id"),
     ),

tools/environments/local.py

@@ -56,6 +56,17 @@ def _build_provider_env_blocklist() -> frozenset:
         "ANTHROPIC_TOKEN",         # OAuth token (not in registry as env var)
         "CLAUDE_CODE_OAUTH_TOKEN",
         "LLM_MODEL",
+        # Expanded isolation for other major providers (Issue #1002)
+        "GOOGLE_API_KEY",          # Gemini / Google AI Studio
+        "DEEPSEEK_API_KEY",        # DeepSeek
+        "MISTRAL_API_KEY",         # Mistral AI
+        "GROQ_API_KEY",            # Groq
+        "TOGETHER_API_KEY",        # Together AI
+        "PERPLEXITY_API_KEY",      # Perplexity
+        "COHERE_API_KEY",          # Cohere
+        "FIREWORKS_API_KEY",       # Fireworks AI
+        "XAI_API_KEY",             # xAI (Grok)
+        "HELICONE_API_KEY",        # LLM Observability proxy
     })
     return frozenset(blocked)
 

tools/transcription_tools.py

@@ -93,6 +93,18 @@ def _load_stt_config() -> dict:
         return {}
 
 
+def is_stt_enabled(stt_config: Optional[dict] = None) -> bool:
+    """Return whether STT is enabled in config."""
+    if stt_config is None:
+        stt_config = _load_stt_config()
+    enabled = stt_config.get("enabled", True)
+    if isinstance(enabled, str):
+        return enabled.strip().lower() in ("true", "1", "yes", "on")
+    if enabled is None:
+        return True
+    return bool(enabled)
+
+
 def _get_provider(stt_config: dict) -> str:
     """Determine which STT provider to use.
 
@@ -101,6 +113,9 @@ def _get_provider(stt_config: dict) -> str:
       2. Auto-detect: local > groq (free) > openai (paid)
       3. Disabled (returns "none")
     """
+    if not is_stt_enabled(stt_config):
+        return "none"
+
     provider = stt_config.get("provider", DEFAULT_PROVIDER)
 
     if provider == "local":
@@ -334,6 +349,13 @@ def transcribe_audio(file_path: str, model: Optional[str] = None) -> Dict[str, A
 
     # Load config and determine provider
     stt_config = _load_stt_config()
+    if not is_stt_enabled(stt_config):
+        return {
+            "success": False,
+            "transcript": "",
+            "error": "STT is disabled in config.yaml (stt.enabled: false).",
+        }
+
     provider = _get_provider(stt_config)
 
     if provider == "local":

tools/vision_tools.py

@@ -3,7 +3,8 @@
 Vision Tools Module
 
 This module provides vision analysis tools that work with image URLs.
-Uses Gemini 3 Flash Preview via OpenRouter API for intelligent image understanding.
+Uses the centralized auxiliary vision router, which can select OpenRouter,
+Nous, Codex, native Anthropic, or a custom OpenAI-compatible endpoint.
 
 Available tools:
 - vision_analyze_tool: Analyze images from URLs with custom prompts
@@ -409,7 +410,7 @@ if __name__ == "__main__":
     
     if not api_available:
         print("❌ No auxiliary vision model available")
-        print("Set OPENROUTER_API_KEY or configure Nous Portal to enable vision tools.")
+        print("Configure a supported multimodal backend (OpenRouter, Nous, Codex, Anthropic, or a custom OpenAI-compatible endpoint).")
         exit(1)
     else:
         print("✅ Vision model available")

tools/voice_mode.py

@@ -703,10 +703,11 @@ def check_voice_requirements() -> Dict[str, Any]:
         ``missing_packages``, and ``details``.
     """
     # Determine STT provider availability
-    from tools.transcription_tools import _get_provider, _load_stt_config, _HAS_FASTER_WHISPER
+    from tools.transcription_tools import _get_provider, _load_stt_config, is_stt_enabled, _HAS_FASTER_WHISPER
     stt_config = _load_stt_config()
+    stt_enabled = is_stt_enabled(stt_config)
     stt_provider = _get_provider(stt_config)
-    stt_available = stt_provider != "none"
+    stt_available = stt_enabled and stt_provider != "none"
 
     missing: List[str] = []
     has_audio = _audio_available()
@@ -725,7 +726,9 @@ def check_voice_requirements() -> Dict[str, Any]:
     else:
         details_parts.append("Audio capture: MISSING (pip install sounddevice numpy)")
 
