Fix TUI input field ANSI leaks and text selection issues

1. Fix ANSI escape code leakage during scroll operations:
   - Add ensureSafeAnsi utility to ensure all ANSI sequences are properly terminated
   - Modify renderWithCursor and renderWithSelection to always include reset codes
   - Add final safety check in text rendering to catch any potential leaks

2. Improve text selection in input field:
   - Add proper mouse drag event handling for text selection
   - Enhance click handlers to support selection operations
   - Fix edge cases in selection rendering
   - Ensure proper reset of ANSI codes after selections

Fixes reported issues where users couldn't copy/paste text in input field and
experienced ANSI leakage during scrolling operations.

agent/anthropic_adapter.py

@@ -390,7 +390,16 @@ def build_anthropic_client(api_key: str, base_url: str = None, timeout: float =
         "timeout": Timeout(timeout=float(_read_timeout), connect=10.0),
     }
     if normalized_base_url:
-        kwargs["base_url"] = normalized_base_url
+        # Azure Anthropic endpoints require an ``api-version`` query parameter.
+        # Pass it via default_query so the SDK appends it to every request URL
+        # without corrupting the base_url (appending it directly produces
+        # malformed paths like /anthropic?api-version=.../v1/messages).
+        _is_azure_endpoint = "azure.com" in normalized_base_url.lower()
+        if _is_azure_endpoint and "api-version" not in normalized_base_url:
+            kwargs["base_url"] = normalized_base_url.rstrip("/")
+            kwargs["default_query"] = {"api-version": "2025-04-15"}
+        else:
+            kwargs["base_url"] = normalized_base_url
     common_betas = _common_betas_for_base_url(normalized_base_url)
 
     if _is_kimi_coding_endpoint(base_url):
@@ -1680,9 +1689,9 @@ def build_anthropic_kwargs(
 
     # ── Strip sampling params on 4.7+ ─────────────────────────────────
     # Opus 4.7 rejects any non-default temperature/top_p/top_k with a 400.
-    # Callers (auxiliary_client, flush_memories, etc.) may set these for
-    # older models; drop them here as a safety net so upstream 4.6 → 4.7
-    # migrations don't require coordinated edits everywhere.
+    # Callers (auxiliary_client, etc.) may set these for older models;
+    # drop them here as a safety net so upstream 4.6 → 4.7 migrations
+    # don't require coordinated edits everywhere.
     if _forbids_sampling_params(model):
         for _sampling_key in ("temperature", "top_p", "top_k"):
             kwargs.pop(_sampling_key, None)

agent/auxiliary_client.py

@@ -42,6 +42,7 @@ import time
 from pathlib import Path  # noqa: F401 — used by test mocks
 from types import SimpleNamespace
 from typing import Any, Dict, List, Optional, Tuple
+from urllib.parse import urlparse, parse_qs, urlunparse
 
 from openai import OpenAI
 
@@ -52,6 +53,17 @@ from utils import base_url_host_matches, base_url_hostname, normalize_proxy_env_
 
 logger = logging.getLogger(__name__)
 
+
+def _extract_url_query_params(url: str):
+    """Extract query params from URL, return (clean_url, default_query dict or None)."""
+    parsed = urlparse(url)
+    if parsed.query:
+        clean = urlunparse(parsed._replace(query=""))
+        params = {k: v[0] for k, v in parse_qs(parsed.query).items()}
+        return clean, params
+    return url, None
+
+
 # Module-level flag: only warn once per process about stale OPENAI_BASE_URL.
 _stale_base_url_warned = False
 
@@ -390,7 +402,7 @@ class _CodexCompletionsAdapter:
         # Note: the Codex endpoint (chatgpt.com/backend-api/codex) does NOT
         # support max_output_tokens or temperature — omit to avoid 400 errors.
 
-        # Tools support for flush_memories and similar callers
+        # Tools support for auxiliary callers (e.g. skills_hub) that pass function schemas
         tools = kwargs.get("tools")
         if tools:
             converted = []
@@ -1157,8 +1169,10 @@ def _try_custom_endpoint() -> Tuple[Optional[Any], Optional[str]]:
         return None, None
     model = _read_main_model() or "gpt-4o-mini"
     logger.debug("Auxiliary client: custom endpoint (%s, api_mode=%s)", model, custom_mode or "chat_completions")
+    _clean_base, _dq = _extract_url_query_params(custom_base)
+    _extra = {"default_query": _dq} if _dq else {}
     if custom_mode == "codex_responses":
-        real_client = OpenAI(api_key=custom_key, base_url=custom_base)
+        real_client = OpenAI(api_key=custom_key, base_url=_clean_base, **_extra)
         return CodexAuxiliaryClient(real_client, model), model
     if custom_mode == "anthropic_messages":
         # Third-party Anthropic-compatible gateway (MiniMax, Zhipu GLM,
@@ -1172,12 +1186,12 @@ def _try_custom_endpoint() -> Tuple[Optional[Any], Optional[str]]:
                 "Custom endpoint declares api_mode=anthropic_messages but the "
                 "anthropic SDK is not installed — falling back to OpenAI-wire."
             )
-            return OpenAI(api_key=custom_key, base_url=custom_base), model
+            return OpenAI(api_key=custom_key, base_url=_clean_base, **_extra), model
         return (
             AnthropicAuxiliaryClient(real_client, model, custom_key, custom_base, is_oauth=False),
             model,
         )
-    return OpenAI(api_key=custom_key, base_url=custom_base), model
+    return OpenAI(api_key=custom_key, base_url=_clean_base, **_extra), model
 
 
 def _try_codex() -> Tuple[Optional[Any], Optional[str]]:
@@ -1349,6 +1363,49 @@ def _is_auth_error(exc: Exception) -> bool:
     return "error code: 401" in err_lower or "authenticationerror" in type(exc).__name__.lower()
 
 
+def _is_unsupported_parameter_error(exc: Exception, param: str) -> bool:
+    """Detect provider 400s for an unsupported request parameter.
+
+    Different OpenAI-compatible endpoints phrase the same class of error a few
+    ways: ``Unsupported parameter: X``, ``unsupported_parameter`` with a
+    ``param`` field, ``X is not supported``, ``unknown parameter: X``,
+    ``unrecognized request argument: X``.  We match on both the parameter
+    name and a generic "unsupported/unknown/unrecognized parameter" marker so
+    call sites can reactively retry without the offending key instead of
+    surfacing a noisy auxiliary failure.
+
+    Generalizes the temperature-specific detector that originally shipped
+    with PR #15621 so the same retry strategy can cover ``max_tokens``,
+    ``seed``, ``top_p``, and any future quirk. Credit @nicholasrae (PR #15416)
+    for the generalization pattern.
+    """
+    param_lower = (param or "").lower()
+    if not param_lower:
+        return False
+    err_lower = str(exc).lower()
+    if param_lower not in err_lower:
+        return False
+    return any(marker in err_lower for marker in (
+        "unsupported parameter",
+        "unsupported_parameter",
+        "not supported",
+        "does not support",
+        "unknown parameter",
+        "unrecognized request argument",
+        "unrecognized parameter",
+        "invalid parameter",
+    ))
+
+
+def _is_unsupported_temperature_error(exc: Exception) -> bool:
+    """Back-compat wrapper: detect API errors where the model rejects ``temperature``.
+
+    Delegates to :func:`_is_unsupported_parameter_error`; kept as a separate
+    public symbol because existing tests and call sites import it by name.
+    """
+    return _is_unsupported_parameter_error(exc, "temperature")
+
+
 def _evict_cached_clients(provider: str) -> None:
     """Drop cached auxiliary clients for a provider so fresh creds are used."""
     normalized = _normalize_aux_provider(provider)
@@ -1782,12 +1839,15 @@ def resolve_provider_client(
                 provider,
             )
             extra = {}
+            _clean_base, _dq = _extract_url_query_params(custom_base)
+            if _dq:
+                extra["default_query"] = _dq
             if base_url_host_matches(custom_base, "api.kimi.com"):
                 extra["default_headers"] = {"User-Agent": "claude-code/0.1.0"}
             elif base_url_host_matches(custom_base, "api.githubcopilot.com"):
                 from hermes_cli.models import copilot_default_headers
                 extra["default_headers"] = copilot_default_headers()
-            client = OpenAI(api_key=custom_key, base_url=custom_base, **extra)
+            client = OpenAI(api_key=custom_key, base_url=_clean_base, **extra)
             client = _wrap_if_needed(client, final_model, custom_base)
             return (_to_async_client(client, final_model) if async_mode
                     else (client, final_model))
@@ -1824,6 +1884,8 @@ def resolve_provider_client(
                     model or custom_entry.get("model") or _read_main_model() or "gpt-4o-mini",
                     provider,
                 )
+                _clean_base2, _dq2 = _extract_url_query_params(custom_base)
+                _extra2 = {"default_query": _dq2} if _dq2 else {}
                 logger.debug(
                     "resolve_provider_client: named custom provider %r (%s, api_mode=%s)",
                     provider, final_model, entry_api_mode or "chat_completions")
@@ -1841,7 +1903,7 @@ def resolve_provider_client(
                             "installed — falling back to OpenAI-wire.",
                             provider,
                         )
-                        client = OpenAI(api_key=custom_key, base_url=custom_base)
+                        client = OpenAI(api_key=custom_key, base_url=_clean_base2, **_extra2)
                         return (_to_async_client(client, final_model) if async_mode
                                 else (client, final_model))
                     sync_anthropic = AnthropicAuxiliaryClient(
@@ -1850,7 +1912,7 @@ def resolve_provider_client(
                     if async_mode:
                         return AsyncAnthropicAuxiliaryClient(sync_anthropic), final_model
                     return sync_anthropic, final_model
-                client = OpenAI(api_key=custom_key, base_url=custom_base)
+                client = OpenAI(api_key=custom_key, base_url=_clean_base2, **_extra2)
                 # codex_responses or inherited auto-detect (via _wrap_if_needed).
                 # _wrap_if_needed reads the closed-over `api_mode` (the task-level
                 # override). Named-provider entry api_mode=codex_responses also
@@ -2760,8 +2822,8 @@ def _build_call_kwargs(
         temperature = fixed_temperature
 
     # Opus 4.7+ rejects any non-default temperature/top_p/top_k — silently
-    # drop here so auxiliary callers that hardcode temperature (e.g. 0.3 on
-    # flush_memories, 0 on structured-JSON extraction) don't 400 the moment
+    # drop here so auxiliary callers that hardcode temperature (e.g. 0 on
+    # structured-JSON extraction) don't 400 the moment
     # the aux model is flipped to 4.7.
     if temperature is not None:
         from agent.anthropic_adapter import _forbids_sampling_params
@@ -2849,7 +2911,7 @@ def call_llm(
 
     Args:
         task: Auxiliary task name ("compression", "vision", "web_extract",
-              "session_search", "skills_hub", "mcp", "flush_memories").
+              "session_search", "skills_hub", "mcp", "title_generation").
               Reads provider:model from config/env. Ignored if provider is set.
         provider: Explicit provider override.
         model: Explicit model override.
@@ -2952,13 +3014,45 @@ def call_llm(
     if _is_anthropic_compat_endpoint(resolved_provider, _client_base):
         kwargs["messages"] = _convert_openai_images_to_anthropic(kwargs["messages"])
 
-    # Handle max_tokens vs max_completion_tokens retry, then payment fallback.
+    # Handle unsupported temperature, max_tokens vs max_completion_tokens retry,
+    # then payment fallback.
     try:
         return _validate_llm_response(
             client.chat.completions.create(**kwargs), task)
     except Exception as first_err:
+        if "temperature" in kwargs and _is_unsupported_temperature_error(first_err):
+            retry_kwargs = dict(kwargs)
+            retry_kwargs.pop("temperature", None)
+            logger.info(
+                "Auxiliary %s: provider rejected temperature; retrying once without it",
+                task or "call",
+            )
+            try:
+                return _validate_llm_response(
+                    client.chat.completions.create(**retry_kwargs), task)
+            except Exception as retry_err:
+                retry_err_str = str(retry_err)
+                # If retry still fails, fall through to the max_tokens /
+                # payment / auth chains below using the temperature-stripped
+                # kwargs.  Re-raise only if the retry hit something those
+                # chains won't handle.
+                if not (
+                    _is_payment_error(retry_err)
+                    or _is_connection_error(retry_err)
+                    or _is_auth_error(retry_err)
+                    or "max_tokens" in retry_err_str
+                    or "unsupported_parameter" in retry_err_str
+                ):
+                    raise
+                first_err = retry_err
+                kwargs = retry_kwargs
+
         err_str = str(first_err)
-        if "max_tokens" in err_str or "unsupported_parameter" in err_str:
+        if max_tokens is not None and (
+            "max_tokens" in err_str
+            or "unsupported_parameter" in err_str
+            or _is_unsupported_parameter_error(first_err, "max_tokens")
+        ):
             kwargs.pop("max_tokens", None)
             kwargs["max_completion_tokens"] = max_tokens
             try:
@@ -3221,8 +3315,35 @@ async def async_call_llm(
         return _validate_llm_response(
             await client.chat.completions.create(**kwargs), task)
     except Exception as first_err:
+        if "temperature" in kwargs and _is_unsupported_temperature_error(first_err):
+            retry_kwargs = dict(kwargs)
+            retry_kwargs.pop("temperature", None)
+            logger.info(
+                "Auxiliary %s (async): provider rejected temperature; retrying once without it",
+                task or "call",
+            )
+            try:
+                return _validate_llm_response(
+                    await client.chat.completions.create(**retry_kwargs), task)
+            except Exception as retry_err:
+                retry_err_str = str(retry_err)
+                if not (
+                    _is_payment_error(retry_err)
+                    or _is_connection_error(retry_err)
+                    or _is_auth_error(retry_err)
+                    or "max_tokens" in retry_err_str
+                    or "unsupported_parameter" in retry_err_str
+                ):
+                    raise
+                first_err = retry_err
+                kwargs = retry_kwargs
+
         err_str = str(first_err)
-        if "max_tokens" in err_str or "unsupported_parameter" in err_str:
+        if max_tokens is not None and (
+            "max_tokens" in err_str
+            or "unsupported_parameter" in err_str
+            or _is_unsupported_parameter_error(first_err, "max_tokens")
+        ):
             kwargs.pop("max_tokens", None)
             kwargs["max_completion_tokens"] = max_tokens
             try:

agent/codex_responses_adapter.py

@@ -44,22 +44,31 @@ _TOOL_CALL_LEAK_PATTERN = re.compile(
 # Multimodal content helpers
 # ---------------------------------------------------------------------------
 
-def _chat_content_to_responses_parts(content: Any) -> List[Dict[str, Any]]:
+def _chat_content_to_responses_parts(content: Any, *, role: str = "user") -> List[Dict[str, Any]]:
     """Convert chat-style multimodal content to Responses API input parts.
 
     Input:  ``[{"type":"text"|"image_url", ...}]`` (native OpenAI Chat format)
-    Output: ``[{"type":"input_text"|"input_image", ...}]`` (Responses format)
+    Output: ``[{"type":"input_text"|"output_text"|"input_image", ...}]`` (Responses format)
+
+    The ``role`` parameter controls the text content type:
+    - ``"user"`` (default) → ``"input_text"``
+    - ``"assistant"`` → ``"output_text"``
+
+    The Responses API rejects ``input_text`` inside assistant messages and
+    ``output_text`` inside user messages, so callers MUST pass the correct
+    role for the message being converted.
 
     Returns an empty list when ``content`` is not a list or contains no
     recognized parts — callers fall back to the string path.
     """
+    text_type = "output_text" if role == "assistant" else "input_text"
     if not isinstance(content, list):
         return []
     converted: List[Dict[str, Any]] = []
     for part in content:
         if isinstance(part, str):
             if part:
-                converted.append({"type": "input_text", "text": part})
+                converted.append({"type": text_type, "text": part})
             continue
         if not isinstance(part, dict):
             continue
@@ -67,7 +76,7 @@ def _chat_content_to_responses_parts(content: Any) -> List[Dict[str, Any]]:
         if ptype in {"text", "input_text", "output_text"}:
             text = part.get("text")
             if isinstance(text, str) and text:
-                converted.append({"type": "input_text", "text": text})
+                converted.append({"type": text_type, "text": text})
             continue
         if ptype in {"image_url", "input_image"}:
             image_ref = part.get("image_url")
@@ -218,6 +227,23 @@ def _responses_tools(tools: Optional[List[Dict[str, Any]]] = None) -> Optional[L
 # Message format conversion
 # ---------------------------------------------------------------------------
 
+_RESPONSE_MESSAGE_STATUSES = {"completed", "incomplete", "in_progress"}
+
+
+def _normalize_responses_message_status(value: Any, *, default: str = "completed") -> str:
+    """Normalize a Responses assistant message status for replay.
+
+    The API accepts completed/incomplete/in_progress on replayed assistant
+    output messages.  Preserve those exactly (modulo case/hyphen spelling) so
+    incomplete Codex continuation turns don't get falsely marked completed.
+    """
+    if isinstance(value, str):
+        status = value.strip().lower().replace("-", "_").replace(" ", "_")
+        if status in _RESPONSE_MESSAGE_STATUSES:
+            return status
+    return default
+
+
 def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
     """Convert internal chat-style messages to Responses input items."""
     items: List[Dict[str, Any]] = []
@@ -233,9 +259,10 @@ def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Di
         if role in {"user", "assistant"}:
             content = msg.get("content", "")
             if isinstance(content, list):
-                content_parts = _chat_content_to_responses_parts(content)
+                content_parts = _chat_content_to_responses_parts(content, role=role)
+                text_type = "output_text" if role == "assistant" else "input_text"
                 content_text = "".join(
-                    p.get("text", "") for p in content_parts if p.get("type") == "input_text"
+                    p.get("text", "") for p in content_parts if p.get("type") == text_type
                 )
             else:
                 content_parts = []
@@ -262,7 +289,57 @@ def _chat_messages_to_responses_input(messages: List[Dict[str, Any]]) -> List[Di
                                 seen_item_ids.add(item_id)
                             has_codex_reasoning = True
 
-                if content_parts:
+                # Replay exact assistant message items (with id/phase) from
+                # previous turns so the API can maintain prefix-cache hits.
+                # OpenAI docs: "preserve and resend phase on all assistant
+                # messages — dropping it can degrade performance."
+                codex_message_items = msg.get("codex_message_items")
+                replayed_message_items = 0
+                if isinstance(codex_message_items, list):
+                    for raw_item in codex_message_items:
+                        if not isinstance(raw_item, dict):
+                            continue
+                        if raw_item.get("type") != "message" or raw_item.get("role") != "assistant":
+                            continue
+                        raw_content_parts = raw_item.get("content")
+                        if not isinstance(raw_content_parts, list):
+                            continue
+
+                        normalized_content_parts = []
+                        for part in raw_content_parts:
+                            if not isinstance(part, dict):
+                                continue
+                            part_type = str(part.get("type") or "").strip()
+                            if part_type not in {"output_text", "text"}:
+                                continue
+                            text = part.get("text", "")
+                            if text is None:
+                                text = ""
+                            if not isinstance(text, str):
+                                text = str(text)
+                            normalized_content_parts.append({"type": "output_text", "text": text})
+
+                        if not normalized_content_parts:
+                            continue
+
+                        replay_item = {
+                            "type": "message",
+                            "role": "assistant",
+                            "status": _normalize_responses_message_status(raw_item.get("status")),
+                            "content": normalized_content_parts,
+                        }
+                        item_id = raw_item.get("id")
+                        if isinstance(item_id, str) and item_id.strip():
+                            replay_item["id"] = item_id.strip()
+                        phase = raw_item.get("phase")
+                        if isinstance(phase, str) and phase.strip():
+                            replay_item["phase"] = phase.strip()
+                        items.append(replay_item)
+                        replayed_message_items += 1
+
+                if replayed_message_items > 0:
+                    pass
+                elif content_parts:
                     items.append({"role": "assistant", "content": content_parts})
                 elif content_text.strip():
                     items.append({"role": "assistant", "content": content_text})
@@ -422,6 +499,47 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                 normalized.append(reasoning_item)
             continue
 
+        if item_type == "message":
+            role = item.get("role")
+            if role != "assistant":
+                raise ValueError(f"Codex Responses input[{idx}] message items must have role='assistant'.")
+            content = item.get("content")
+            if not isinstance(content, list):
+                raise ValueError(f"Codex Responses input[{idx}] message item must have content list.")
+            normalized_content = []
+            for part_idx, part in enumerate(content):
+                if not isinstance(part, dict):
+                    raise ValueError(
+                        f"Codex Responses input[{idx}] message content[{part_idx}] must be an object."
+                    )
+                part_type = part.get("type")
+                if part_type not in {"output_text", "text"}:
+                    raise ValueError(
+                        f"Codex Responses input[{idx}] message content[{part_idx}] has unsupported type {part_type!r}."
+                    )
+                text = part.get("text", "")
+                if text is None:
+                    text = ""
+                if not isinstance(text, str):
+                    text = str(text)
+                normalized_content.append({"type": "output_text", "text": text})
+            if not normalized_content:
+                raise ValueError(f"Codex Responses input[{idx}] message item must contain at least one text part.")
+            normalized_item: Dict[str, Any] = {
+                "type": "message",
+                "role": "assistant",
+                "status": _normalize_responses_message_status(item.get("status")),
+                "content": normalized_content,
+            }
+            item_id = item.get("id")
+            if isinstance(item_id, str) and item_id.strip():
+                normalized_item["id"] = item_id.strip()
+            phase = item.get("phase")
+            if isinstance(phase, str) and phase.strip():
+                normalized_item["phase"] = phase.strip()
+            normalized.append(normalized_item)
+            continue
+
         role = item.get("role")
         if role in {"user", "assistant"}:
             content = item.get("content", "")
@@ -429,13 +547,16 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                 content = ""
             if isinstance(content, list):
                 # Multimodal content from ``_chat_messages_to_responses_input``
-                # is already in Responses format (``input_text`` / ``input_image``).
-                # Validate each part and pass through.
+                # is already in Responses format (``input_text`` / ``output_text``
+                # / ``input_image``).  Validate each part and pass through.
+                # Use the correct text type for the role — ``output_text`` for
+                # assistant messages, ``input_text`` for user messages.
+                text_type = "output_text" if role == "assistant" else "input_text"
                 validated: List[Dict[str, Any]] = []
                 for part_idx, part in enumerate(content):
                     if isinstance(part, str):
                         if part:
-                            validated.append({"type": "input_text", "text": part})
+                            validated.append({"type": text_type, "text": part})
                         continue
                     if not isinstance(part, dict):
                         raise ValueError(
@@ -446,7 +567,7 @@ def _preflight_codex_input_items(raw_items: Any) -> List[Dict[str, Any]]:
                         text = part.get("text", "")
                         if not isinstance(text, str):
                             text = str(text or "")
-                        validated.append({"type": "input_text", "text": text})
+                        validated.append({"type": text_type, "text": text})
                     elif ptype in {"input_image", "image_url"}:
                         image_ref = part.get("image_url", "")
                         detail = part.get("detail")
@@ -703,6 +824,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
     content_parts: List[str] = []
     reasoning_parts: List[str] = []
     reasoning_items_raw: List[Dict[str, Any]] = []
+    message_items_raw: List[Dict[str, Any]] = []
     tool_calls: List[Any] = []
     has_incomplete_items = response_status in {"queued", "in_progress", "incomplete"}
     saw_commentary_phase = False
@@ -721,6 +843,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
 
         if item_type == "message":
             item_phase = getattr(item, "phase", None)
+            normalized_phase = None
             if isinstance(item_phase, str):
                 normalized_phase = item_phase.strip().lower()
                 if normalized_phase in {"commentary", "analysis"}:
@@ -730,6 +853,18 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
             message_text = _extract_responses_message_text(item)
             if message_text:
                 content_parts.append(message_text)
+                raw_message_item: Dict[str, Any] = {
+                    "type": "message",
+                    "role": "assistant",
+                    "status": _normalize_responses_message_status(item_status),
+                    "content": [{"type": "output_text", "text": message_text}],
+                }
+                item_id = getattr(item, "id", None)
+                if isinstance(item_id, str) and item_id:
+                    raw_message_item["id"] = item_id
+                if normalized_phase:
+                    raw_message_item["phase"] = normalized_phase
+                message_items_raw.append(raw_message_item)
         elif item_type == "reasoning":
             reasoning_text = _extract_responses_reasoning_text(item)
             if reasoning_text:
@@ -842,6 +977,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
         reasoning_content=None,
         reasoning_details=None,
         codex_reasoning_items=reasoning_items_raw or None,
+        codex_message_items=message_items_raw or None,
     )
 
     if tool_calls:

agent/credential_pool.py

@@ -14,6 +14,7 @@ from datetime import datetime
 from typing import Any, Dict, List, Optional, Set, Tuple
 
 from hermes_constants import OPENROUTER_BASE_URL
+from hermes_cli.config import get_env_value
 import hermes_cli.auth as auth_mod
 from hermes_cli.auth import (
     CODEX_ACCESS_TOKEN_REFRESH_SKEW_SECONDS,
@@ -1273,7 +1274,8 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
         def _is_source_suppressed(_p, _s):  # type: ignore[misc]
             return False
     if provider == "openrouter":
-        token = os.getenv("OPENROUTER_API_KEY", "").strip()
+        # Check both os.environ and ~/.hermes/.env file
+        token = (get_env_value("OPENROUTER_API_KEY") or "").strip()
         if token:
             source = "env:OPENROUTER_API_KEY"
             if _is_source_suppressed(provider, source):
@@ -1299,7 +1301,7 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
 
     env_url = ""
     if pconfig.base_url_env_var:
-        env_url = os.getenv(pconfig.base_url_env_var, "").strip().rstrip("/")
+        env_url = (get_env_value(pconfig.base_url_env_var) or "").strip().rstrip("/")
 
     env_vars = list(pconfig.api_key_env_vars)
     if provider == "anthropic":
@@ -1310,7 +1312,8 @@ def _seed_from_env(provider: str, entries: List[PooledCredential]) -> Tuple[bool
         ]
 
     for env_var in env_vars:
-        token = os.getenv(env_var, "").strip()
+        # Check both os.environ and ~/.hermes/.env file
+        token = (get_env_value(env_var) or "").strip()
         if not token:
             continue
         source = f"env:{env_var}"

agent/model_metadata.py

@@ -106,9 +106,11 @@ _endpoint_model_metadata_cache_time: Dict[str, float] = {}
 _ENDPOINT_MODEL_CACHE_TTL = 300
 
 # Descending tiers for context length probing when the model is unknown.
-# We start at 128K (a safe default for most modern models) and step down
-# on context-length errors until one works.
+# We start at 256K (covers GPT-5.x, many current large-context models) and
+# step down on context-length errors until one works.  Tier[0] is also the
+# default fallback when no detection method succeeds.
 CONTEXT_PROBE_TIERS = [
+    256_000,
     128_000,
     64_000,
     32_000,
@@ -143,10 +145,11 @@ DEFAULT_CONTEXT_LENGTHS = {
     "claude": 200000,
     # OpenAI — GPT-5 family (most have 400k; specific overrides first)
     # Source: https://developers.openai.com/api/docs/models
-    # GPT-5.5 (launched Apr 23 2026). 400k is the fallback for providers we
-    # can't probe live. ChatGPT Codex OAuth actually caps lower (272k as of
-    # Apr 2026) and is resolved via _resolve_codex_oauth_context_length().
-    "gpt-5.5": 400000,
+    # GPT-5.5 (launched Apr 23 2026) is 1.05M on the direct OpenAI API and
+    # ChatGPT Codex OAuth caps it at 272K; both paths resolve via their own
+    # provider-aware branches (_resolve_codex_oauth_context_length + models.dev).
+    # This hardcoded value is only reached when every probe misses.
+    "gpt-5.5": 1050000,
     "gpt-5.4-nano": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4-mini": 400000,           # 400k (not 1.05M like full 5.4)
     "gpt-5.4": 1050000,               # GPT-5.4, GPT-5.4 Pro (1.05M context)
@@ -162,7 +165,17 @@ DEFAULT_CONTEXT_LENGTHS = {
     "gemma-4-31b": 256000,
     "gemma-3": 131072,
     "gemma": 8192,  # fallback for older gemma models
-    # DeepSeek
+    # DeepSeek — V4 family ships with a 1M context window. The legacy
+    # aliases ``deepseek-chat`` / ``deepseek-reasoner`` are server-side
+    # mapped to the non-thinking / thinking modes of ``deepseek-v4-flash``
+    # and inherit the same 1M window. The ``deepseek`` substring entry
+    # below remains as a 128K fallback for older / unknown DeepSeek model
+    # ids (e.g. via custom endpoints).
+    # https://api-docs.deepseek.com/zh-cn/quick_start/pricing
+    "deepseek-v4-pro": 1_000_000,
+    "deepseek-v4-flash": 1_000_000,
+    "deepseek-chat": 1_000_000,
+    "deepseek-reasoner": 1_000_000,
     "deepseek": 128000,
     # Meta
     "llama": 131072,
@@ -1193,6 +1206,7 @@ def get_model_context_length(
     api_key: str = "",
     config_context_length: int | None = None,
     provider: str = "",
+    custom_providers: list | None = None,
 ) -> int:
     """Get the context length for a model.
 
@@ -1213,6 +1227,23 @@ def get_model_context_length(
     if config_context_length is not None and isinstance(config_context_length, int) and config_context_length > 0:
         return config_context_length
 
+    # 0b. custom_providers per-model override — check before any probe.
+    # This closes the gap where /model switch and display paths used to fall
+    # back to 128K despite the user having a per-model context_length set.
+    # See #15779.
+    if custom_providers and base_url and model:
+        try:
+            from hermes_cli.config import get_custom_provider_context_length
+            cp_ctx = get_custom_provider_context_length(
+                model=model,
+                base_url=base_url,
+                custom_providers=custom_providers,
+            )
+            if cp_ctx:
+                return cp_ctx
+        except Exception:
+            pass  # fall through to probing
+
     # Normalise provider-prefixed model names (e.g. "local:model-name" →
     # "model-name") so cache lookups and server queries use the bare ID that
     # local servers actually know about.  Ollama "model:tag" colons are preserved.
@@ -1352,7 +1383,7 @@ def get_model_context_length(
     # 6. OpenRouter live API metadata (provider-unaware fallback)
     metadata = fetch_model_metadata()
     if model in metadata:
-        return metadata[model].get("context_length", 128000)
+        return metadata[model].get("context_length", DEFAULT_FALLBACK_CONTEXT)
 
     # 8. Hardcoded defaults (fuzzy match — longest key first for specificity)
     # Only check `default_model in model` (is the key a substring of the input).

agent/nous_rate_guard.py

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

agent/onboarding.py

@@ -0,0 +1,191 @@
+"""
+Contextual first-touch onboarding hints.
+
+Instead of blocking first-run questionnaires, show a one-time hint the *first*
+time a user hits a behavior fork — message-while-running, first long-running
+tool, etc.  Each hint is shown once per install (tracked in ``config.yaml`` under
+``onboarding.seen.<flag>``) and then never again.
+
+Keep this module tiny and dependency-free so both the CLI and gateway can import
+it without pulling in heavy modules.
+"""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Any, Mapping, Optional
+
+logger = logging.getLogger(__name__)
+
+
+# -------------------------------------------------------------------------
+# Flag names (stable — used as config.yaml keys under onboarding.seen)
+# -------------------------------------------------------------------------
+
+BUSY_INPUT_FLAG = "busy_input_prompt"
+TOOL_PROGRESS_FLAG = "tool_progress_prompt"
+OPENCLAW_RESIDUE_FLAG = "openclaw_residue_cleanup"
+
+
+# -------------------------------------------------------------------------
+# Hint content
+# -------------------------------------------------------------------------
+
+def busy_input_hint_gateway(mode: str) -> str:
+    """Hint shown the first time a user messages while the agent is busy.
+
+    ``mode`` is the effective busy_input_mode that was just applied, so the
+    message matches reality ("I just interrupted…" vs "I just queued…").
+    """
+    if mode == "queue":
+        return (
+            "💡 First-time tip — I queued your message instead of interrupting. "
+            "Send `/busy interrupt` to make new messages stop the current task "
+            "immediately, or `/busy status` to check. This notice won't appear again."
+        )
+    if mode == "steer":
+        return (
+            "💡 First-time tip — I steered your message into the current run; "
+            "it will arrive after the next tool call instead of interrupting. "
+            "Send `/busy interrupt` or `/busy queue` to change this, or "
+            "`/busy status` to check. This notice won't appear again."
+        )
+    return (
+        "💡 First-time tip — I just interrupted my current task to answer you. "
+        "Send `/busy queue` to queue follow-ups for after the current task instead, "
+        "`/busy steer` to inject them mid-run without interrupting, or "
+        "`/busy status` to check. This notice won't appear again."
+    )
+
+
+def busy_input_hint_cli(mode: str) -> str:
+    """CLI version of the busy-input hint (plain text, no markdown)."""
+    if mode == "queue":
+        return (
+            "(tip) Your message was queued for the next turn. "
+            "Use /busy interrupt to make Enter stop the current run instead, "
+            "or /busy steer to inject mid-run. This tip only shows once."
+        )
+    if mode == "steer":
+        return (
+            "(tip) Your message was steered into the current run; it arrives "
+            "after the next tool call. Use /busy interrupt or /busy queue to "
+            "change this. This tip only shows once."
+        )
+    return (
+        "(tip) Your message interrupted the current run. "
+        "Use /busy queue to queue messages for the next turn instead, "
+        "or /busy steer to inject mid-run. This tip only shows once."
+    )
+
+
+def tool_progress_hint_gateway() -> str:
+    return (
+        "💡 First-time tip — that tool took a while and I'm streaming every step. "
+        "If the progress messages feel noisy, send `/verbose` to cycle modes "
+        "(all → new → off). This notice won't appear again."
+    )
+
+
+def tool_progress_hint_cli() -> str:
+    return (
+        "(tip) That tool ran for a while. Use /verbose to cycle tool-progress "
+        "display modes (all -> new -> off -> verbose). This tip only shows once."
+    )
+
+
+def openclaw_residue_hint_cli() -> str:
+    """Banner shown the first time Hermes starts and finds ``~/.openclaw/``.
+
+    OpenClaw-era config, memory, and skill paths in ``~/.openclaw/`` will
+    otherwise attract the agent (memory entries like ``~/.openclaw/config.yaml``
+    get carried forward and the agent dutifully reads them). ``hermes claw
+    cleanup`` renames the directory so the agent stops finding it.
+    """
+    return (
+        "Heads up — an OpenClaw workspace was detected at ~/.openclaw/.\n"
+        "After migrating, the agent can still get confused and read that "
+        "directory's config/memory instead of Hermes's.\n"
+        "Run `hermes claw cleanup` to archive it (rename → .openclaw.pre-migration). "
+        "This tip only shows once; rerun it any time with `hermes claw cleanup`."
+    )
+
+
+def detect_openclaw_residue(home: Optional[Path] = None) -> bool:
+    """Return True if an OpenClaw workspace directory is present in ``$HOME``.
+
+    Pure filesystem check — no side effects. ``home`` override exists for tests.
+    """
+    base = home or Path.home()
+    try:
+        return (base / ".openclaw").is_dir()
+    except OSError:
+        return False
+
+
+# -------------------------------------------------------------------------
+# State read / write
+# -------------------------------------------------------------------------
+
+def _get_seen_dict(config: Mapping[str, Any]) -> Mapping[str, Any]:
+    onboarding = config.get("onboarding") if isinstance(config, Mapping) else None
+    if not isinstance(onboarding, Mapping):
+        return {}
+    seen = onboarding.get("seen")
+    return seen if isinstance(seen, Mapping) else {}
+
+
+def is_seen(config: Mapping[str, Any], flag: str) -> bool:
+    """Return True if the user has already been shown this first-touch hint."""
+    return bool(_get_seen_dict(config).get(flag))
+
+
+def mark_seen(config_path: Path, flag: str) -> bool:
+    """Persist ``onboarding.seen.<flag> = True`` to ``config_path``.
+
+    Uses the atomic YAML writer so a concurrent process can't observe a
+    partially-written file.  Returns True on success, False on any error
+    (including the config file being absent — onboarding is best-effort).
+    """
+    try:
+        import yaml
+        from utils import atomic_yaml_write
+    except Exception as e:  # pragma: no cover — dependency issue
+        logger.debug("onboarding: failed to import yaml/utils: %s", e)
+        return False
+
+    try:
+        cfg: dict = {}
+        if config_path.exists():
+            with open(config_path, encoding="utf-8") as f:
+                cfg = yaml.safe_load(f) or {}
+        if not isinstance(cfg.get("onboarding"), dict):
+            cfg["onboarding"] = {}
+        seen = cfg["onboarding"].get("seen")
+        if not isinstance(seen, dict):
+            seen = {}
+            cfg["onboarding"]["seen"] = seen
+        if seen.get(flag) is True:
+            return True  # already marked — nothing to do
+        seen[flag] = True
+        atomic_yaml_write(config_path, cfg)
+        return True
+    except Exception as e:
+        logger.debug("onboarding: failed to mark flag %s: %s", flag, e)
+        return False
+
+
+__all__ = [
+    "BUSY_INPUT_FLAG",
+    "TOOL_PROGRESS_FLAG",
+    "OPENCLAW_RESIDUE_FLAG",
+    "busy_input_hint_gateway",
+    "busy_input_hint_cli",
+    "tool_progress_hint_gateway",
+    "tool_progress_hint_cli",
+    "openclaw_residue_hint_cli",
+    "detect_openclaw_residue",
+    "is_seen",
+    "mark_seen",
+]

agent/prompt_builder.py

@@ -422,6 +422,29 @@ PLATFORM_HINTS = {
         "your response. Images are sent as native photos, and other files arrive as downloadable "
         "documents."
     ),
+    "yuanbao": (
+        "You are on Yuanbao (腾讯元宝), a Chinese AI assistant platform. "
+        "Markdown formatting is supported (code blocks, tables, bold/italic). "
+        "You CAN send media files natively — to deliver a file to the user, include "
+        "MEDIA:/absolute/path/to/file in your response. The file will be sent as a native "
+        "Yuanbao attachment: images (.jpg, .png, .webp, .gif) are sent as photos, "
+        "and other files (.pdf, .docx, .txt, .zip, etc.) arrive as downloadable documents "
+        "(max 50 MB). You can also include image URLs in markdown format ![alt](url) and "
+        "they will be downloaded and sent as native photos. "
+        "Do NOT tell the user you lack file-sending capability — use MEDIA: syntax "
+        "whenever a file delivery is appropriate.\n\n"
+        "Stickers (贴纸 / 表情包 / TIM face): Yuanbao has a built-in sticker catalogue. "
+        "When the user sends a sticker (you see '[emoji: 名称]' in their message) or asks "
+        "you to send/reply-with a 贴纸/表情/表情包, you MUST use the sticker tools:\n"
+        "  1. Call yb_search_sticker with a Chinese keyword (e.g. '666', '比心', '吃瓜', "
+        "     '捂脸', '合十') to discover matching sticker_ids.\n"
+        "  2. Call yb_send_sticker with the chosen sticker_id or name — this sends a real "
+        "     TIMFaceElem that renders as a native sticker in the chat.\n"
+        "DO NOT draw sticker-like PNGs with execute_code/Pillow/matplotlib and then send "
+        "them via MEDIA: or send_image_file. That produces a fake low-quality 'sticker' "
+        "image and is the WRONG path. Bare Unicode emoji in text is also not a substitute "
+        "— when a sticker is the right response, use yb_send_sticker."
+    ),
 }
 
 # ---------------------------------------------------------------------------
@@ -825,6 +848,11 @@ def build_skills_system_prompt(
             "Skills also encode the user's preferred approach, conventions, and quality standards "
             "for tasks like code review, planning, and testing — load them even for tasks you "
             "already know how to do, because the skill defines how it should be done here.\n"
+            "Whenever the user asks you to configure, set up, install, enable, disable, modify, "
+            "or troubleshoot Hermes Agent itself — its CLI, config, models, providers, tools, "
+            "skills, voice, gateway, plugins, or any feature — load the `hermes-agent` skill "
+            "first. It has the actual commands (e.g. `hermes config set …`, `hermes tools`, "
+            "`hermes setup`) so you don't have to guess or invent workarounds.\n"
             "If a skill has issues, fix it with skill_manage(action='patch').\n"
             "After difficult/iterative tasks, offer to save as a skill. "
             "If a skill you loaded was missing steps, had wrong commands, or needed "

agent/shell_hooks.py

@@ -754,7 +754,11 @@ def _resolve_effective_accept(
     if env in ("1", "true", "yes", "on"):
         return True
     cfg_val = cfg.get("hooks_auto_accept", False)
-    return bool(cfg_val)
+    if isinstance(cfg_val, bool):
+        return cfg_val
+    if isinstance(cfg_val, str):
+        return cfg_val.strip().lower() in ("1", "true", "yes", "on")
+    return False
 
 
 # ---------------------------------------------------------------------------

agent/skill_commands.py

@@ -329,7 +329,7 @@ def build_skill_invocation_message(
 
     loaded_skill, skill_dir, skill_name = loaded
     activation_note = (
-        f'[SYSTEM: The user has invoked the "{skill_name}" skill, indicating they want '
+        f'[IMPORTANT: The user has invoked the "{skill_name}" skill, indicating they want '
         "you to follow its instructions. The full skill content is loaded below.]"
     )
     return _build_skill_message(
@@ -368,7 +368,7 @@ def build_preloaded_skills_prompt(
 
         loaded_skill, skill_dir, skill_name = loaded
         activation_note = (
-            f'[SYSTEM: The user launched this CLI session with the "{skill_name}" skill '
+            f'[IMPORTANT: The user launched this CLI session with the "{skill_name}" skill '
             "preloaded. Treat its instructions as active guidance for the duration of this "
             "session unless the user overrides them.]"
         )

agent/transports/__init__.py

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

agent/transports/chat_completions.py

@@ -31,15 +31,15 @@ class ChatCompletionsTransport(ProviderTransport):
     def convert_messages(self, messages: List[Dict[str, Any]], **kwargs) -> List[Dict[str, Any]]:
         """Messages are already in OpenAI format — sanitize Codex leaks only.
 