-    if stt_provider == "local":
+    if not stt_enabled:
+        details_parts.append("STT provider: DISABLED in config (stt.enabled: false)")
+    elif stt_provider == "local":
         details_parts.append("STT provider: OK (local faster-whisper)")
     elif stt_provider == "groq":
         details_parts.append("STT provider: OK (Groq)")

website/docs/user-guide/messaging/discord.md

@@ -8,6 +8,21 @@ description: "Set up Hermes Agent as a Discord bot"
 
 Hermes Agent integrates with Discord as a bot, letting you chat with your AI assistant through direct messages or server channels. The bot receives your messages, processes them through the Hermes Agent pipeline (including tool use, memory, and reasoning), and responds in real time. It supports text, voice messages, file attachments, and slash commands.
 
+Before setup, here's the part most people want to know: how Hermes behaves once it's in your server.
+
+## How Hermes Behaves
+
+| Context | Behavior |
+|---------|----------|
+| **DMs** | Hermes responds to every message. No `@mention` needed. |
+| **Server channels** | By default, Hermes only responds when you `@mention` it. If you post in a channel without mentioning it, Hermes ignores the message. |
+| **Free-response channels** | You can make specific channels mention-free with `DISCORD_FREE_RESPONSE_CHANNELS`, or disable mentions globally with `DISCORD_REQUIRE_MENTION=false`. |
+| **Threads** | Hermes replies in the same thread. Mention rules still apply unless that thread or its parent channel is configured as free-response. |
+
+:::tip
+If you want a normal shared bot channel where people can talk to Hermes without tagging it every time, add that channel to `DISCORD_FREE_RESPONSE_CHANNELS`.
+:::
+
 This guide walks you through the full setup process — from creating your bot on Discord's Developer Portal to sending your first message.
 
 ## Step 1: Create a Discord Application
@@ -200,12 +215,6 @@ DISCORD_HOME_CHANNEL_NAME="#bot-updates"
 
 Replace the ID with the actual channel ID (right-click → Copy Channel ID with Developer Mode on).
 
-## Bot Behavior
-
-- **Server channels**: By default the bot requires an `@mention` before it responds in server channels. You can disable that globally with `DISCORD_REQUIRE_MENTION=false` or allow specific channels to be mention-free via `DISCORD_FREE_RESPONSE_CHANNELS`.
-- **Direct messages**: DMs always work, even without the Message Content Intent enabled (Discord exempts DMs from this requirement). However, you should still enable the intent for server channel support.
-- **Conversations**: Each channel or DM maintains its own conversation context.
-
 ## Voice Messages
 
 Hermes Agent supports Discord voice messages:

website/docs/user-guide/messaging/slack.md

@@ -193,8 +193,8 @@ Understanding how Hermes behaves in different contexts:
 | Context | Behavior |
 |---------|----------|
 | **DMs** | Bot responds to every message — no @mention needed |
-| **Channels** | Bot **only responds when @mentioned** (e.g., `@Hermes Agent what time is it?`) |
-| **Threads** | Bot replies in threads when the triggering message is in a thread |
+| **Channels** | Bot **only responds when @mentioned** (e.g., `@Hermes Agent what time is it?`). In channels, Hermes replies in a thread attached to that message. |
+| **Threads** | If you @mention Hermes inside an existing thread, it replies in that same thread. |
 
 :::tip
 In channels, always @mention the bot. Simply typing a message without mentioning it will be ignored.