-        Strips Codex Responses API fields (``codex_reasoning_items`` on the
-        message, ``call_id``/``response_item_id`` on tool_calls) that strict
-        chat-completions providers reject with 400/422.
+        Strips Codex Responses API fields (``codex_reasoning_items`` /
+        ``codex_message_items`` on the message, ``call_id``/``response_item_id``
+        on tool_calls) that strict chat-completions providers reject with 400/422.
         """
         needs_sanitize = False
         for msg in messages:
             if not isinstance(msg, dict):
                 continue
-            if "codex_reasoning_items" in msg:
+            if "codex_reasoning_items" in msg or "codex_message_items" in msg:
                 needs_sanitize = True
                 break
             tool_calls = msg.get("tool_calls")
@@ -59,6 +59,7 @@ class ChatCompletionsTransport(ProviderTransport):
             if not isinstance(msg, dict):
                 continue
             msg.pop("codex_reasoning_items", None)
+            msg.pop("codex_message_items", None)
             tool_calls = msg.get("tool_calls")
             if isinstance(tool_calls, list):
                 for tc in tool_calls:

agent/transports/codex.py

@@ -120,6 +120,24 @@ class ResponsesApiTransport(ProviderTransport):
         if request_overrides:
             kwargs.update(request_overrides)
 
+        if is_codex_backend:
+            prompt_cache_key = kwargs.get("prompt_cache_key")
+            cache_scope_id = str(prompt_cache_key or session_id or "").strip()
+            if cache_scope_id:
+                existing_extra_headers = kwargs.get("extra_headers")
+                merged_extra_headers: Dict[str, str] = {}
+                if isinstance(existing_extra_headers, dict):
+                    merged_extra_headers.update(
+                        {
+                            str(key): str(value)
+                            for key, value in existing_extra_headers.items()
+                            if key and value is not None
+                        }
+                    )
+                merged_extra_headers["session_id"] = cache_scope_id
+                merged_extra_headers["x-client-request-id"] = cache_scope_id
+                kwargs["extra_headers"] = merged_extra_headers
+
         max_tokens = params.get("max_tokens")
         if max_tokens is not None and not is_codex_backend:
             kwargs["max_output_tokens"] = max_tokens
@@ -160,6 +178,8 @@ class ResponsesApiTransport(ProviderTransport):
         provider_data = {}
         if msg and hasattr(msg, "codex_reasoning_items") and msg.codex_reasoning_items:
             provider_data["codex_reasoning_items"] = msg.codex_reasoning_items
+        if msg and hasattr(msg, "codex_message_items") and msg.codex_message_items:
+            provider_data["codex_message_items"] = msg.codex_message_items
         if msg and hasattr(msg, "reasoning_details") and msg.reasoning_details:
             provider_data["reasoning_details"] = msg.reasoning_details
 

agent/transports/types.py

@@ -97,7 +97,7 @@ class NormalizedResponse:
     Response-level ``provider_data`` examples:
 
     * Anthropic: ``{"reasoning_details": [...]}``
-    * Codex: ``{"codex_reasoning_items": [...]}``
+    * Codex: ``{"codex_reasoning_items": [...], "codex_message_items": [...]}``
     * Others: ``None``
     """
 
@@ -126,6 +126,11 @@ class NormalizedResponse:
         pd = self.provider_data or {}
         return pd.get("codex_reasoning_items")
 
+    @property
+    def codex_message_items(self):
+        pd = self.provider_data or {}
+        return pd.get("codex_message_items")
+
 
 # ---------------------------------------------------------------------------
 # Factory helpers

cli-config.yaml.example

@@ -606,6 +606,7 @@ platform_toolsets:
   signal: [hermes-signal]
   homeassistant: [hermes-homeassistant]
   qqbot: [hermes-qqbot]
+  yuanbao: [hermes-yuanbao]
 
 # =============================================================================
 # Gateway Platform Settings
@@ -824,7 +825,9 @@ delegation:
 # Display
 # =============================================================================
 display:
-  # Use compact banner mode
+  # Use compact banner mode (hides the ASCII-art banner, shows a single line).
+  #   true:  Compact single-line banner
+  #   false: Full ASCII banner with tool/skill summary (default)
   compact: false
 
   # Tool progress display level (CLI and gateway)
@@ -838,12 +841,19 @@ display:
   # Gateway-only natural mid-turn assistant updates.
   # When true, completed assistant status messages are sent as separate chat
   # messages. This is independent of tool_progress and gateway streaming.
+  #   true:  Send mid-turn assistant updates as separate messages (default)
+  #   false: Only send the final response
   interim_assistant_messages: true
 
-  # What Enter does when Hermes is already busy in the CLI.
+  # What Enter does when Hermes is already busy (CLI and gateway platforms).
   #   interrupt: Interrupt the current run and redirect Hermes (default)
   #   queue:     Queue your message for the next turn
-  # Ctrl+C always interrupts regardless of this setting.
+  #   steer:     Inject your message mid-run via /steer, arriving at the agent
+  #              after the next tool call — no interrupt, no role violation.
+  #              Falls back to 'queue' if the agent isn't running yet or if
+  #              images are attached (steer only carries text).
+  # Ctrl+C (or /stop in gateway) always interrupts regardless of this setting.
+  # Toggle at runtime with /busy <interrupt|queue|steer>.
   busy_input_mode: interrupt
 
   # Background process notifications (gateway/messaging only).
@@ -859,17 +869,22 @@ display:
   # Play terminal bell when agent finishes a response.
   # Useful for long-running tasks — your terminal will ding when the agent is done.
   # Works over SSH. Most terminals can be configured to flash the taskbar or play a sound.
+  #   true:  Ring the terminal bell on each response
+  #   false: Silent (default)
   bell_on_complete: false
 
   # Show model reasoning/thinking before each response.
   # When enabled, a dim box shows the model's thought process above the response.
   # Toggle at runtime with /reasoning show or /reasoning hide.
+  #   true:  Show the reasoning box
+  #   false: Hide reasoning (default)
   show_reasoning: false
 
   # Stream tokens to the terminal as they arrive instead of waiting for the
   # full response. The response box opens on first token and text appears
   # line-by-line. Tool calls are still captured silently.
-  # Stream tokens to the terminal in real-time. Disable to wait for full responses.
+  #   true:  Stream tokens as they arrive (default)
+  #   false: Wait for the full response before rendering
   streaming: true
 
   # ───────────────────────────────────────────────────────────────────────────
@@ -879,10 +894,15 @@ display:
   # response box label, and branding text. Change at runtime with /skin <name>.
   #
   # Built-in skins:
-  #   default  — Classic Hermes gold/kawaii
-  #   ares     — Crimson/bronze war-god theme with spinner wings
-  #   mono     — Clean grayscale monochrome
-  #   slate    — Cool blue developer-focused
+  #   default        — Classic Hermes gold/kawaii
+  #   ares           — Crimson/bronze war-god theme with spinner wings
+  #   mono           — Clean grayscale monochrome
+  #   slate          — Cool blue developer-focused
+  #   daylight       — Bright light-mode theme
+  #   warm-lightmode — Warm paper-tone light-mode theme
+  #   poseidon       — Sea-green/teal Olympian theme
+  #   sisyphus       — Earthy stone-and-moss theme
+  #   charizard      — Fiery orange dragon theme
   #
   # Custom skins: drop a YAML file in ~/.hermes/skins/<name>.yaml
   # Schema (all fields optional, missing values inherit from default):

cli.py

@@ -22,6 +22,7 @@ import re
 import concurrent.futures
 import base64
 import atexit
+import errno
 import tempfile
 import time
 import uuid
@@ -416,6 +417,11 @@ def load_cli_config() -> Dict[str, Any]:
             "base_url": "",    # Direct OpenAI-compatible endpoint for subagents
             "api_key": "",     # API key for delegation.base_url (falls back to OPENAI_API_KEY)
         },
+        "onboarding": {
+            # First-touch hint flags (see agent/onboarding.py).  Each hint is
+            # shown once per install then latched here.
+            "seen": {},
+        },
     }
     
     # Track whether the config file explicitly set terminal config.
@@ -968,6 +974,7 @@ def _run_state_db_auto_maintenance(session_db) -> None:
         return
     try:
         from hermes_cli.config import load_config as _load_full_config
+        from hermes_constants import get_hermes_home as _get_hermes_home
         cfg = (_load_full_config().get("sessions") or {})
         if not cfg.get("auto_prune", False):
             return
@@ -975,11 +982,35 @@ def _run_state_db_auto_maintenance(session_db) -> None:
             retention_days=int(cfg.get("retention_days", 90)),
             min_interval_hours=int(cfg.get("min_interval_hours", 24)),
             vacuum=bool(cfg.get("vacuum_after_prune", True)),
+            sessions_dir=_get_hermes_home() / "sessions",
         )
     except Exception as exc:
         logger.debug("state.db auto-maintenance skipped: %s", exc)
 
 
+def _run_checkpoint_auto_maintenance() -> None:
+    """Call ``checkpoint_manager.maybe_auto_prune_checkpoints`` using current config.
+
+    Reads the ``checkpoints:`` section from config.yaml via
+    :func:`hermes_cli.config.load_config`. Honours ``auto_prune`` /
+    ``retention_days`` / ``delete_orphans`` / ``min_interval_hours``.
+    Never raises — maintenance must never block interactive startup.
+    """
+    try:
+        from hermes_cli.config import load_config as _load_full_config
+        cfg = (_load_full_config().get("checkpoints") or {})
+        if not cfg.get("auto_prune", False):
+            return
+        from tools.checkpoint_manager import maybe_auto_prune_checkpoints
+        maybe_auto_prune_checkpoints(
+            retention_days=int(cfg.get("retention_days", 7)),
+            min_interval_hours=int(cfg.get("min_interval_hours", 24)),
+            delete_orphans=bool(cfg.get("delete_orphans", True)),
+        )
+    except Exception as exc:
+        logger.debug("checkpoint auto-maintenance skipped: %s", exc)
+
+
 def _prune_stale_worktrees(repo_root: str, max_age_hours: int = 24) -> None:
     """Remove stale worktrees and orphaned branches on startup.
 
@@ -1372,7 +1403,7 @@ def _resolve_attachment_path(raw_path: str) -> Path | None:
 
 
 def _format_process_notification(evt: dict) -> "str | None":
-    """Format a process notification event into a [SYSTEM: ...] message.
+    """Format a process notification event into a [IMPORTANT: ...] message.
 
     Handles both completion events (notify_on_complete) and watch pattern
     match events from the unified completion_queue.
@@ -1382,14 +1413,14 @@ def _format_process_notification(evt: dict) -> "str | None":
     _cmd = evt.get("command", "unknown")
 
     if evt_type == "watch_disabled":
-        return f"[SYSTEM: {evt.get('message', '')}]"
+        return f"[IMPORTANT: {evt.get('message', '')}]"
 
     if evt_type == "watch_match":
         _pat = evt.get("pattern", "?")
         _out = evt.get("output", "")
         _sup = evt.get("suppressed", 0)
         text = (
-            f"[SYSTEM: Background process {_sid} matched "
+            f"[IMPORTANT: Background process {_sid} matched "
             f"watch pattern \"{_pat}\".\n"
             f"Command: {_cmd}\n"
             f"Matched output:\n{_out}"
@@ -1403,7 +1434,7 @@ def _format_process_notification(evt: dict) -> "str | None":
     _exit = evt.get("exit_code", "?")
     _out = evt.get("output", "")
     return (
-        f"[SYSTEM: Background process {_sid} completed "
+        f"[IMPORTANT: Background process {_sid} completed "
         f"(exit code {_exit}).\n"
         f"Command: {_cmd}\n"
         f"Output:\n{_out}]"
@@ -1842,9 +1873,16 @@ class HermesCLI:
         self.bell_on_complete = CLI_CONFIG["display"].get("bell_on_complete", False)
         # show_reasoning: display model thinking/reasoning before the response
         self.show_reasoning = CLI_CONFIG["display"].get("show_reasoning", False)
-        # busy_input_mode: "interrupt" (Enter interrupts current run) or "queue" (Enter queues for next turn)
-        _bim = CLI_CONFIG["display"].get("busy_input_mode", "interrupt")
-        self.busy_input_mode = "queue" if str(_bim).strip().lower() == "queue" else "interrupt"
+        # busy_input_mode: "interrupt" (Enter interrupts current run),
+        # "queue" (Enter queues for next turn), or "steer" (Enter injects
+        # mid-run via /steer, arriving after the next tool call).
+        _bim = str(CLI_CONFIG["display"].get("busy_input_mode", "interrupt")).strip().lower()
+        if _bim == "queue":
+            self.busy_input_mode = "queue"
+        elif _bim == "steer":
+            self.busy_input_mode = "steer"
+        else:
+            self.busy_input_mode = "interrupt"
 
         self.verbose = verbose if verbose is not None else (self.tool_progress_mode == "verbose")
         
@@ -2039,6 +2077,11 @@ class HermesCLI:
         # Never blocks startup on failure.
         _run_state_db_auto_maintenance(self._session_db)
 
+        # Opportunistic shadow-repo cleanup — deletes orphan/stale
+        # checkpoint repos under ~/.hermes/checkpoints/.  Opt-in via
+        # checkpoints.auto_prune, idempotent via .last_prune marker.
+        _run_checkpoint_auto_maintenance()
+
         # Deferred title: stored in memory until the session is created in the DB
         self._pending_title: Optional[str] = None
         
@@ -3176,7 +3219,14 @@ class HermesCLI:
         # the configured model (e.g. "qwen3.6-plus"), causing 400 errors.
         runtime_model = runtime.get("model")
         if runtime_model and isinstance(runtime_model, str):
-            self.model = runtime_model
+            # Only use runtime model if: model is unset, or model equals provider name
+            should_use_runtime_model = (
+                not self.model or  # No model configured yet
+                self.model == self.provider or  # Model is the provider slug
+                self.model == runtime.get("name")  # Model matches provider display name
+            )
+            if should_use_runtime_model:
+                self.model = runtime_model
 
         # If model is still empty (e.g. user ran `hermes auth add openai-codex`
         # without `hermes model`), fall back to the provider's first catalog
@@ -4311,7 +4361,7 @@ class HermesCLI:
 
         _cprint(f"\n  {_DIM}Tip: Just type your message to chat with Hermes!{_RST}")
         _cprint(f"  {_DIM}Multi-line: Alt+Enter for a new line{_RST}")
-        _cprint(f"  {_DIM}Draft editor: Ctrl+G{_RST}")
+        _cprint(f"  {_DIM}Draft editor: Ctrl+G (Alt+G in VSCode/Cursor){_RST}")
         if _is_termux_environment():
             _cprint(f"  {_DIM}Attach image: /image {_termux_example_image_path()} or start your prompt with a local image path{_RST}\n")
         else:
@@ -4661,10 +4711,6 @@ class HermesCLI:
     def new_session(self, silent=False):
         """Start a fresh session with a new session ID and cleared agent state."""
         if self.agent and self.conversation_history:
-            try:
-                self.agent.flush_memories(self.conversation_history)
-            except (Exception, KeyboardInterrupt):
-                pass
             # Trigger memory extraction on the old session before session_id rotates.
             self.agent.commit_memory_session(self.conversation_history)
             self._notify_session_boundary("on_session_finalize")
@@ -4906,6 +4952,12 @@ class HermesCLI:
         if self.agent:
             self.agent.session_id = new_session_id
             self.agent.session_start = now
+            # Redirect the JSON session log to the new branch session file so
+            # messages written after branching land in the correct file.
+            if hasattr(self.agent, "session_log_file") and hasattr(self.agent, "logs_dir"):
+                self.agent.session_log_file = (
+                    self.agent.logs_dir / f"session_{new_session_id}.json"
+                )
             self.agent.reset_session_state()
             if hasattr(self.agent, "_last_flushed_db_idx"):
                 self.agent._last_flushed_db_idx = len(self.conversation_history)
@@ -4927,22 +4979,37 @@ class HermesCLI:
         _cprint(f"  Branch session:   {new_session_id}")
 
     def save_conversation(self):
-        """Save the current conversation to a file."""
+        """Save the current conversation to a JSON snapshot under ~/.hermes/sessions/saved/.
+
+        The snapshot is a convenience export for sharing or off-line inspection;
+        every message is already persisted incrementally to the SQLite session
+        DB, so the live session remains resumable via ``hermes --resume <id>``
+        regardless of whether the user ever runs ``/save``.
+        """
         if not self.conversation_history:
             print("(;_;) No conversation to save.")
             return
-        
+
         timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
-        filename = f"hermes_conversation_{timestamp}.json"
-        
+        saved_dir = get_hermes_home() / "sessions" / "saved"
         try:
-            with open(filename, "w", encoding="utf-8") as f:
+            saved_dir.mkdir(parents=True, exist_ok=True)
+        except Exception as e:
+            print(f"(x_x) Failed to create save directory {saved_dir}: {e}")
+            return
+        path = saved_dir / f"hermes_conversation_{timestamp}.json"
+
+        try:
+            with open(path, "w", encoding="utf-8") as f:
                 json.dump({
                     "model": self.model,
+                    "session_id": self.session_id,
                     "session_start": self.session_start.isoformat(),
                     "messages": self.conversation_history,
                 }, f, indent=2, ensure_ascii=False)
-            print(f"(^_^)v Conversation saved to: {filename}")
+            print(f"(^_^)v Conversation snapshot saved to: {path}")
+            if self.session_id:
+                print(f"       Resume the live session with: hermes --resume {self.session_id}")
         except Exception as e:
             print(f"(x_x) Failed to save: {e}")
     
@@ -5149,27 +5216,29 @@ class HermesCLI:
         _cprint(f"  ✓ Model switched: {result.new_model}")
         _cprint(f"    Provider: {provider_label}")
 
+        # Context: always resolve via the provider-aware chain so Codex OAuth,
+        # Copilot, and Nous-enforced caps win over the raw models.dev entry
+        # (e.g. gpt-5.5 is 1.05M on openai but 272K on Codex OAuth).
         mi = result.model_info
+        try:
+            from hermes_cli.model_switch import resolve_display_context_length
+            ctx = resolve_display_context_length(
+                result.new_model,
+                result.target_provider,
+                base_url=result.base_url or self.base_url or "",
+                api_key=result.api_key or self.api_key or "",
+                model_info=mi,
+            )
+            if ctx:
+                _cprint(f"    Context: {ctx:,} tokens")
+        except Exception:
+            pass
         if mi:
-            if mi.context_window:
-                _cprint(f"    Context: {mi.context_window:,} tokens")
             if mi.max_output:
                 _cprint(f"    Max output: {mi.max_output:,} tokens")
             if mi.has_cost_data():
                 _cprint(f"    Cost: {mi.format_cost()}")
             _cprint(f"    Capabilities: {mi.format_capabilities()}")
-        else:
-            try:
-                from agent.model_metadata import get_model_context_length
-                ctx = get_model_context_length(
-                    result.new_model,
-                    base_url=result.base_url or self.base_url,
-                    api_key=result.api_key or self.api_key,
-                    provider=result.target_provider,
-                )
-                _cprint(f"    Context: {ctx:,} tokens")
-            except Exception:
-                pass
 
         cache_enabled = (
             (base_url_host_matches(result.base_url or "", "openrouter.ai") and "claude" in result.new_model.lower())
@@ -5270,24 +5339,22 @@ class HermesCLI:
         # Parse --provider and --global flags
         model_input, explicit_provider, persist_global = parse_model_flags(raw_args)
 
+        # Load providers for switch_model (picker path needs them below)
         user_provs = None
         custom_provs = None
+        try:
+            from hermes_cli.config import get_compatible_custom_providers, load_config
+            cfg = load_config()
+            user_provs = cfg.get("providers")
+            custom_provs = get_compatible_custom_providers(cfg)
+        except Exception:
+            pass
 
         # No args at all: open prompt_toolkit-native picker modal
         if not model_input and not explicit_provider:
             model_display = self.model or "unknown"
             provider_display = get_label(self.provider) if self.provider else "unknown"
 
-            user_provs = None
-            custom_provs = None
-            try:
-                from hermes_cli.config import get_compatible_custom_providers, load_config
-                cfg = load_config()
-                user_provs = cfg.get("providers")
-                custom_provs = get_compatible_custom_providers(cfg)
-            except Exception:
-                pass
-
             try:
                 providers = list_authenticated_providers(
                     current_provider=self.provider or "",
@@ -6120,8 +6187,6 @@ class HermesCLI:
             self._handle_agents_command()
         elif canonical == "background":
             self._handle_background_command(cmd_original)
-        elif canonical == "btw":
-            self._handle_btw_command(cmd_original)
         elif canonical == "queue":
             # Extract prompt after "/queue " or "/q "
             parts = cmd_original.split(None, 1)
@@ -6300,6 +6365,12 @@ class HermesCLI:
         turn_route = self._resolve_turn_agent_config(prompt)
 
         def run_background():
+            set_sudo_password_callback(self._sudo_password_callback)
+            set_approval_callback(self._approval_callback)
+            try:
+                set_secret_capture_callback(self._secret_capture_callback)
+            except Exception:
+                pass
             try:
                 bg_agent = AIAgent(
                     model=turn_route["model"],
@@ -6397,6 +6468,12 @@ class HermesCLI:
                 print()
                 _cprint(f"  ❌ Background task #{task_num} failed: {e}")
             finally:
+                try:
+                    set_sudo_password_callback(None)
+                    set_approval_callback(None)
+                    set_secret_capture_callback(None)
+                except Exception:
+                    pass
                 self._background_tasks.pop(task_id, None)
                 # Clear spinner only if no foreground agent owns it
                 if not self._agent_running:
@@ -6408,122 +6485,6 @@ class HermesCLI:
         self._background_tasks[task_id] = thread
         thread.start()
 
-    def _handle_btw_command(self, cmd: str):
-        """Handle /btw <question> — ephemeral side question using session context.
-
-        Snapshots the current conversation history, spawns a no-tools agent in
-        a background thread, and prints the answer without persisting anything
-        to the main session.
-        """
-        parts = cmd.strip().split(maxsplit=1)
-        if len(parts) < 2 or not parts[1].strip():
-            _cprint("  Usage: /btw <question>")
-            _cprint("  Example: /btw what module owns session title sanitization?")
-            _cprint("  Answers using session context. No tools, not persisted.")
-            return
-
-        question = parts[1].strip()
-        task_id = f"btw_{datetime.now().strftime('%H%M%S')}_{uuid.uuid4().hex[:6]}"
-
-        if not self._ensure_runtime_credentials():
-            _cprint("  (>_<) Cannot start /btw: no valid credentials.")
-            return
-
-        turn_route = self._resolve_turn_agent_config(question)
-        history_snapshot = list(self.conversation_history)
-
-        preview = question[:60] + ("..." if len(question) > 60 else "")
-        _cprint(f'  💬 /btw: "{preview}"')
-
-        def run_btw():
-            try:
-                btw_agent = AIAgent(
-                    model=turn_route["model"],
-                    api_key=turn_route["runtime"].get("api_key"),
-                    base_url=turn_route["runtime"].get("base_url"),
-                    provider=turn_route["runtime"].get("provider"),
-                    api_mode=turn_route["runtime"].get("api_mode"),
-                    acp_command=turn_route["runtime"].get("command"),
-                    acp_args=turn_route["runtime"].get("args"),
-                    max_iterations=8,
-                    enabled_toolsets=[],
-                    quiet_mode=True,
-                    verbose_logging=False,
-                    session_id=task_id,
-                    platform="cli",
-                    reasoning_config=self.reasoning_config,
-                    service_tier=self.service_tier,
-                    request_overrides=turn_route.get("request_overrides"),
-                    providers_allowed=self._providers_only,
-                    providers_ignored=self._providers_ignore,
-                    providers_order=self._providers_order,
-                    provider_sort=self._provider_sort,
-                    provider_require_parameters=self._provider_require_params,
-                    provider_data_collection=self._provider_data_collection,
-                    fallback_model=self._fallback_model,
-                    session_db=None,
-                    skip_memory=True,
-                    skip_context_files=True,
-                    persist_session=False,
-                )
-
-                btw_prompt = (
-                    "[Ephemeral /btw side question. Answer using the conversation "
-                    "context. No tools available. Be direct and concise.]\n\n"
-                    + question
-                )
-                result = btw_agent.run_conversation(
-                    user_message=btw_prompt,
-                    conversation_history=history_snapshot,
-                    task_id=task_id,
-                )
-
-                response = (result.get("final_response") or "") if result else ""
-                if not response and result and result.get("error"):
-                    response = f"Error: {result['error']}"
-
-                # TUI refresh before printing
-                if self._app:
-                    self._app.invalidate()
-                    time.sleep(0.05)
-                print()
-
-                if response:
-                    try:
-                        from hermes_cli.skin_engine import get_active_skin
-                        _skin = get_active_skin()
-                        _resp_color = _skin.get_color("response_border", "#4F6D4A")
-                    except Exception:
-                        _resp_color = "#4F6D4A"
-
-                    ChatConsole().print(Panel(
-                        _render_final_assistant_content(response, mode=self.final_response_markdown),
-                        title=f"[{_resp_color} bold]⚕ /btw[/]",
-                        title_align="left",
-                        border_style=_resp_color,
-                        box=rich_box.HORIZONTALS,
-                        padding=(1, 4),
-                    ))
-                else:
-                    _cprint("  💬 /btw: (no response)")
-
-                if self.bell_on_complete:
-                    sys.stdout.write("\a")
-                    sys.stdout.flush()
-
-            except Exception as e:
-                if self._app:
-                    self._app.invalidate()
-                    time.sleep(0.05)
-                print()
-                _cprint(f"  ❌ /btw failed: {e}")
-            finally:
-                if self._app:
-                    self._invalidate(min_interval=0)
-
-        thread = threading.Thread(target=run_btw, daemon=True, name=f"btw-{task_id}")
-        thread.start()
-
     @staticmethod
     def _try_launch_chrome_debug(port: int, system: str) -> bool:
         """Try to launch Chrome/Chromium with remote debugging enabled.
@@ -6907,24 +6868,36 @@ class HermesCLI:
             /busy               Show current busy input mode
             /busy status        Show current busy input mode
             /busy queue         Queue input for the next turn instead of interrupting
+            /busy steer         Inject Enter mid-run via /steer (after next tool call)
             /busy interrupt     Interrupt the current run on Enter (default)
         """
         parts = cmd.strip().split(maxsplit=1)
         if len(parts) < 2 or parts[1].strip().lower() == "status":
             _cprint(f"  {_ACCENT}Busy input mode: {self.busy_input_mode}{_RST}")
-            _cprint(f"  {_DIM}Enter while busy: {'queues for next turn' if self.busy_input_mode == 'queue' else 'interrupts current run'}{_RST}")
-            _cprint(f"  {_DIM}Usage: /busy [queue|interrupt|status]{_RST}")
+            if self.busy_input_mode == "queue":
+                _behavior = "queues for next turn"
+            elif self.busy_input_mode == "steer":
+                _behavior = "steers into current run (after next tool call)"
+            else:
+                _behavior = "interrupts current run"
+            _cprint(f"  {_DIM}Enter while busy: {_behavior}{_RST}")
+            _cprint(f"  {_DIM}Usage: /busy [queue|steer|interrupt|status]{_RST}")
             return
 
         arg = parts[1].strip().lower()
-        if arg not in {"queue", "interrupt"}:
+        if arg not in {"queue", "interrupt", "steer"}:
             _cprint(f"  {_DIM}(._.) Unknown argument: {arg}{_RST}")
-            _cprint(f"  {_DIM}Usage: /busy [queue|interrupt|status]{_RST}")
+            _cprint(f"  {_DIM}Usage: /busy [queue|steer|interrupt|status]{_RST}")
             return
 
         self.busy_input_mode = arg
         if save_config_value("display.busy_input_mode", arg):
-            behavior = "Enter will queue follow-up input while Hermes is busy." if arg == "queue" else "Enter will interrupt the current run while Hermes is busy."
+            if arg == "queue":
+                behavior = "Enter will queue follow-up input while Hermes is busy."
+            elif arg == "steer":
+                behavior = "Enter will steer your message into the current run (after the next tool call)."
+            else:
+                behavior = "Enter will interrupt the current run while Hermes is busy."
             _cprint(f"  {_ACCENT}✓ Busy input mode set to '{arg}' (saved to config){_RST}")
             _cprint(f"  {_DIM}{behavior}{_RST}")
         else:
@@ -7326,7 +7299,7 @@ class HermesCLI:
             change_detail = ". ".join(change_parts) + ". " if change_parts else ""
             self.conversation_history.append({
                 "role": "user",
-                "content": f"[SYSTEM: MCP servers have been reloaded. {change_detail}{tool_summary}. The tool list for this conversation has been updated accordingly.]",
+                "content": f"[IMPORTANT: MCP servers have been reloaded. {change_detail}{tool_summary}. The tool list for this conversation has been updated accordingly.]",
             })
 
             # Persist session immediately so the session log reflects the
@@ -7408,6 +7381,31 @@ class HermesCLI:
                     _cprint(f"  {line}")
                 except Exception:
                     pass
+                # First-touch onboarding: on the first tool in this process
+                # that takes longer than the threshold while we're in the
+                # noisiest progress mode, print a one-time hint about
+                # /verbose.  Latched on self so it fires at most once per
+                # process; persisted to config.yaml so it never fires again
+                # across processes either.
+                try:
+                    if (
+                        not getattr(self, "_long_tool_hint_fired", False)
+                        and self.tool_progress_mode == "all"
+                        and duration >= 30.0
+                    ):
+                        from agent.onboarding import (
+                            TOOL_PROGRESS_FLAG,
+                            is_seen,
+                            mark_seen,
+                            tool_progress_hint_cli,
+                        )
+                        if not is_seen(CLI_CONFIG, TOOL_PROGRESS_FLAG):
+                            self._long_tool_hint_fired = True
+                            _cprint(f"  {_DIM}{tool_progress_hint_cli()}{_RST}")
+                            mark_seen(_hermes_home / "config.yaml", TOOL_PROGRESS_FLAG)
+                            CLI_CONFIG.setdefault("onboarding", {}).setdefault("seen", {})[TOOL_PROGRESS_FLAG] = True
+                except Exception:
+                    pass
             self._invalidate()
             return
         if event_type != "tool.started":
@@ -9075,6 +9073,30 @@ class HermesCLI:
             _welcome_text = "Welcome to Hermes Agent! Type your message or /help for commands."
             _welcome_color = "#FFF8DC"
         self._console_print(f"[{_welcome_color}]{_welcome_text}[/]")
+        # First-time OpenClaw-residue banner — fires once if ~/.openclaw/ exists
+        # after an OpenClaw→Hermes migration (especially migrations done by
+        # OpenClaw's own tool, which doesn't archive the source directory).
+        try:
+            from agent.onboarding import (
+                OPENCLAW_RESIDUE_FLAG,
+                detect_openclaw_residue,
+                is_seen,
+                mark_seen,
+                openclaw_residue_hint_cli,
+            )
+            if not is_seen(self.config, OPENCLAW_RESIDUE_FLAG) and detect_openclaw_residue():
+                try:
+                    _resid_color = _welcome_skin.get_color("banner_dim", "#B8860B")
+                except Exception:
+                    _resid_color = "#B8860B"
+                self._console_print(f"[{_resid_color}]{openclaw_residue_hint_cli()}[/]")
+                try:
+                    from hermes_cli.config import get_config_path as _get_cfg_path_resid
+                    mark_seen(_get_cfg_path_resid(), OPENCLAW_RESIDUE_FLAG)
+                except Exception:
+                    pass  # best-effort — banner will fire again next session
+        except Exception:
+            pass  # banner is non-critical — never break startup
         # Show a random tip to help users discover features
         try:
             from hermes_cli.tips import get_random_tip
@@ -9276,12 +9298,34 @@ class HermesCLI:
                 # Bundle text + images as a tuple when images are present
                 payload = (text, images) if images else text
                 if self._agent_running and not (text and _looks_like_slash_command(text)):
-                    if self.busy_input_mode == "queue":
+                    _effective_mode = self.busy_input_mode
+                    if _effective_mode == "steer":
+                        # Route Enter through /steer — inject mid-run after the
+                        # next tool call.  Images can't ride along (steer only
+                        # appends text), so fall back to queue when images are
+                        # attached.  If the agent lacks steer() or rejects the
+                        # payload, also fall back to queue so nothing is lost.
+                        if images or not text:
+                            _effective_mode = "queue"
+                        else:
+                            accepted = False
+                            try:
+                                if self.agent is not None and hasattr(self.agent, "steer"):
+                                    accepted = bool(self.agent.steer(text))
+                            except Exception as exc:
+                                _cprint(f"  {_DIM}Steer failed ({exc}) — queued for next turn.{_RST}")
+                                accepted = False
+                            if accepted:
+                                preview = text[:80] + ("..." if len(text) > 80 else "")
+                                _cprint(f"  {_ACCENT}⏩ Steered: '{preview}'{_RST}")
+                            else:
+                                _effective_mode = "queue"
+                    if _effective_mode == "queue":
                         # Queue for the next turn instead of interrupting
                         self._pending_input.put(payload)
                         preview = text if text else f"[{len(images)} image{'s' if len(images) != 1 else ''} attached]"
                         _cprint(f"  Queued for the next turn: {preview[:80]}{'...' if len(preview) > 80 else ''}")
-                    else:
+                    elif _effective_mode == "interrupt":
                         self._interrupt_queue.put(payload)
                         # Debug: log to file when message enters interrupt queue
                         try:
@@ -9291,6 +9335,24 @@ class HermesCLI:
                                          f"agent_running={self._agent_running}\n")
                         except Exception:
                             pass
+                    # First-touch onboarding: on the very first busy-while-running
+                    # event for this install, print a one-line tip explaining the
+                    # /busy knob.  Flag persists to config.yaml and never fires
+                    # again.  Guarded for exceptions so onboarding can't break
+                    # the input loop.
+                    try:
+                        from agent.onboarding import (
+                            BUSY_INPUT_FLAG,
+                            busy_input_hint_cli,
+                            is_seen,
+                            mark_seen,
+                        )
+                        if not is_seen(CLI_CONFIG, BUSY_INPUT_FLAG):
+                            _cprint(f"  {_DIM}{busy_input_hint_cli(self.busy_input_mode)}{_RST}")
+                            mark_seen(_hermes_home / "config.yaml", BUSY_INPUT_FLAG)
+                            CLI_CONFIG.setdefault("onboarding", {}).setdefault("seen", {})[BUSY_INPUT_FLAG] = True
+                    except Exception:
+                        pass
                 else:
                     self._pending_input.put(payload)
                 event.app.current_buffer.reset(append_to_history=True)
@@ -9305,14 +9367,18 @@ class HermesCLI:
             """Ctrl+Enter (c-j) inserts a newline. Most terminals send c-j for Ctrl+Enter."""
             event.current_buffer.insert_text('\n')
 
-        @kb.add(
-            'c-g',
-            filter=Condition(
-                lambda: not self._clarify_state and not self._approval_state and not self._sudo_state and not self._secret_state
-            ),
+        # VSCode/Cursor bind Ctrl+G to "Find Next" at the editor level, so
+        # the keystroke never reaches the embedded terminal. Alt+G is unbound
+        # in those IDEs and arrives here as ('escape', 'g') — register it as
+        # a fallback so the editor handoff works inside Cursor/VSCode too.
+        _editor_filter = Condition(
+            lambda: not self._clarify_state and not self._approval_state and not self._sudo_state and not self._secret_state
         )
+
+        @kb.add('c-g', filter=_editor_filter)
+        @kb.add('escape', 'g', filter=_editor_filter)
         def handle_open_in_editor(event):
-            """Ctrl+G opens the current draft in an external editor."""
+            """Ctrl+G (or Alt+G in VSCode/Cursor) opens the current draft in an external editor."""
             cli_ref._open_external_editor(event.current_buffer)
 
         @kb.add('tab', eager=True)
@@ -9776,6 +9842,11 @@ class HermesCLI:
                 completer=_completer,
             ),
         )
+        # Keep prompt_toolkit on its simple tempfile path. Setting
+        # buffer.tempfile = "prompt.md" triggers its complex-tempfile branch,
+        # which tries to mkdir() the mkdtemp() directory again and raises
+        # EEXIST. The suffix keeps markdown highlighting without that bug.
+        input_area.buffer.tempfile_suffix = '.md'
 
         # Dynamic height: accounts for both explicit newlines AND visual
         # wrapping of long lines so the input area always fits its content.
@@ -9898,7 +9969,7 @@ class HermesCLI:
                 status = cli_ref._command_status or "Processing command..."
                 return f"{frame} {status}"
             if cli_ref._agent_running:
-                return "type a message + Enter to interrupt, Ctrl+C to cancel"
+                return "msg=interrupt · /queue · /bg · /steer · Ctrl+C cancel"
             if cli_ref._voice_mode:
                 return "type or Ctrl+B to record"
             return ""
@@ -10728,6 +10799,8 @@ class HermesCLI:
                 return  # silently suppress
             if isinstance(exc, KeyError) and "is not registered" in str(exc):
                 return  # suppress selector registration failures (#6393)
+            if isinstance(exc, OSError) and getattr(exc, "errno", None) == errno.EIO:
+                return  # suppress I/O errors from broken stdout on interrupt (#13710)
             # Fall back to default handler for everything else
             loop.default_exception_handler(context)
 
@@ -10760,9 +10833,11 @@ class HermesCLI:
         except (EOFError, KeyboardInterrupt, BrokenPipeError):
             pass
         except (KeyError, OSError) as _stdin_err:
-            # Catch selector registration failures from broken stdin (#6393).
-            # This is the fallback for cases that slip past the fstat() guard.
-            if "is not registered" in str(_stdin_err) or "Bad file descriptor" in str(_stdin_err):
+            # Catch selector registration failures from broken stdin (#6393)
+            # and I/O errors from broken stdout during interrupt (#13710).
+            if isinstance(_stdin_err, OSError) and getattr(_stdin_err, "errno", None) == errno.EIO:
+                pass  # suppress broken-stdout I/O errors on interrupt (#13710)
+            elif "is not registered" in str(_stdin_err) or "Bad file descriptor" in str(_stdin_err):
                 print(
                     f"\nError: stdin is not usable ({_stdin_err}).\n"
                     "This can happen with certain Python installations (e.g. uv-managed cPython on macOS).\n"
@@ -10781,12 +10856,6 @@ class HermesCLI:
                     self.agent.interrupt()
                 except Exception:
                     pass
-            # Flush memories before exit (only for substantial conversations)
-            if self.agent and self.conversation_history:
-                try:
-                    self.agent.flush_memories(self.conversation_history)
-                except (Exception, KeyboardInterrupt):
-                    pass
             # Shut down voice recorder (release persistent audio stream)
             if hasattr(self, '_voice_recorder') and self._voice_recorder:
                 try:

cron/jobs.py

@@ -16,7 +16,7 @@ import uuid
 from datetime import datetime, timedelta
 from pathlib import Path
 from hermes_constants import get_hermes_home
-from typing import Optional, Dict, List, Any
+from typing import Optional, Dict, List, Any, Union
 
 logger = logging.getLogger(__name__)
 
@@ -417,6 +417,7 @@ def create_job(
     provider: Optional[str] = None,
     base_url: Optional[str] = None,
     script: Optional[str] = None,
+    context_from: Optional[Union[str, List[str]]] = None,
     enabled_toolsets: Optional[List[str]] = None,
     workdir: Optional[str] = None,
 ) -> Dict[str, Any]:
@@ -438,6 +439,9 @@ def create_job(
         script: Optional path to a Python script whose stdout is injected into the
                 prompt each run.  The script runs before the agent turn, and its output
                 is prepended as context.  Useful for data collection / change detection.
+        context_from: Optional job ID (or list of job IDs) whose most recent output
+                      is injected into the prompt as context before each run.
+                      Useful for chaining cron jobs: job A finds data, job B processes it.
         enabled_toolsets: Optional list of toolset names to restrict the agent to.
                           When set, only tools from these toolsets are loaded, reducing
                           token overhead. When omitted, all default tools are loaded.
@@ -481,6 +485,14 @@ def create_job(
     normalized_toolsets = normalized_toolsets or None
     normalized_workdir = _normalize_workdir(workdir)
 
+    # Normalize context_from: accept str or list of str, store as list or None
+    if isinstance(context_from, str):
+        context_from = [context_from.strip()] if context_from.strip() else None
+    elif isinstance(context_from, list):
+        context_from = [str(j).strip() for j in context_from if str(j).strip()] or None
+    else:
+        context_from = None
+
     label_source = (prompt or (normalized_skills[0] if normalized_skills else None)) or "cron job"
     job = {
         "id": job_id,
@@ -492,6 +504,7 @@ def create_job(
         "provider": normalized_provider,
         "base_url": normalized_base_url,
         "script": normalized_script,
+        "context_from": context_from,
         "schedule": parsed_schedule,
         "schedule_display": parsed_schedule.get("display", schedule),
         "repeat": {

cron/scheduler.py

@@ -77,7 +77,7 @@ _KNOWN_DELIVERY_PLATFORMS = frozenset({
     "telegram", "discord", "slack", "whatsapp", "signal",
     "matrix", "mattermost", "homeassistant", "dingtalk", "feishu",
     "wecom", "wecom_callback", "weixin", "sms", "email", "webhook", "bluebubbles",
-    "qqbot",
+    "qqbot", "yuanbao",
 })
 
 # Platforms that support a configured cron/notification home target, mapped to
@@ -337,6 +337,7 @@ def _deliver_result(job: dict, content: str, adapters=None, loop=None) -> Option
         "sms": Platform.SMS,
         "bluebubbles": Platform.BLUEBUBBLES,
         "qqbot": Platform.QQBOT,
+        "yuanbao": Platform.YUANBAO,
     }
 
     # Optionally wrap the content with a header/footer so the user knows this
@@ -671,10 +672,51 @@ def _build_job_prompt(job: dict, prerun_script: Optional[tuple] = None) -> str:
                 f"{prompt}"
             )
 
+    # Inject output from referenced cron jobs as context.
+    context_from = job.get("context_from")
+    if context_from:
+        from cron.jobs import OUTPUT_DIR
+        if isinstance(context_from, str):
+            context_from = [context_from]
+        for source_job_id in context_from:
+            # Guard against path traversal — valid job IDs are 12-char hex strings
+            if not source_job_id or not all(c in "0123456789abcdef" for c in source_job_id):
+                logger.warning("context_from: skipping invalid job_id %r", source_job_id)
+                continue
+            try:
+                job_output_dir = OUTPUT_DIR / source_job_id
+                if not job_output_dir.exists():
+                    continue  # silent skip — no output yet
+                output_files = sorted(
+                    job_output_dir.glob("*.md"),
+                    key=lambda f: f.stat().st_mtime,
+                    reverse=True,
+                )
+                if not output_files:
+                    continue  # silent skip — no output yet
+                latest_output = output_files[0].read_text(encoding="utf-8").strip()
+                # Truncate to 8K characters to avoid prompt bloat
+                _MAX_CONTEXT_CHARS = 8000
+                if len(latest_output) > _MAX_CONTEXT_CHARS:
+                    latest_output = latest_output[:_MAX_CONTEXT_CHARS] + "\n\n[... output truncated ...]"
+                if latest_output:
+                    prompt = (
+                        f"## Output from job '{source_job_id}'\n"
+                        "The following is the most recent output from a preceding "
+                        "cron job. Use it as context for your analysis.\n\n"
+                        f"```\n{latest_output}\n```\n\n"
+                        f"{prompt}"
+                    )
+                else:
+                    continue  # silent skip — empty output
+            except (OSError, PermissionError) as e:
+                logger.warning("context_from: failed to read output for job %r: %s", source_job_id, e)
+                # silent skip — do not pollute the prompt with error messages
+
     # Always prepend cron execution guidance so the agent knows how
     # delivery works and can suppress delivery when appropriate.
     cron_hint = (
-        "[SYSTEM: You are running as a scheduled cron job. "
+        "[IMPORTANT: You are running as a scheduled cron job. "
         "DELIVERY: Your final response will be automatically delivered "
         "to the user — do NOT use send_message or try to deliver "
         "the output yourself. Just produce your report/output as your "
@@ -710,7 +752,7 @@ def _build_job_prompt(job: dict, prerun_script: Optional[tuple] = None) -> str:
             parts.append("")
         parts.extend(
             [
-                f'[SYSTEM: The user has invoked the "{skill_name}" skill, indicating they want you to follow its instructions. The full skill content is loaded below.]',
+                f'[IMPORTANT: The user has invoked the "{skill_name}" skill, indicating they want you to follow its instructions. The full skill content is loaded below.]',
                 "",
                 content,
             ]
@@ -718,7 +760,7 @@ def _build_job_prompt(job: dict, prerun_script: Optional[tuple] = None) -> str:
 
     if skipped:
         notice = (
-            f"[SYSTEM: The following skill(s) were listed for this job but could not be found "
+            f"[IMPORTANT: The following skill(s) were listed for this job but could not be found "
             f"and were skipped: {', '.join(skipped)}. "
             f"Start your response with a brief notice so the user is aware, e.g.: "
             f"'⚠️ Skill(s) not found and skipped: {', '.join(skipped)}']"
@@ -1267,6 +1309,17 @@ def tick(verbose: bool = True, adapters=None, loop=None) -> int:
                     _futures.append(_tick_pool.submit(_ctx.run, _process_job, job))
                 _results.extend(f.result() for f in _futures)
 
+        # Best-effort sweep of MCP stdio subprocesses that survived their
+        # session teardown during this tick.  Runs AFTER every job has
+        # finished so active sessions (including live user chats) are
+        # never touched — only PIDs explicitly detected as orphans in
+        # tools.mcp_tool._run_stdio's finally block are reaped.
+        try:
+            from tools.mcp_tool import _kill_orphaned_mcp_children
+            _kill_orphaned_mcp_children()
+        except Exception as _e:
+            logger.debug("Post-tick MCP orphan cleanup failed: %s", _e)
+
         return sum(_results)
     finally:
         if fcntl:

docker/entrypoint.sh

@@ -41,6 +41,15 @@ if [ "$(id -u)" = "0" ]; then
             echo "Warning: chown failed (rootless container?) — continuing anyway"
     fi
 
+    # Ensure config.yaml is readable by the hermes runtime user even if it was
+    # edited on the host after initial ownership setup. Must run here (as root)
+    # rather than after the gosu drop, otherwise a non-root caller like
+    # `docker run -u $(id -u):$(id -g)` hits "Operation not permitted" (#15865).
+    if [ -f "$HERMES_HOME/config.yaml" ]; then
+        chown hermes:hermes "$HERMES_HOME/config.yaml" 2>/dev/null || true
+        chmod 640 "$HERMES_HOME/config.yaml" 2>/dev/null || true
+    fi
+
     echo "Dropping root privileges"
     exec gosu hermes "$0" "$@"
 fi
@@ -67,13 +76,6 @@ if [ ! -f "$HERMES_HOME/config.yaml" ]; then
     cp "$INSTALL_DIR/cli-config.yaml.example" "$HERMES_HOME/config.yaml"
 fi
 
-# Ensure the main config file remains accessible to the hermes runtime user
-# even if it was edited on the host after initial ownership setup.
-if [ -f "$HERMES_HOME/config.yaml" ]; then
-    chown hermes:hermes "$HERMES_HOME/config.yaml"
-    chmod 640 "$HERMES_HOME/config.yaml"
-fi
-
 # SOUL.md
 if [ ! -f "$HERMES_HOME/SOUL.md" ]; then
     cp "$INSTALL_DIR/docker/SOUL.md" "$HERMES_HOME/SOUL.md"

gateway/channel_directory.py

@@ -57,7 +57,7 @@ def _session_entry_name(origin: Dict[str, Any]) -> str:
 # Build / refresh
 # ---------------------------------------------------------------------------
 
-def build_channel_directory(adapters: Dict[Any, Any]) -> Dict[str, Any]:
+async def build_channel_directory(adapters: Dict[Any, Any]) -> Dict[str, Any]:
     """
     Build a channel directory from connected platform adapters and session data.
 
@@ -72,7 +72,7 @@ def build_channel_directory(adapters: Dict[Any, Any]) -> Dict[str, Any]:
             if platform == Platform.DISCORD:
                 platforms["discord"] = _build_discord(adapter)
             elif platform == Platform.SLACK:
-                platforms["slack"] = _build_slack(adapter)
+                platforms["slack"] = await _build_slack(adapter)
         except Exception as e:
             logger.warning("Channel directory: failed to build %s: %s", platform.value, e)
 
@@ -136,21 +136,66 @@ def _build_discord(adapter) -> List[Dict[str, str]]:
     return channels
 
 
-def _build_slack(adapter) -> List[Dict[str, str]]:
-    """List Slack channels the bot has joined."""
-    # Slack adapter may expose a web client
-    client = getattr(adapter, "_app", None) or getattr(adapter, "_client", None)
-    if not client:
+async def _build_slack(adapter) -> List[Dict[str, Any]]:
+    """List Slack channels the bot has joined across all workspaces.
+
+    Uses ``users.conversations`` against each workspace's web client. Pulls
+    public + private channels the bot is a member of, then merges in DMs
+    discovered from session history (IMs aren't useful to enumerate
+    proactively).
+    """
+    team_clients = getattr(adapter, "_team_clients", None) or {}
+    if not team_clients:
         return _build_from_sessions("slack")
 
-    try:
-        from tools.send_message_tool import _send_slack  # noqa: F401
-        # Use the Slack Web API directly if available
-    except Exception:
-        pass
+    channels: List[Dict[str, Any]] = []
+    seen_ids: set = set()
 
-    # Fallback to session data
-    return _build_from_sessions("slack")
+    for team_id, client in team_clients.items():
+        try:
+            cursor: Optional[str] = None
+            for _page in range(20):  # safety cap on pagination
+                response = await client.users_conversations(
+                    types="public_channel,private_channel",
+                    exclude_archived=True,
+                    limit=200,
+                    cursor=cursor,
+                )
+                if not response.get("ok"):
+                    logger.warning(
+                        "Channel directory: users.conversations not ok for team %s: %s",
+                        team_id,
+                        response.get("error", "unknown"),
+                    )
+                    break
+                for ch in response.get("channels", []):
+                    cid = ch.get("id")
+                    name = ch.get("name")
+                    if not cid or not name or cid in seen_ids:
+                        continue
+                    seen_ids.add(cid)
+                    channels.append({
+                        "id": cid,
+                        "name": name,
+                        "type": "private" if ch.get("is_private") else "channel",
+                    })
+                cursor = (response.get("response_metadata") or {}).get("next_cursor")
+                if not cursor:
+                    break
+        except Exception as e:
+            logger.warning(
+                "Channel directory: failed to list Slack channels for team %s: %s",
+                team_id, e,
+            )
+            continue
+
+    # Merge in DM/group entries discovered from session history.
+    for entry in _build_from_sessions("slack"):
+        if entry.get("id") not in seen_ids:
+            channels.append(entry)
+            seen_ids.add(entry.get("id"))
+
+    return channels
 
 
 def _build_from_sessions(platform_name: str) -> List[Dict[str, str]]:
@@ -223,6 +268,14 @@ def resolve_channel_name(platform_name: str, name: str) -> Optional[str]:
     if not channels:
         return None
 
+    # 0. Exact ID match — case-sensitive, no normalization. Lets callers pass
+    # raw platform IDs (e.g. Slack "C0B0QV5434G") even when the format guard
+    # in _parse_target_ref hasn't recognized them as explicit.
+    raw = name.strip()
+    for ch in channels:
+        if ch.get("id") == raw:
+            return ch["id"]
+
     query = _normalize_channel_query(name)
 
     # 1. Exact name match, including the display labels shown by send_message(action="list")

gateway/config.py

@@ -67,6 +67,7 @@ class Platform(Enum):
     WEIXIN = "weixin"
     BLUEBUBBLES = "bluebubbles"
     QQBOT = "qqbot"
+    YUANBAO = "yuanbao"
 
 
 @dataclass
@@ -195,6 +196,14 @@ class StreamingConfig:
     edit_interval: float = 1.0    # Seconds between message edits (Telegram rate-limits at ~1/s)
     buffer_threshold: int = 40    # Chars before forcing an edit
     cursor: str = " ▉"           # Cursor shown during streaming
+    # Ported from openclaw/openclaw#72038.  When >0, the final edit for
+    # a long-running streamed response is delivered as a fresh message
+    # if the original preview has been visible for at least this many
+    # seconds, so the platform's visible timestamp reflects completion
+    # time instead of the preview creation time.  Currently applied to
+    # Telegram only (other platforms ignore the setting).  Default 60s
+    # matches the OpenClaw rollout.  Set to 0 to disable.
+    fresh_final_after_seconds: float = 60.0
 
     def to_dict(self) -> Dict[str, Any]:
         return {
@@ -203,6 +212,7 @@ class StreamingConfig:
             "edit_interval": self.edit_interval,
             "buffer_threshold": self.buffer_threshold,
             "cursor": self.cursor,
+            "fresh_final_after_seconds": self.fresh_final_after_seconds,
         }
 
     @classmethod
@@ -215,6 +225,9 @@ class StreamingConfig:
             edit_interval=float(data.get("edit_interval", 1.0)),
             buffer_threshold=int(data.get("buffer_threshold", 40)),
             cursor=data.get("cursor", " ▉"),
+            fresh_final_after_seconds=float(
+                data.get("fresh_final_after_seconds", 60.0)
+            ),
         )
 
 
@@ -314,6 +327,9 @@ class GatewayConfig:
             # QQBot uses extra dict for app credentials
             elif platform == Platform.QQBOT and config.extra.get("app_id") and config.extra.get("client_secret"):
                 connected.append(platform)
+            # Yuanbao uses extra dict for app credentials
+            elif platform == Platform.YUANBAO and config.extra.get("app_id") and config.extra.get("app_secret"):
+                connected.append(platform)
             # DingTalk uses client_id/client_secret from config.extra or env vars
             elif platform == Platform.DINGTALK and (
                 config.extra.get("client_id") or os.getenv("DINGTALK_CLIENT_ID")
@@ -570,6 +586,8 @@ def load_gateway_config() -> GatewayConfig:
                     )
                 if "reply_prefix" in platform_cfg:
                     bridged["reply_prefix"] = platform_cfg["reply_prefix"]
+                if "reply_in_thread" in platform_cfg:
+                    bridged["reply_in_thread"] = platform_cfg["reply_in_thread"]
                 if "require_mention" in platform_cfg:
                     bridged["require_mention"] = platform_cfg["require_mention"]
                 if "free_response_channels" in platform_cfg:
@@ -584,7 +602,7 @@ def load_gateway_config() -> GatewayConfig:
                     bridged["group_policy"] = platform_cfg["group_policy"]
                 if "group_allow_from" in platform_cfg:
                     bridged["group_allow_from"] = platform_cfg["group_allow_from"]
-                if plat == Platform.DISCORD and "channel_skill_bindings" in platform_cfg:
+                if plat in (Platform.DISCORD, Platform.SLACK) and "channel_skill_bindings" in platform_cfg:
                     bridged["channel_skill_bindings"] = platform_cfg["channel_skill_bindings"]
                 if "channel_prompts" in platform_cfg:
                     channel_prompts = platform_cfg["channel_prompts"]
@@ -609,6 +627,8 @@ def load_gateway_config() -> GatewayConfig:
             if isinstance(slack_cfg, dict):
                 if "require_mention" in slack_cfg and not os.getenv("SLACK_REQUIRE_MENTION"):
                     os.environ["SLACK_REQUIRE_MENTION"] = str(slack_cfg["require_mention"]).lower()
+                if "strict_mention" in slack_cfg and not os.getenv("SLACK_STRICT_MENTION"):
+                    os.environ["SLACK_STRICT_MENTION"] = str(slack_cfg["strict_mention"]).lower()
                 if "allow_bots" in slack_cfg and not os.getenv("SLACK_ALLOW_BOTS"):
                     os.environ["SLACK_ALLOW_BOTS"] = str(slack_cfg["allow_bots"]).lower()
                 frc = slack_cfg.get("free_response_channels")
@@ -918,8 +938,12 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
     slack_token = os.getenv("SLACK_BOT_TOKEN")
     if slack_token:
         if Platform.SLACK not in config.platforms:
+            # No yaml config for Slack — env-only setup, enable it
             config.platforms[Platform.SLACK] = PlatformConfig()
-        config.platforms[Platform.SLACK].enabled = True
+            config.platforms[Platform.SLACK].enabled = True
+        # If yaml config exists, respect its enabled flag (don't override
+        # explicit enabled: false). Token is still stored so skills that
+        # send Slack messages can use it without activating the gateway adapter.
         config.platforms[Platform.SLACK].token = slack_token
     slack_home = os.getenv("SLACK_HOME_CHANNEL")
     if slack_home and Platform.SLACK in config.platforms:
@@ -1276,6 +1300,48 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
                 name=os.getenv("QQBOT_HOME_CHANNEL_NAME") or os.getenv(qq_home_name_env, "Home"),
             )
 
+    # Yuanbao — YUANBAO_APP_ID preferred
+    yuanbao_app_id = os.getenv("YUANBAO_APP_ID") or os.getenv("YUANBAO_APP_KEY")
+    yuanbao_app_secret = os.getenv("YUANBAO_APP_SECRET")
+    if yuanbao_app_id and yuanbao_app_secret:
+        if Platform.YUANBAO not in config.platforms:
+            config.platforms[Platform.YUANBAO] = PlatformConfig()
+        config.platforms[Platform.YUANBAO].enabled = True
+        extra = config.platforms[Platform.YUANBAO].extra
+        extra["app_id"] = yuanbao_app_id
+        extra["app_secret"] = yuanbao_app_secret
+        yuanbao_bot_id = os.getenv("YUANBAO_BOT_ID")
+        if yuanbao_bot_id:
+            extra["bot_id"] = yuanbao_bot_id
+        yuanbao_ws_url = os.getenv("YUANBAO_WS_URL")
+        if yuanbao_ws_url:
+            extra["ws_url"] = yuanbao_ws_url
+        yuanbao_api_domain = os.getenv("YUANBAO_API_DOMAIN")
+        if yuanbao_api_domain:
+            extra["api_domain"] = yuanbao_api_domain
+        yuanbao_route_env = os.getenv("YUANBAO_ROUTE_ENV")
+        if yuanbao_route_env:
+            extra["route_env"] = yuanbao_route_env
+        yuanbao_home = os.getenv("YUANBAO_HOME_CHANNEL")
+        if yuanbao_home:
+            config.platforms[Platform.YUANBAO].home_channel = HomeChannel(
+                platform=Platform.YUANBAO,
+                chat_id=yuanbao_home,
+                name=os.getenv("YUANBAO_HOME_CHANNEL_NAME", "Home"),
+            )
+        yuanbao_dm_policy = os.getenv("YUANBAO_DM_POLICY")
+        if yuanbao_dm_policy:
+            extra["dm_policy"] = yuanbao_dm_policy.strip().lower()
+        yuanbao_dm_allow_from = os.getenv("YUANBAO_DM_ALLOW_FROM")
+        if yuanbao_dm_allow_from:
+            extra["dm_allow_from"] = yuanbao_dm_allow_from
+        yuanbao_group_policy = os.getenv("YUANBAO_GROUP_POLICY")
+        if yuanbao_group_policy:
+            extra["group_policy"] = yuanbao_group_policy.strip().lower()
+        yuanbao_group_allow_from = os.getenv("YUANBAO_GROUP_ALLOW_FROM")
+        if yuanbao_group_allow_from:
+            extra["group_allow_from"] = yuanbao_group_allow_from
+
     # Session settings
     idle_minutes = os.getenv("SESSION_IDLE_MINUTES")
     if idle_minutes:

gateway/display_config.py

@@ -79,7 +79,9 @@ _PLATFORM_DEFAULTS: dict[str, dict[str, Any]] = {
     "discord":     _TIER_HIGH,
 
     # Tier 2 — edit support, often customer/workspace channels
-    "slack":           _TIER_MEDIUM,
+    # Slack: tool_progress off by default — Bolt posts cannot be edited like CLI;
+    # "new"/"all" spam permanent lines in channels (hermes-agent#14663).
+    "slack":           {**_TIER_MEDIUM, "tool_progress": "off"},
     "mattermost":      _TIER_MEDIUM,
     "matrix":          _TIER_MEDIUM,
     "feishu":          _TIER_MEDIUM,

gateway/mirror.py

@@ -28,6 +28,7 @@ def mirror_to_session(
     message_text: str,
     source_label: str = "cli",
     thread_id: Optional[str] = None,
+    user_id: Optional[str] = None,
 ) -> bool:
     """
     Append a delivery-mirror message to the target session's transcript.
@@ -39,9 +40,20 @@ def mirror_to_session(
     All errors are caught -- this is never fatal.
     """
     try:
-        session_id = _find_session_id(platform, str(chat_id), thread_id=thread_id)
+        session_id = _find_session_id(
+            platform,
+            str(chat_id),
+            thread_id=thread_id,
+            user_id=user_id,
+        )
         if not session_id:
-            logger.debug("Mirror: no session found for %s:%s:%s", platform, chat_id, thread_id)
+            logger.debug(
+                "Mirror: no session found for %s:%s:%s:%s",
+                platform,
+                chat_id,
+                thread_id,
+                user_id,
+            )
             return False
 
         mirror_msg = {
@@ -59,17 +71,33 @@ def mirror_to_session(
         return True
 
     except Exception as e:
-        logger.debug("Mirror failed for %s:%s:%s: %s", platform, chat_id, thread_id, e)
+        logger.debug(
+            "Mirror failed for %s:%s:%s:%s: %s",
+            platform,
+            chat_id,
+            thread_id,
+            user_id,
+            e,
+        )
         return False
 
 
-def _find_session_id(platform: str, chat_id: str, thread_id: Optional[str] = None) -> Optional[str]:
+def _find_session_id(
+    platform: str,
+    chat_id: str,
+    thread_id: Optional[str] = None,
+    user_id: Optional[str] = None,
+) -> Optional[str]:
     """
     Find the active session_id for a platform + chat_id pair.
 
     Scans sessions.json entries and matches where origin.chat_id == chat_id
     on the right platform.  DM session keys don't embed the chat_id
     (e.g. "agent:main:telegram:dm"), so we check the origin dict.
+
+    When *user_id* is provided, prefer exact sender matches. If multiple
+    same-chat candidates exist and none matches the user, return None instead
+    of guessing and contaminating another participant's session.
     """
     if not _SESSIONS_INDEX.exists():
         return None
@@ -81,8 +109,7 @@ def _find_session_id(platform: str, chat_id: str, thread_id: Optional[str] = Non
         return None
 
     platform_lower = platform.lower()
-    best_match = None
-    best_updated = ""
+    candidates = []
 
     for _key, entry in data.items():
         origin = entry.get("origin") or {}
@@ -96,12 +123,31 @@ def _find_session_id(platform: str, chat_id: str, thread_id: Optional[str] = Non
             origin_thread_id = origin.get("thread_id")
             if thread_id is not None and str(origin_thread_id or "") != str(thread_id):
                 continue
-            updated = entry.get("updated_at", "")
-            if updated > best_updated:
-                best_updated = updated
-                best_match = entry.get("session_id")
+            candidates.append(entry)
 
-    return best_match
+    if not candidates:
+        return None
+
+    if user_id:
+        exact_user_matches = [
+            entry for entry in candidates
+            if str((entry.get("origin") or {}).get("user_id") or "") == str(user_id)
+        ]
+        if exact_user_matches:
+            candidates = exact_user_matches
+        elif len(candidates) > 1:
+            return None
+    elif len(candidates) > 1:
+        distinct_user_ids = {
+            str((entry.get("origin") or {}).get("user_id") or "").strip()
+            for entry in candidates
+            if str((entry.get("origin") or {}).get("user_id") or "").strip()
+        }
+        if len(distinct_user_ids) > 1:
+            return None
+
+    best_entry = max(candidates, key=lambda entry: entry.get("updated_at", ""))
+    return best_entry.get("session_id")
 
 
 def _append_to_jsonl(session_id: str, message: dict) -> None:

gateway/platforms/__init__.py

@@ -10,10 +10,12 @@ Each adapter handles:
 
 from .base import BasePlatformAdapter, MessageEvent, SendResult
 from .qqbot import QQAdapter
+from .yuanbao import YuanbaoAdapter
 
 __all__ = [
     "BasePlatformAdapter",
     "MessageEvent",
     "SendResult",
     "QQAdapter",
+    "YuanbaoAdapter",
 ]

gateway/platforms/api_server.py

@@ -9,6 +9,7 @@ Exposes an HTTP server with endpoints:
 - GET  /v1/models                  — lists hermes-agent as an available model
 - POST /v1/runs                    — start a run, returns run_id immediately (202)
 - GET  /v1/runs/{run_id}/events    — SSE stream of structured lifecycle events
+- POST /v1/runs/{run_id}/stop    — interrupt a running agent
 - GET  /health                     — health check
 - GET  /health/detailed            — rich status for cross-container dashboard probing
 
@@ -586,6 +587,9 @@ class APIServerAdapter(BasePlatformAdapter):
         self._run_streams: Dict[str, "asyncio.Queue[Optional[Dict]]"] = {}
         # Creation timestamps for orphaned-run TTL sweep
         self._run_streams_created: Dict[str, float] = {}
+        # Active run agent/task references for stop support
+        self._active_run_agents: Dict[str, Any] = {}
+        self._active_run_tasks: Dict[str, "asyncio.Task"] = {}
         self._session_db: Optional[Any] = None  # Lazy-init SessionDB for session continuity
 
     @staticmethod
@@ -2441,6 +2445,7 @@ class APIServerAdapter(BasePlatformAdapter):
                     stream_delta_callback=_text_cb,
                     tool_progress_callback=event_cb,
                 )
+                self._active_run_agents[run_id] = agent
                 def _run_sync():
                     r = agent.run_conversation(
                         user_message=user_message,
@@ -2480,8 +2485,11 @@ class APIServerAdapter(BasePlatformAdapter):
                     q.put_nowait(None)
                 except Exception:
                     pass
+                self._active_run_agents.pop(run_id, None)
+                self._active_run_tasks.pop(run_id, None)
 
         task = asyncio.create_task(_run_and_close())
+        self._active_run_tasks[run_id] = task
         try:
             self._background_tasks.add(task)
         except TypeError:
@@ -2540,6 +2548,44 @@ class APIServerAdapter(BasePlatformAdapter):
 
         return response
 
+    async def _handle_stop_run(self, request: "web.Request") -> "web.Response":
+        """POST /v1/runs/{run_id}/stop — interrupt a running agent."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+
+        run_id = request.match_info["run_id"]
+        agent = self._active_run_agents.get(run_id)
+        task = self._active_run_tasks.get(run_id)
+
+        if agent is None and task is None:
+            return web.json_response(_openai_error(f"Run not found: {run_id}", code="run_not_found"), status=404)
+
+        if agent is not None:
+            try:
+                agent.interrupt("Stop requested via API")
+            except Exception:
+                pass
+
+        if task is not None and not task.done():
+            task.cancel()
+            # Bounded wait: run_conversation() executes in the default
+            # executor thread which task.cancel() cannot preempt — we rely on
+            # agent.interrupt() above to break the loop. Cap the wait so a
+            # slow/unresponsive interrupt can't hang this handler.
+            try:
+                await asyncio.wait_for(asyncio.shield(task), timeout=5.0)
+            except asyncio.TimeoutError:
+                logger.warning(
+                    "[api_server] stop for run %s timed out after 5s; "
+                    "agent may still be finishing the current step",
+                    run_id,
+                )
+            except (asyncio.CancelledError, Exception):
+                pass
+
+        return web.json_response({"run_id": run_id, "status": "stopping"})
+
     async def _sweep_orphaned_runs(self) -> None:
         """Periodically clean up run streams that were never consumed."""
         while True:
@@ -2554,6 +2600,8 @@ class APIServerAdapter(BasePlatformAdapter):
                 logger.debug("[api_server] sweeping orphaned run %s", run_id)
                 self._run_streams.pop(run_id, None)
                 self._run_streams_created.pop(run_id, None)
+                self._active_run_agents.pop(run_id, None)
+                self._active_run_tasks.pop(run_id, None)
 
     # ------------------------------------------------------------------
     # BasePlatformAdapter interface
@@ -2589,6 +2637,7 @@ class APIServerAdapter(BasePlatformAdapter):
             # Structured event streaming
             self._app.router.add_post("/v1/runs", self._handle_runs)
             self._app.router.add_get("/v1/runs/{run_id}/events", self._handle_run_events)
+            self._app.router.add_post("/v1/runs/{run_id}/stop", self._handle_stop_run)
             # Start background sweep to clean up orphaned (unconsumed) run streams
             sweep_task = asyncio.create_task(self._sweep_orphaned_runs())
             try:

gateway/platforms/base.py

@@ -336,6 +336,39 @@ def proxy_kwargs_for_aiohttp(proxy_url: str | None) -> tuple[dict, dict]:
     return {}, {"proxy": proxy_url}
 
 
+def is_host_excluded_by_no_proxy(hostname: str, no_proxy_value: str | None = None) -> bool:
+    """Return True when ``hostname`` matches a ``NO_PROXY`` entry.
+
+    Supports comma- or whitespace-separated entries with optional leading dots
+    and ``*.`` wildcards, which match both the apex domain and subdomains.
+    """
+    raw = no_proxy_value
+    if raw is None:
+        raw = os.environ.get("NO_PROXY") or os.environ.get("no_proxy") or ""
+
+    raw = raw.strip()
+    if not raw:
+        return False
+
+    lower_hostname = hostname.lower()
+    for entry in re.split(r"[\s,]+", raw):
+        normalized = entry.strip().lower()
+        if not normalized:
+            continue
+        if normalized == "*":
+            return True
+
+        if normalized.startswith("*."):
+            normalized = normalized[2:]
+        elif normalized.startswith("."):
+            normalized = normalized[1:]
+
+        if lower_hostname == normalized or lower_hostname.endswith(f".{normalized}"):
+            return True
+
+    return False
+
+
 from dataclasses import dataclass, field
 from datetime import datetime
 from pathlib import Path
@@ -693,7 +726,15 @@ SUPPORTED_DOCUMENT_TYPES = {
     ".pdf": "application/pdf",
     ".md": "text/markdown",
     ".txt": "text/plain",
+    ".csv": "text/csv",
     ".log": "text/plain",
+    ".json": "application/json",
+    ".xml": "application/xml",
+    ".yaml": "application/yaml",
+    ".yml": "application/yaml",
+    ".toml": "application/toml",
+    ".ini": "text/plain",
+    ".cfg": "text/plain",
     ".zip": "application/zip",
     ".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
     ".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
@@ -982,6 +1023,61 @@ def resolve_channel_prompt(
     return None
 
 
+def resolve_channel_skills(
+    config_extra: dict,
+    channel_id: str,
+    parent_id: str | None = None,
+) -> list[str] | None:
+    """Resolve auto-loaded skill(s) for a channel/thread from platform config.
+
+    Looks up ``channel_skill_bindings`` in the adapter's ``config.extra`` dict.
+
+    Config format::
+
+        channel_skill_bindings:
+          - id: "C0123"          # Slack channel ID or Discord channel/forum ID
+            skills: ["skill-a", "skill-b"]
+          - id: "D0ABCDE"
+            skill: "solo-skill"  # single string also accepted
+
+    Prefers an exact match on *channel_id*; falls back to *parent_id*
+    (useful for forum threads / Slack threads inheriting the parent channel's
+    binding).
+
+    Returns a deduplicated list of skill names (order preserved), or None if
+    no match is found.
+    """
+    bindings = config_extra.get("channel_skill_bindings") or []
+    if not isinstance(bindings, list) or not bindings:
+        return None
+    ids_to_check: set[str] = set()
+    if channel_id:
+        ids_to_check.add(str(channel_id))
+    if parent_id:
+        ids_to_check.add(str(parent_id))
+    if not ids_to_check:
+        return None
+    for entry in bindings:
+        if not isinstance(entry, dict):
+            continue
+        entry_id = str(entry.get("id", ""))
+        if entry_id in ids_to_check:
+            skills = entry.get("skills") or entry.get("skill")
+            if isinstance(skills, str):
+                s = skills.strip()
+                return [s] if s else None
+            if isinstance(skills, list) and skills:
+                seen: list[str] = []
+                for name in skills:
+                    if not isinstance(name, str):
+                        continue
+                    nm = name.strip()
+                    if nm and nm not in seen:
+                        seen.append(nm)
+                return seen or None
+    return None
+
+
 class BasePlatformAdapter(ABC):
     """
     Base class for platform adapters.
@@ -1025,7 +1121,20 @@ class BasePlatformAdapter(ABC):
         self._post_delivery_callbacks: Dict[str, Any] = {}
         self._expected_cancelled_tasks: set[asyncio.Task] = set()
         self._busy_session_handler: Optional[Callable[[MessageEvent, str], Awaitable[bool]]] = None
-        # Chats where auto-TTS on voice input is disabled (set by /voice off)
+        # Auto-TTS on voice input: ``_auto_tts_default`` is the global default
+        # (``voice.auto_tts`` in config.yaml, pushed by GatewayRunner on connect).
+        # Per-chat overrides live in two sets populated from ``_voice_mode``:
+        #   - ``_auto_tts_enabled_chats``: chat explicitly opted in via ``/voice on``
+        #     or ``/voice tts`` (mode is ``voice_only`` or ``all``). Fires even when
+        #     the global default is False.
+        #   - ``_auto_tts_disabled_chats``: chat explicitly opted out via
+        #     ``/voice off`` (mode is ``off``). Suppresses auto-TTS even when the
+        #     global default is True.
+        # The gate in _process_message() is:
+        #   fire if chat in _auto_tts_enabled_chats
+        #     OR (_auto_tts_default and chat not in _auto_tts_disabled_chats)
+        self._auto_tts_default: bool = False
+        self._auto_tts_enabled_chats: set = set()
         self._auto_tts_disabled_chats: set = set()
         # Chats where typing indicator is paused (e.g. during approval waits).
         # _keep_typing skips send_typing when the chat_id is in this set.
@@ -1047,6 +1156,21 @@ class BasePlatformAdapter(ABC):
     def fatal_error_retryable(self) -> bool:
         return self._fatal_error_retryable
 
+    def _should_auto_tts_for_chat(self, chat_id: str) -> bool:
+        """Whether auto-TTS on voice input should fire for ``chat_id``.
+
+        Decision layers (Issue #16007):
+          1. Explicit ``/voice on`` or ``/voice tts`` → always fire (even if
+             ``voice.auto_tts`` is False).
+          2. Explicit ``/voice off`` → never fire.
+          3. Fall back to the global ``voice.auto_tts`` config default.
+        """
+        if chat_id in self._auto_tts_enabled_chats:
+            return True
+        if chat_id in self._auto_tts_disabled_chats:
+            return False
+        return bool(self._auto_tts_default)
+
     def set_fatal_error_handler(self, handler: Callable[["BasePlatformAdapter"], Awaitable[None] | None]) -> None:
         self._fatal_error_handler = handler
 
@@ -1230,6 +1354,27 @@ class BasePlatformAdapter(ABC):
         """
         return SendResult(success=False, error="Not supported")
 
+    async def delete_message(
+        self,
+        chat_id: str,
+        message_id: str,
+    ) -> bool:
+        """
+        Delete a previously sent message.  Optional — platforms that don't
+        support deletion return ``False`` and callers fall back to leaving
+        the message in place.
+
+        Used by the stream consumer's fresh-final cleanup path (see
+        openclaw/openclaw#72038) to remove long-lived preview messages
+        after sending the completed reply as a fresh message so the
+        platform's visible timestamp reflects completion time.
+
+        Returns ``True`` on successful deletion, ``False`` otherwise.
+        Subclasses should override for platforms with a deletion API
+        (e.g. Telegram ``deleteMessage``).
+        """
+        return False
+
     async def send_typing(self, chat_id: str, metadata=None) -> None:
         """
         Send a typing indicator.
@@ -2214,12 +2359,14 @@ class BasePlatformAdapter(ABC):
                     logger.info("[%s] extract_local_files found %d file(s) in response", self.name, len(local_files))
                 
                 # Auto-TTS: if voice message, generate audio FIRST (before sending text)
-                # Skipped when the chat has voice mode disabled (/voice off)
+                # Gated via ``_should_auto_tts_for_chat``: fires when the chat has
+                # an explicit ``/voice on|tts`` opt-in OR when ``voice.auto_tts`` is
+                # True globally and no ``/voice off`` has been issued.
                 _tts_path = None
-                if (event.message_type == MessageType.VOICE
+                if (self._should_auto_tts_for_chat(event.source.chat_id)
+                        and event.message_type == MessageType.VOICE
                         and text_content
-                        and not media_files
-                        and event.source.chat_id not in self._auto_tts_disabled_chats):
+                        and not media_files):
                     try:
                         from tools.tts_tool import text_to_speech_tool, check_tts_requirements
                         if check_tts_requirements():
@@ -2543,6 +2690,9 @@ class BasePlatformAdapter(ABC):
         user_id_alt: Optional[str] = None,
         chat_id_alt: Optional[str] = None,
         is_bot: bool = False,
+        guild_id: Optional[str] = None,
+        parent_chat_id: Optional[str] = None,
+        message_id: Optional[str] = None,
     ) -> SessionSource:
         """Helper to build a SessionSource for this platform."""
         # Normalize empty topic to None
@@ -2560,6 +2710,9 @@ class BasePlatformAdapter(ABC):
             user_id_alt=user_id_alt,
             chat_id_alt=chat_id_alt,
             is_bot=is_bot,
+            guild_id=str(guild_id) if guild_id else None,
+            parent_chat_id=str(parent_chat_id) if parent_chat_id else None,
+            message_id=str(message_id) if message_id else None,
         )
     
     @abstractmethod

gateway/platforms/discord.py

@@ -2315,11 +2315,6 @@ class DiscordAdapter(BasePlatformAdapter):
         async def slash_background(interaction: discord.Interaction, prompt: str):
             await self._run_simple_slash(interaction, f"/background {prompt}", "Background task started~")
 
-        @tree.command(name="btw", description="Ephemeral side question using session context")
-        @discord.app_commands.describe(question="Your side question (no tools, not persisted)")
-        async def slash_btw(interaction: discord.Interaction, question: str):
-            await self._run_simple_slash(interaction, f"/btw {question}")
-
         # ── Auto-register any gateway-available commands not yet on the tree ──
         # This ensures new commands added to COMMAND_REGISTRY in
         # hermes_cli/commands.py automatically appear as Discord slash
@@ -2684,21 +2679,8 @@ class DiscordAdapter(BasePlatformAdapter):
                 skills: ["skill-a", "skill-b"]
         Also checks parent_id so forum threads inherit the forum's bindings.
         """
-        bindings = self.config.extra.get("channel_skill_bindings", [])
-        if not bindings:
-            return None
-        ids_to_check = {channel_id}
-        if parent_id:
-            ids_to_check.add(parent_id)
-        for entry in bindings:
-            entry_id = str(entry.get("id", ""))
-            if entry_id in ids_to_check:
-                skills = entry.get("skills") or entry.get("skill")
-                if isinstance(skills, str):
-                    return [skills]
-                if isinstance(skills, list) and skills:
-                    return list(dict.fromkeys(skills))  # dedup, preserve order
-        return None
+        from gateway.platforms.base import resolve_channel_skills
+        return resolve_channel_skills(self.config.extra, channel_id, parent_id)
 
     def _resolve_channel_prompt(self, channel_id: str, parent_id: str | None = None) -> str | None:
         """Resolve a Discord per-channel prompt, preferring the exact channel over its parent."""
@@ -3261,6 +3243,7 @@ class DiscordAdapter(BasePlatformAdapter):
             if auto_thread and not skip_thread and not is_voice_linked_channel and not is_reply_message:
                 thread = await self._auto_create_thread(message)
                 if thread:
+                    parent_channel_id = str(message.channel.id)
                     is_thread = True
                     thread_id = str(thread.id)
                     auto_threaded_channel = thread
@@ -3320,6 +3303,9 @@ class DiscordAdapter(BasePlatformAdapter):
             thread_id=thread_id,
             chat_topic=chat_topic,
             is_bot=getattr(message.author, "bot", False),
+            guild_id=str(message.guild.id) if message.guild else None,
+            parent_chat_id=parent_channel_id,
+            message_id=str(message.id),
         )
 
         # Build media URLs -- download image attachments to local cache so the

gateway/platforms/helpers.py

@@ -57,6 +57,15 @@ class MessageDeduplicator:
         if len(self._seen) > self._max_size:
             cutoff = now - self._ttl
             self._seen = {k: v for k, v in self._seen.items() if v > cutoff}
+            if len(self._seen) > self._max_size:
+                # TTL pruning alone does not cap the cache when every entry is
+                # still fresh. Keep the newest entries so the helper's
+                # max_size bound is enforced under sustained traffic.
+                newest = sorted(
+                    self._seen.items(),
+                    key=lambda item: item[1],
+                )[-self._max_size:]
+                self._seen = dict(newest)
         return False
 
     def clear(self):

gateway/platforms/telegram.py

@@ -1209,6 +1209,31 @@ class TelegramAdapter(BasePlatformAdapter):
             )
             return SendResult(success=False, error=str(e))
 
+    async def delete_message(self, chat_id: str, message_id: str) -> bool:
+        """Delete a previously sent Telegram message.
+
+        Used by the stream consumer's fresh-final cleanup path (ported
+        from openclaw/openclaw#72038) to remove long-lived preview
+        messages after sending the completed reply as a fresh message.
+        Telegram's Bot API ``deleteMessage`` works for bot-posted
+        messages in the last 48 hours.  Failures are non-fatal — the
+        caller leaves the preview in place and logs at debug level.
+        """
+        if not self._bot:
+            return False
+        try:
+            await self._bot.delete_message(
+                chat_id=int(chat_id),
+                message_id=int(message_id),
+            )
+            return True
+        except Exception as e:
+            logger.debug(
+                "[%s] Failed to delete Telegram message %s: %s",
+                self.name, message_id, e,
+            )
+            return False
+
     async def send_update_prompt(
         self, chat_id: str, prompt: str, default: str = "",
         session_key: str = "",

gateway/platforms/yuanbao_media.py

@@ -0,0 +1,647 @@
+"""
+yuanbao_media.py — 元宝平台媒体处理模块
+
+提供 COS 上传、文件下载、TIM 媒体消息构建等功能。
+移植自 TypeScript 版 media.ts（yuanbao-openclaw-plugin），
+使用 httpx 替代 cos-nodejs-sdk-v5，避免引入额外 SDK 依赖。
+
+COS 上传流程：
+  1. 调用 genUploadInfo 获取临时凭证（tmpSecretId/tmpSecretKey/sessionToken）
+  2. 用临时凭证通过 HMAC-SHA1 签名构建 Authorization 头
+  3. HTTP PUT 上传到 COS
+
+TIM 消息体构建：
+  - buildImageMsgBody() → TIMImageElem
+  - buildFileMsgBody()  → TIMFileElem
+"""
+
+from __future__ import annotations
+
+import hashlib
+import hmac
+import logging
+import os
+import re
+import secrets
+import struct
+import time
+import urllib.parse
+from datetime import datetime, timezone, timedelta
+from typing import Optional, Any
+
+import httpx
+
+logger = logging.getLogger(__name__)
+
+# ============ 常量 ============
+
+UPLOAD_INFO_PATH = "/api/resource/genUploadInfo"
+DEFAULT_API_DOMAIN = "yuanbao.tencent.com"
+DEFAULT_MAX_SIZE_MB = 50
+
+# COS 加速域名后缀（优先使用全球加速）
+COS_USE_ACCELERATE = True
+
+# ============ 类型映射 ============
+
+# MIME → image_format 数字（TIM 协议字段）
+_MIME_TO_IMAGE_FORMAT: dict[str, int] = {
+    "image/jpeg": 1,
+    "image/jpg": 1,
+    "image/gif": 2,
+    "image/png": 3,
+    "image/bmp": 4,
+    "image/webp": 255,
+    "image/heic": 255,
+    "image/tiff": 255,
+}
+
+# 文件扩展名 → MIME
+_EXT_TO_MIME: dict[str, str] = {
+    ".jpg": "image/jpeg",
+    ".jpeg": "image/jpeg",
+    ".png": "image/png",
+    ".gif": "image/gif",
+    ".webp": "image/webp",
+    ".bmp": "image/bmp",
+    ".heic": "image/heic",
+    ".tiff": "image/tiff",
+    ".ico": "image/x-icon",
+    ".pdf": "application/pdf",
+    ".doc": "application/msword",
+    ".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    ".xls": "application/vnd.ms-excel",
+    ".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+    ".ppt": "application/vnd.ms-powerpoint",
+    ".pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+    ".txt": "text/plain",
+    ".zip": "application/zip",
+    ".tar": "application/x-tar",
+    ".gz": "application/gzip",
+    ".mp3": "audio/mpeg",
+    ".mp4": "video/mp4",
+    ".wav": "audio/wav",
+    ".ogg": "audio/ogg",
+    ".webm": "video/webm",
+}
+
+
+# ============ 工具函数 ============
+
+def guess_mime_type(filename: str) -> str:
+    """根据文件扩展名猜测 MIME 类型。"""
+    ext = os.path.splitext(filename)[-1].lower()
+    return _EXT_TO_MIME.get(ext, "application/octet-stream")
+
+
+def is_image(filename: str, mime_type: str = "") -> bool:
+    """判断是否为图片类型。"""
+    if mime_type.startswith("image/"):
+        return True
+    ext = os.path.splitext(filename)[-1].lower()
+    return ext in {".jpg", ".jpeg", ".png", ".gif", ".webp", ".bmp", ".heic", ".tiff", ".ico"}
+
+
+def get_image_format(mime_type: str) -> int:
+    """获取 TIM 图片格式编号。"""
+    return _MIME_TO_IMAGE_FORMAT.get(mime_type.lower(), 255)
+
+
+def md5_hex(data: bytes) -> str:
+    """计算 MD5 十六进制摘要。"""
+    return hashlib.md5(data).hexdigest()
+
+
+def generate_file_id() -> str:
+    """生成随机文件 ID（32 位 hex）。"""
+    return secrets.token_hex(16)
+
+
+
+# ============ 图片尺寸解析（纯 Python，无需 Pillow） ============
+
+def parse_image_size(data: bytes) -> Optional[dict[str, int]]:
+    """
+    解析图片宽高（支持 JPEG/PNG/GIF/WebP），无需第三方依赖。
+    返回 {"width": w, "height": h} 或 None（无法识别）。
+    """
+    return (
+        _parse_png_size(data)
+        or _parse_jpeg_size(data)
+        or _parse_gif_size(data)
+        or _parse_webp_size(data)
+    )
+
+
+def _parse_png_size(buf: bytes) -> Optional[dict[str, int]]:
+    if len(buf) < 24:
+        return None
+    if buf[:4] != b"\x89PNG":
+        return None
+    w = struct.unpack(">I", buf[16:20])[0]
+    h = struct.unpack(">I", buf[20:24])[0]
+    return {"width": w, "height": h}
+
+
+def _parse_jpeg_size(buf: bytes) -> Optional[dict[str, int]]:
+    if len(buf) < 4 or buf[0] != 0xFF or buf[1] != 0xD8:
+        return None
+    i = 2
+    while i < len(buf) - 9:
+        if buf[i] != 0xFF:
+            i += 1
+            continue
+        marker = buf[i + 1]
+        if marker in (0xC0, 0xC2):
+            h = struct.unpack(">H", buf[i + 5: i + 7])[0]
+            w = struct.unpack(">H", buf[i + 7: i + 9])[0]
+            return {"width": w, "height": h}
+        if i + 3 < len(buf):
+            i += 2 + struct.unpack(">H", buf[i + 2: i + 4])[0]
+        else:
+            break
+    return None
+
+
+def _parse_gif_size(buf: bytes) -> Optional[dict[str, int]]:
+    if len(buf) < 10:
+        return None
+    sig = buf[:6].decode("ascii", errors="replace")
+    if sig not in ("GIF87a", "GIF89a"):
+        return None
+    w = struct.unpack("<H", buf[6:8])[0]
+    h = struct.unpack("<H", buf[8:10])[0]
+    return {"width": w, "height": h}
+
+
+def _parse_webp_size(buf: bytes) -> Optional[dict[str, int]]:
+    if len(buf) < 16:
+        return None
+    if buf[:4] != b"RIFF" or buf[8:12] != b"WEBP":
+        return None
+    chunk = buf[12:16].decode("ascii", errors="replace")
+    if chunk == "VP8 ":
+        if len(buf) >= 30 and buf[23] == 0x9D and buf[24] == 0x01 and buf[25] == 0x2A:
+            w = struct.unpack("<H", buf[26:28])[0] & 0x3FFF
+            h = struct.unpack("<H", buf[28:30])[0] & 0x3FFF
+            return {"width": w, "height": h}
+    elif chunk == "VP8L":
+        if len(buf) >= 25 and buf[20] == 0x2F:
+            bits = struct.unpack("<I", buf[21:25])[0]
+            w = (bits & 0x3FFF) + 1
+            h = ((bits >> 14) & 0x3FFF) + 1
+            return {"width": w, "height": h}
+    elif chunk == "VP8X":
+        if len(buf) >= 30:
+            w = (buf[24] | (buf[25] << 8) | (buf[26] << 16)) + 1
+            h = (buf[27] | (buf[28] << 8) | (buf[29] << 16)) + 1
+            return {"width": w, "height": h}
+    return None
+
+
+# ============ URL 下载 ============
+
+async def download_url(
+    url: str,
+    max_size_mb: int = DEFAULT_MAX_SIZE_MB,
+) -> tuple[bytes, str]:
+    """
+    下载 URL 内容，返回 (bytes, content_type)。
+
+    Args:
+        url:          HTTP(S) URL
+        max_size_mb:  最大允许大小（MB），超过则抛出异常
+
+    Returns:
+        (data_bytes, content_type_string)
+
+    Raises:
+        ValueError:  内容超过大小限制
+        httpx.HTTPError: 网络/HTTP 错误
+    """
+    max_bytes = max_size_mb * 1024 * 1024
+    async with httpx.AsyncClient(timeout=30.0, follow_redirects=True) as client:
+        # 先 HEAD 检查大小
+        try:
+            head = await client.head(url)
+            content_length = int(head.headers.get("content-length", 0) or 0)
+            if content_length > 0 and content_length > max_bytes:
+                raise ValueError(
+                    f"文件过大: {content_length / 1024 / 1024:.1f} MB > {max_size_mb} MB"
+                )
+        except httpx.HTTPStatusError:
+            pass  # 部分服务器不支持 HEAD，忽略
+
+        # GET 下载（流式读取，防止超限）
+        async with client.stream("GET", url) as resp:
+            resp.raise_for_status()
+
+            content_type = resp.headers.get("content-type", "").split(";")[0].strip()
+
+            chunks: list[bytes] = []
+            downloaded = 0
+            async for chunk in resp.aiter_bytes(65536):
+                downloaded += len(chunk)
+                if downloaded > max_bytes:
+                    raise ValueError(
+                        f"文件过大: 已超过 {max_size_mb} MB 限制"
+                    )
+                chunks.append(chunk)
+
+        data = b"".join(chunks)
+        return data, content_type
+
+
+# ============ COS 鉴权（HMAC-SHA1） ============
+
+def _cos_sign(
+    method: str,
+    path: str,
+    params: dict[str, str],
+    headers: dict[str, str],
+    secret_id: str,
+    secret_key: str,
+    start_time: Optional[int] = None,
+    expire_seconds: int = 3600,
+) -> str:
+    """
+    构建 COS 请求签名（q-sign-algorithm=sha1 方案）。
+    参考：https://cloud.tencent.com/document/product/436/7778
+
+    Args:
+        method:         HTTP 方法（小写，如 "put"）
+        path:           URL 路径（URL encode 后的小写）
+        params:         URL 查询参数 dict（用于签名）
+        headers:        参与签名的请求头 dict（key 需小写）
+        secret_id:      临时 SecretId（tmpSecretId）
+        secret_key:     临时 SecretKey（tmpSecretKey）
+        start_time:     签名起始 Unix 时间戳（默认 now）
+        expire_seconds: 签名有效期（秒，默认 3600）
+
+    Returns:
+        Authorization header 值（完整字符串）
+    """
+    now = int(time.time())
+    q_sign_time = f"{start_time or now};{(start_time or now) + expire_seconds}"
+
+    # Step 1: SignKey = HMAC-SHA1(SecretKey, q-sign-time)
+    sign_key = hmac.new(
+        secret_key.encode("utf-8"),
+        q_sign_time.encode("utf-8"),
+        hashlib.sha1,
+    ).hexdigest()
+
+    # Step 2: HttpString
+    # 参数和头部需按字典序排列，key 小写
+    sorted_params = sorted((k.lower(), urllib.parse.quote(str(v), safe="") ) for k, v in params.items())
+    sorted_headers = sorted((k.lower(), urllib.parse.quote(str(v), safe="") ) for k, v in headers.items())
+
+    url_param_list = ";".join(k for k, _ in sorted_params)
+    url_params = "&".join(f"{k}={v}" for k, v in sorted_params)
+    header_list = ";".join(k for k, _ in sorted_headers)
+    header_str = "&".join(f"{k}={v}" for k, v in sorted_headers)
+
+    http_string = "\n".join([
+        method.lower(),
+        path,
+        url_params,
+        header_str,
+        "",
+    ])
+
+    # Step 3: StringToSign = sha1 hash of HttpString
+    sha1_of_http = hashlib.sha1(http_string.encode("utf-8")).hexdigest()
+    string_to_sign = "\n".join([
+        "sha1",
+        q_sign_time,
+        sha1_of_http,
+        "",
+    ])
+
+    # Step 4: Signature = HMAC-SHA1(SignKey, StringToSign)
+    signature = hmac.new(
+        sign_key.encode("utf-8"),
+        string_to_sign.encode("utf-8"),
+        hashlib.sha1,
+    ).hexdigest()
+
+    return (
+        f"q-sign-algorithm=sha1"
+        f"&q-ak={secret_id}"
+        f"&q-sign-time={q_sign_time}"
+        f"&q-key-time={q_sign_time}"
+        f"&q-header-list={header_list}"
+        f"&q-url-param-list={url_param_list}"
+        f"&q-signature={signature}"
+    )
+
+
+# ============ 主要公开 API ============
+
+async def get_cos_credentials(
+    app_key: str,
+    api_domain: str,
+    token: str,
+    filename: str = "file",
+    file_id: Optional[str] = None,
+    bot_id: str = "",
+    route_env: str = "",
+) -> dict:
+    """
+    调用 genUploadInfo 接口获取 COS 临时密钥及上传配置。
+
+    Args:
+        app_key:        应用 Key（用于 X-ID 头）
+        api_domain:     API 域名（如 https://bot.yuanbao.tencent.com）
+        token:          当前有效的签票 token（X-Token 头）
+        filename:       待上传的文件名（含扩展名）
+        file_id:        客户端生成的唯一文件 ID（不传则自动生成）
+        bot_id:         Bot 账号 ID（用于 X-ID 头）
+
+    Returns:
+        COS 上传配置 dict，包含以下字段：
+            bucketName         (str)  — COS Bucket 名称
+            region             (str)  — COS 地域
+            location           (str)  — 上传 Key（对象路径）
+            encryptTmpSecretId (str)  — 临时 SecretId
+            encryptTmpSecretKey(str)  — 临时 SecretKey
+            encryptToken       (str)  — SessionToken
+            startTime          (int)  — 凭证起始时间戳（Unix）
+            expiredTime        (int)  — 凭证过期时间戳（Unix）
+            resourceUrl        (str)  — 上传后的公网访问 URL
+            resourceID         (str)  — 资源 ID（可选）
+
+    Raises:
+        RuntimeError: 接口返回非 0 code 或字段缺失
+    """
+    if file_id is None:
+        file_id = generate_file_id()
+
+    upload_url = f"{api_domain.rstrip('/')}{UPLOAD_INFO_PATH}"
+
+    headers = {
+        "Content-Type": "application/json",
+        "X-Token": token,
+        "X-ID": bot_id or app_key,
+        "X-Source": "web",
+    }
+    if route_env:
+        headers["X-Route-Env"] = route_env
+    body = {
+        "fileName": filename,
+        "fileId": file_id,
+        "docFrom": "localDoc",
+        "docOpenId": "",
+    }
+
+    async with httpx.AsyncClient(timeout=15.0) as client:
+        resp = await client.post(upload_url, json=body, headers=headers)
+        resp.raise_for_status()
+        result: dict[str, Any] = resp.json()
+
+    code = result.get("code")
+    if code != 0 and code is not None:
+        raise RuntimeError(
+            f"genUploadInfo 失败: code={code}, msg={result.get('msg', '')}"
+        )
+
+    data = result.get("data") or result
+    required_fields = ["bucketName", "location"]
+    missing = [f for f in required_fields if not data.get(f)]
+    if missing:
+        raise RuntimeError(
+            f"genUploadInfo 返回字段不完整: 缺少字段 {missing}"
+        )
+
+    return data
+
+
+async def upload_to_cos(
+    file_bytes: bytes,
+    filename: str,
+    content_type: str,
+    credentials: dict,
+    bucket: str,
+    region: str,
+) -> dict:
+    """
+    通过 httpx PUT 请求将文件上传到 COS。
+    使用临时凭证（tmpSecretId/tmpSecretKey/sessionToken）构建 HMAC-SHA1 签名。
+
+    Args:
+        file_bytes:   文件二进制内容
+        filename:     文件名（用于辅助计算 MIME、UUID）
+        content_type: MIME 类型（如 "image/jpeg"）
+        credentials:  get_cos_credentials() 返回的 dict，包含：
+                        encryptTmpSecretId  → tmpSecretId
+                        encryptTmpSecretKey → tmpSecretKey
+                        encryptToken        → sessionToken
+                        location            → COS key（对象路径）
+                        resourceUrl         → 上传后公网 URL
+                        startTime           → 凭证起始时间（Unix）
+                        expiredTime         → 凭证过期时间（Unix）
+        bucket:       COS Bucket 名称（如 chatbot-1234567890）
+        region:       COS 地域（如 ap-guangzhou）
+
+    Returns:
+        上传结果 dict，包含：
+            url       (str)           — COS 公网访问 URL
+            uuid      (str)           — 文件内容 MD5
+            size      (int)           — 文件大小（字节）
+            width     (int, optional) — 图片宽度（仅图片）
+            height    (int, optional) — 图片高度（仅图片）
+
+    Raises:
+        httpx.HTTPStatusError: COS 返回非 2xx 状态
+        RuntimeError:          credentials 字段缺失
+    """
+    secret_id: str = credentials.get("encryptTmpSecretId", "")
+    secret_key: str = credentials.get("encryptTmpSecretKey", "")
+    session_token: str = credentials.get("encryptToken", "")
+    cos_key: str = credentials.get("location", "")
+    resource_url: str = credentials.get("resourceUrl", "")
+    start_time: Optional[int] = credentials.get("startTime")
+    expired_time: Optional[int] = credentials.get("expiredTime")
+
+    if not secret_id or not secret_key or not cos_key:
+        raise RuntimeError(
+            f"COS credentials 不完整: secretId={bool(secret_id)}, "
+            f"secretKey={bool(secret_key)}, location={bool(cos_key)}"
+        )
+
+    # 构建 COS 上传 URL（优先使用全球加速域名）
+    if COS_USE_ACCELERATE:
+        cos_host = f"{bucket}.cos.accelerate.myqcloud.com"
+    else:
+        cos_host = f"{bucket}.cos.{region}.myqcloud.com"
+
+    # URL encode cos_key（保留 /）
+    encoded_key = urllib.parse.quote(cos_key, safe="/")
+    cos_url = f"https://{cos_host}/{encoded_key.lstrip('/')}"
+
+    # 确定 Content-Type
+    if not content_type or content_type == "application/octet-stream":
+        if is_image(filename):
+            content_type = guess_mime_type(filename)
+        else:
+            content_type = "application/octet-stream"
+
+    # 计算文件 MD5 + size
+    file_uuid = md5_hex(file_bytes)
+    file_size = len(file_bytes)
+
+    # 参与签名的请求头
+    sign_headers = {
+        "host": cos_host,
+        "content-type": content_type,
+        "x-cos-security-token": session_token,
+    }
+
+    # 计算签名有效期
+    now = int(time.time())
+    sign_start = start_time if start_time else now
+    sign_expire = (expired_time - now) if expired_time and expired_time > now else 3600
+
+    authorization = _cos_sign(
+        method="put",
+        path=f"/{encoded_key.lstrip('/')}",
+        params={},
+        headers=sign_headers,
+        secret_id=secret_id,
+        secret_key=secret_key,
+        start_time=sign_start,
+        expire_seconds=sign_expire,
+    )
+
+    put_headers = {
+        "Authorization": authorization,
+        "Content-Type": content_type,
+        "x-cos-security-token": session_token,
+    }
+
+    logger.info(
+        "COS PUT: bucket=%s region=%s key=%s size=%d mime=%s",
+        bucket, region, cos_key, file_size, content_type,
+    )
+
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        resp = await client.put(
+            cos_url,
+            content=file_bytes,
+            headers=put_headers,
+        )
+        resp.raise_for_status()
+
+    # 解析图片尺寸（仅图片类型）
+    result: dict[str, Any] = {
+        "url": resource_url or cos_url,
+        "uuid": file_uuid,
+        "size": file_size,
+    }
+
+    if content_type.startswith("image/"):
+        size_info = parse_image_size(file_bytes)
+        if size_info:
+            result["width"] = size_info["width"]
+            result["height"] = size_info["height"]
+
+    logger.info(
+        "COS 上传成功: url=%s size=%d",
+        result["url"], file_size,
+    )
+    return result
+
+
+# ============ TIM 媒体消息构建 ============
+
+def build_image_msg_body(
+    url: str,
+    uuid: Optional[str] = None,
+    filename: Optional[str] = None,
+    size: int = 0,
+    width: int = 0,
+    height: int = 0,
+    mime_type: str = "",
+) -> list[dict]:
+    """
+    构建腾讯 IM TIMImageElem 消息体。
+    参考：https://cloud.tencent.com/document/product/269/2720
+
+    Args:
+        url:       图片公网访问 URL（COS resourceUrl）
+        uuid:      文件 UUID（MD5 或其他唯一标识）
+        filename:  文件名（uuid 为空时作为备用）
+        size:      文件大小（字节）
+        width:     图片宽度（像素）
+        height:    图片高度（像素）
+        mime_type: MIME 类型（用于确定 image_format）
+
+    Returns:
+        TIMImageElem 消息体列表（适合直接放入 msg_body）
+    """
+    _uuid = uuid or filename or _basename_from_url(url) or "image"
+    image_format = get_image_format(mime_type) if mime_type else 255
+
+    return [
+        {
+            "msg_type": "TIMImageElem",
+            "msg_content": {
+                "uuid": _uuid,
+                "image_format": image_format,
+                "image_info_array": [
+                    {
+                        "type": 1,       # 1 = 原图
+                        "size": size,
+                        "width": width,
+                        "height": height,
+                        "url": url,
+                    }
+                ],
+            },
+        }
+    ]
+
+
+def build_file_msg_body(
+    url: str,
+    filename: str,
+    uuid: Optional[str] = None,
+    size: int = 0,
+) -> list[dict]:
+    """
+    构建腾讯 IM TIMFileElem 消息体。
+    参考：https://cloud.tencent.com/document/product/269/2720
+
+    Args:
+        url:      文件公网访问 URL（COS resourceUrl）
+        filename: 文件名（含扩展名）
+        uuid:     文件 UUID（MD5 或其他唯一标识，不传则使用 filename）
+        size:     文件大小（字节）
+
+    Returns:
+        TIMFileElem 消息体列表（适合直接放入 msg_body）
+    """
+    _uuid = uuid or filename
+
+    return [
+        {
+            "msg_type": "TIMFileElem",
+            "msg_content": {
+                "uuid": _uuid,
+                "file_name": filename,
+                "file_size": size,
+                "url": url,
+            },
+        }
+    ]
+
+
+# ============ 内部工具 ============
+
+def _basename_from_url(url: str) -> str:
+    """从 URL 提取文件名。"""
+    try:
+        parsed = urllib.parse.urlparse(url)
+        return os.path.basename(parsed.path)
+    except Exception:
+        return ""

gateway/platforms/yuanbao_sticker.py

@@ -0,0 +1,558 @@
+"""
+Yuanbao sticker (TIMFaceElem) support.
+
+Ported from yuanbao-openclaw-plugin/src/sticker/.
+
+TIMFaceElem wire format:
+    {
+        "msg_type": "TIMFaceElem",
+        "msg_content": {
+            "index": 0,          # always 0 per Yuanbao convention
+            "data": "<json>",    # serialised sticker metadata
+        }
+    }
+
+The `data` field carries a JSON string with the sticker's metadata so the
+receiver can look up the correct asset in the emoji pack.
+"""
+
+from __future__ import annotations
+
+import json
+import random
+import re
+import unicodedata
+from typing import Optional
+
+# ---------------------------------------------------------------------------
+# Sticker catalogue – ported from builtin-stickers.json
+# Key   : canonical name (Chinese)
+# Value : {sticker_id, package_id, name, description, width, height, formats}
+# ---------------------------------------------------------------------------
+STICKER_MAP: dict[str, dict] = {
+    "六六六": {
+        "sticker_id": "278", "package_id": "1003", "name": "六六六",
+        "description": "666 厉害 牛 棒 绝了 好强 awesome",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "我想开了": {
+        "sticker_id": "262", "package_id": "1003", "name": "我想开了",
+        "description": "想开 佛系 释怀 顿悟 看淡了 无所谓",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "害羞": {
+        "sticker_id": "130", "package_id": "1003", "name": "害羞",
+        "description": "腼腆 不好意思 脸红 娇羞 羞涩 捂脸",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "比心": {
+        "sticker_id": "252", "package_id": "1003", "name": "比心",
+        "description": "笔芯 爱你 爱心手势 love heart 喜欢你",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "委屈": {
+        "sticker_id": "125", "package_id": "1003", "name": "委屈",
+        "description": "难过 想哭 可怜巴巴 瘪嘴 受伤 被欺负",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "亲亲": {
+        "sticker_id": "146", "package_id": "1003", "name": "亲亲",
+        "description": "么么 mua 亲一下 kiss 飞吻 啵",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "酷": {
+        "sticker_id": "131", "package_id": "1003", "name": "酷",
+        "description": "帅 墨镜 cool 高冷 有型 swagger",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "睡": {
+        "sticker_id": "145", "package_id": "1003", "name": "睡",
+        "description": "睡觉 困 zzZ 打盹 躺平 休眠 sleepy",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "发呆": {
+        "sticker_id": "152", "package_id": "1003", "name": "发呆",
+        "description": "懵 愣住 放空 呆滞 出神 脑子空白",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "可怜": {
+        "sticker_id": "157", "package_id": "1003", "name": "可怜",
+        "description": "卖萌 求饶 委屈巴巴 弱小 拜托 眼巴巴",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "摊手": {
+        "sticker_id": "200", "package_id": "1003", "name": "摊手",
+        "description": "无奈 没办法 耸肩 随便 那咋整 whatever",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "头大": {
+        "sticker_id": "213", "package_id": "1003", "name": "头大",
+        "description": "头疼 烦恼 郁闷 难搞 崩溃 一团乱",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "吓": {
+        "sticker_id": "256", "package_id": "1003", "name": "吓",
+        "description": "害怕 惊恐 震惊 吓一跳 恐怖 怂",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "吐血": {
+        "sticker_id": "203", "package_id": "1003", "name": "吐血",
+        "description": "无语 崩溃 被雷 内伤 一口老血 屮",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "哼": {
+        "sticker_id": "185", "package_id": "1003", "name": "哼",
+        "description": "傲娇 生气 不满 撇嘴 不理 赌气",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "嘿嘿": {
+        "sticker_id": "220", "package_id": "1003", "name": "嘿嘿",
+        "description": "坏笑 猥琐笑 偷笑 憨笑 得意 你懂的",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "头秃": {
+        "sticker_id": "218", "package_id": "1003", "name": "头秃",
+        "description": "程序员 加班 焦虑 没头发 秃了 肝爆",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "暗中观察": {
+        "sticker_id": "221", "package_id": "1003", "name": "暗中观察",
+        "description": "窥屏 潜水 偷偷看 角落 围观 屏住呼吸",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "我酸了": {
+        "sticker_id": "224", "package_id": "1003", "name": "我酸了",
+        "description": "嫉妒 柠檬精 羡慕 吃柠檬 眼红 恰柠檬",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "打call": {
+        "sticker_id": "246", "package_id": "1003", "name": "打call",
+        "description": "应援 加油 支持 喝彩 助威 call",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "庆祝": {
+        "sticker_id": "251", "package_id": "1003", "name": "庆祝",
+        "description": "祝贺 开心 耶 party 胜利 干杯",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "奋斗": {
+        "sticker_id": "151", "package_id": "1003", "name": "奋斗",
+        "description": "努力 加油 拼搏 冲 干劲 卷起来",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "惊讶": {
+        "sticker_id": "143", "package_id": "1003", "name": "惊讶",
+        "description": "震惊 哇 不敢相信 OMG 居然 这么离谱",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "疑问": {
+        "sticker_id": "144", "package_id": "1003", "name": "疑问",
+        "description": "问号 不懂 啥 为什么 啥情况 懵逼问",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "仔细分析": {
+        "sticker_id": "248", "package_id": "1003", "name": "仔细分析",
+        "description": "思考 推敲 认真 研究 琢磨 让我想想",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "撅嘴": {
+        "sticker_id": "184", "package_id": "1003", "name": "撅嘴",
+        "description": "嘟嘴 卖萌 不高兴 撒娇 嘴翘",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "泪奔": {
+        "sticker_id": "199", "package_id": "1003", "name": "泪奔",
+        "description": "大哭 伤心 破防 感动哭 泪流满面 呜呜",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "尊嘟假嘟": {
+        "sticker_id": "276", "package_id": "1003", "name": "尊嘟假嘟",
+        "description": "真的假的 真假 可爱问 你骗我 是不是",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "略略略": {
+        "sticker_id": "113", "package_id": "1003", "name": "略略略",
+        "description": "调皮 吐舌 不服 略 气死你 鬼脸",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "困": {
+        "sticker_id": "180", "package_id": "1003", "name": "困",
+        "description": "想睡 倦 打哈欠 睁不开眼 好困啊 sleepy",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "折磨": {
+        "sticker_id": "181", "package_id": "1003", "name": "折磨",
+        "description": "难受 痛苦 煎熬 蚌埠住了 受不了 要命",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "抠鼻": {
+        "sticker_id": "182", "package_id": "1003", "name": "抠鼻",
+        "description": "不屑 无聊 淡定 无所谓 鄙视 挖鼻",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "鼓掌": {
+        "sticker_id": "183", "package_id": "1003", "name": "鼓掌",
+        "description": "拍手 叫好 赞同 666 喝彩 掌声",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "斜眼笑": {
+        "sticker_id": "204", "package_id": "1003", "name": "斜眼笑",
+        "description": "滑稽 坏笑 doge 意味深长 阴阳怪气 嘿嘿嘿",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "辣眼睛": {
+        "sticker_id": "216", "package_id": "1003", "name": "辣眼睛",
+        "description": "看不下去 cringe 毁三观 太丑了 瞎了",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "哦哟": {
+        "sticker_id": "217", "package_id": "1003", "name": "哦哟",
+        "description": "惊讶 起哄 哇哦 有戏 不简单 哟",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "吃瓜": {
+        "sticker_id": "222", "package_id": "1003", "name": "吃瓜",
+        "description": "围观 看戏 八卦 路人 看热闹 板凳",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "狗头": {
+        "sticker_id": "225", "package_id": "1003", "name": "狗头",
+        "description": "doge 保命 开玩笑 滑稽 反讽 懂的都懂",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "敬礼": {
+        "sticker_id": "227", "package_id": "1003", "name": "敬礼",
+        "description": "salute 尊重 收到 遵命 致敬 报告",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "哦": {
+        "sticker_id": "231", "package_id": "1003", "name": "哦",
+        "description": "知道了 明白 敷衍 嗯 这样啊 收到",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "拿到红包": {
+        "sticker_id": "236", "package_id": "1003", "name": "拿到红包",
+        "description": "红包 谢谢老板 发财 开心 抢到了 欧气",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "牛吖": {
+        "sticker_id": "239", "package_id": "1003", "name": "牛吖",
+        "description": "牛 厉害 强 666 佩服 大佬",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "贴贴": {
+        "sticker_id": "272", "package_id": "1003", "name": "贴贴",
+        "description": "抱抱 亲昵 蹭蹭 亲密 靠靠 撒娇贴",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "爱心": {
+        "sticker_id": "138", "package_id": "1003", "name": "爱心",
+        "description": "心 love 喜欢你 红心 示爱 么么哒",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "晚安": {
+        "sticker_id": "170", "package_id": "1003", "name": "晚安",
+        "description": "好梦 睡了 night 早点休息 安啦 moon",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "太阳": {
+        "sticker_id": "176", "package_id": "1003", "name": "太阳",
+        "description": "晴天 早上好 阳光 morning 好天气 日",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "柠檬": {
+        "sticker_id": "266", "package_id": "1003", "name": "柠檬",
+        "description": "酸 嫉妒 柠檬精 羡慕 我酸 恰柠檬",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "大冤种": {
+        "sticker_id": "267", "package_id": "1003", "name": "大冤种",
+        "description": "倒霉 吃亏 自嘲 好心没好报 背锅 工具人",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "吐了": {
+        "sticker_id": "132", "package_id": "1003", "name": "吐了",
+        "description": "恶心 yue 受不了 嫌弃 想吐 生理不适",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "怒": {
+        "sticker_id": "134", "package_id": "1003", "name": "怒",
+        "description": "生气 愤怒 火大 暴躁 气炸 怼",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "玫瑰": {
+        "sticker_id": "165", "package_id": "1003", "name": "玫瑰",
+        "description": "花 示爱 表白 浪漫 送你花 情人节",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "凋谢": {
+        "sticker_id": "119", "package_id": "1003", "name": "凋谢",
+        "description": "花谢 失恋 难过 枯萎 心碎 凉了",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "点赞": {
+        "sticker_id": "159", "package_id": "1003", "name": "点赞",
+        "description": "赞 认同 好棒 good like 大拇指 顶",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "握手": {
+        "sticker_id": "164", "package_id": "1003", "name": "握手",
+        "description": "合作 你好 商务 hello deal 成交 友好",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "抱拳": {
+        "sticker_id": "163", "package_id": "1003", "name": "抱拳",
+        "description": "谢谢 失敬 江湖 承让 拜托 有礼",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "ok": {
+        "sticker_id": "169", "package_id": "1003", "name": "ok",
+        "description": "好的 收到 没问题 okay 行 可以 懂了",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "拳头": {
+        "sticker_id": "174", "package_id": "1003", "name": "拳头",
+        "description": "加油 干 冲 fight 力量 击拳 硬气",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "鞭炮": {
+        "sticker_id": "191", "package_id": "1003", "name": "鞭炮",
+        "description": "过年 喜庆 爆竹 春节 噼里啪啦 红",
+        "width": 128, "height": 128, "formats": "png",
+    },
+    "烟花": {
+        "sticker_id": "258", "package_id": "1003", "name": "烟花",
+        "description": "庆典 漂亮 新年 嘭 绽放 节日快乐",
+        "width": 128, "height": 128, "formats": "png",
+    },
+}
+
+
+def get_sticker_by_name(name: str) -> Optional[dict]:
+    """
+    按名称查找贴纸，支持模糊匹配。
+
+    匹配优先级：
+      1. 完全相等（name）
+      2. name 包含查询词（前缀/子串）
+      3. description 包含查询词（同义词搜索）
+      4. 通用模糊评分（与 sticker-search 同算法），命中即返回得分最高的一条
+
+    返回 sticker dict，找不到返回 None。
+    """
+    if not name:
+        return None
+
+    query = name.strip()
+
+    if query in STICKER_MAP:
+        return STICKER_MAP[query]
+
+    for key, sticker in STICKER_MAP.items():
+        if query in key or key in query:
+            return sticker
+
+    for sticker in STICKER_MAP.values():
+        desc = sticker.get("description", "")
+        if query in desc:
+            return sticker
+
+    matches = search_stickers(query, limit=1)
+    return matches[0] if matches else None
+
+
+def get_random_sticker(category: str = None) -> dict:
+    """
+    随机返回一个贴纸。
+
+    若指定 category，则在 description 中含有该关键词的贴纸里随机选取；
+    category 为 None 时从全表随机。
+    """
+    if category:
+        candidates = [
+            s for s in STICKER_MAP.values()
+            if category in s.get("description", "") or category in s.get("name", "")
+        ]
+        if candidates:
+            return random.choice(candidates)
+    return random.choice(list(STICKER_MAP.values()))
+
+
+def get_sticker_by_id(sticker_id: str) -> Optional[dict]:
+    """按 sticker_id 精确查找贴纸。"""
+    if not sticker_id:
+        return None
+    sid = str(sticker_id).strip()
+    for sticker in STICKER_MAP.values():
+        if sticker.get("sticker_id") == sid:
+            return sticker
+    return None
+
+
+# ---------------------------------------------------------------------------
+# 模糊搜索（对齐 chatbot-web yuanbao-openclaw-plugin/sticker-cache.ts.searchStickers）
+# ---------------------------------------------------------------------------
+
+_PUNCT_RE = re.compile(r"[\s\u3000\-_·.,，。!！?？\"“”'‘’、/\\]+")
+
+
+def _normalize_text(raw: str) -> str:
+    return unicodedata.normalize("NFKC", str(raw or "")).strip().lower()
+
+
+def _compact_text(raw: str) -> str:
+    return _PUNCT_RE.sub("", _normalize_text(raw))
+
+
+def _multiset_char_hit_ratio(needle: str, haystack: str) -> float:
+    if not needle:
+        return 0.0
+    bag: dict[str, int] = {}
+    for ch in haystack:
+        bag[ch] = bag.get(ch, 0) + 1
+    hits = 0
+    for ch in needle:
+        n = bag.get(ch, 0)
+        if n > 0:
+            hits += 1
+            bag[ch] = n - 1
+    return hits / len(needle)
+
+
+def _bigram_jaccard(a: str, b: str) -> float:
+    if len(a) < 2 or len(b) < 2:
+        return 0.0
+    A = {a[i:i + 2] for i in range(len(a) - 1)}
+    B = {b[i:i + 2] for i in range(len(b) - 1)}
+    inter = len(A & B)
+    union = len(A) + len(B) - inter
+    return inter / union if union else 0.0
+
+
+def _longest_subsequence_ratio(needle: str, haystack: str) -> float:
+    if not needle:
+        return 0.0
+    j = 0
+    for ch in haystack:
+        if j >= len(needle):
+            break
+        if ch == needle[j]:
+            j += 1
+    return j / len(needle)
+
+
+def _score_field(haystack: str, query: str) -> float:
+    hay = _normalize_text(haystack)
+    q = _normalize_text(query)
+    if not hay or not q:
+        return 0.0
+    hay_c = _compact_text(haystack)
+    q_c = _compact_text(query)
+    best = 0.0
+    if hay == q:
+        best = max(best, 100.0)
+    if q in hay:
+        best = max(best, 92 + min(6, len(q)))
+    if len(q) >= 2 and hay.startswith(q):
+        best = max(best, 88.0)
+    if q_c and q_c in hay_c:
+        best = max(best, 86.0)
+    best = max(best, _multiset_char_hit_ratio(q_c, hay_c) * 62)
+    best = max(best, _bigram_jaccard(q_c, hay_c) * 58)
+    best = max(best, _longest_subsequence_ratio(q_c, hay_c) * 52)
+    if len(q) == 1 and q in hay:
+        best = max(best, 68.0)
+    return best
+
+
+def search_stickers(query: str, limit: int = 10) -> list[dict]:
+    """
+    在内置贴纸表中按模糊匹配排序返回前 N 条结果。
+
+    评分综合 name/description 字段的子串、字符多重集覆盖、bigram Jaccard、子序列比例。
+    name 权重略高于 description（×0.88）。空 query 时按字典顺序返回前 N 条。
+    """
+    safe_limit = max(1, min(500, int(limit) if limit else 10))
+    if not query or not _normalize_text(query):
+        return list(STICKER_MAP.values())[:safe_limit]
+
+    scored: list[tuple[float, dict]] = []
+    for sticker in STICKER_MAP.values():
+        name_s = _score_field(sticker.get("name", ""), query)
+        desc_s = _score_field(sticker.get("description", ""), query) * 0.88
+        sid = str(sticker.get("sticker_id", "")).strip()
+        q_norm = _normalize_text(query)
+        id_s = 0.0
+        if sid and q_norm:
+            sid_norm = _normalize_text(sid)
+            if sid_norm == q_norm:
+                id_s = 100.0
+            elif q_norm in sid_norm:
+                id_s = 84.0
+        scored.append((max(name_s, desc_s, id_s), sticker))
+
+    scored.sort(key=lambda x: x[0], reverse=True)
+    top = scored[0][0] if scored else 0
+    if top <= 0:
+        return [s for _, s in scored[:safe_limit]]
+
+    if top >= 22:
+        floor = 18.0
+    elif top >= 12:
+        floor = max(10.0, top * 0.5)
+    else:
+        floor = max(6.0, top * 0.35)
+
+    filtered = [pair for pair in scored if pair[0] >= floor]
+    out = filtered if filtered else scored
+    return [s for _, s in out[:safe_limit]]
+
+
+def build_face_msg_body(
+    face_index: int,
+    face_type: int = 1,
+    data: Optional[str] = None,
+) -> list:
+    """
+    构造 TIMFaceElem 消息体。
+
+    Yuanbao 约定：
+      - index 固定传 0（服务端通过 data 字段识别具体表情）
+      - data 为 JSON 字符串，包含 sticker_id / package_id 等字段
+
+    Args:
+        face_index: 保留字段，暂时不影响 wire format（Yuanbao 固定 index=0）。
+                    当 face_index > 0 时视为旧版 QQ 表情 ID，直接放入 index。
+        face_type:  保留字段（兼容旧接口，当前未使用）。
+        data:       已序列化的 JSON 字符串；为 None 时仅传 index。
+
+    Returns:
+        符合 Yuanbao TIM 协议的 msg_body list，如::
+
+            [{"msg_type": "TIMFaceElem", "msg_content": {"index": 0, "data": "..."}}]
+    """
+    msg_content: dict = {"index": face_index}
+    if data is not None:
+        msg_content["data"] = data
+    return [{"msg_type": "TIMFaceElem", "msg_content": msg_content}]
+
+
+def build_sticker_msg_body(sticker: dict) -> list:
+    """
+    从 STICKER_MAP 中的 sticker dict 直接构造 TIMFaceElem 消息体。
+
+    这是 send_sticker() 的内部辅助，确保 data 字段与原始 JS 插件一致。
+    """
+    data_payload = json.dumps(
+        {
+            "sticker_id": sticker["sticker_id"],
+            "package_id": sticker["package_id"],
+            "width": sticker.get("width", 128),
+            "height": sticker.get("height", 128),
+            "formats": sticker.get("formats", "png"),
+            "name": sticker["name"],
+        },
+        ensure_ascii=False,
+        separators=(",", ":"),
+    )
+    return build_face_msg_body(face_index=0, data=data_payload)

gateway/session.py

@@ -87,6 +87,9 @@ class SessionSource:
     user_id_alt: Optional[str] = None  # Platform-specific stable alt ID (Signal UUID, Feishu union_id)
     chat_id_alt: Optional[str] = None  # Signal group internal ID
     is_bot: bool = False  # True when the message author is a bot/webhook (Discord)
+    guild_id: Optional[str] = None  # Discord guild / Slack workspace / Matrix server scope
+    parent_chat_id: Optional[str] = None  # Parent channel when chat_id refers to a thread
+    message_id: Optional[str] = None  # ID of the triggering message (for pin/reply/react)
     
     @property
     def description(self) -> str:
@@ -124,8 +127,14 @@ class SessionSource:
             d["user_id_alt"] = self.user_id_alt
         if self.chat_id_alt:
             d["chat_id_alt"] = self.chat_id_alt
+        if self.guild_id:
+            d["guild_id"] = self.guild_id
+        if self.parent_chat_id:
+            d["parent_chat_id"] = self.parent_chat_id
+        if self.message_id:
+            d["message_id"] = self.message_id
         return d
-    
+
     @classmethod
     def from_dict(cls, data: Dict[str, Any]) -> "SessionSource":
         return cls(
@@ -139,6 +148,9 @@ class SessionSource:
             chat_topic=data.get("chat_topic"),
             user_id_alt=data.get("user_id_alt"),
             chat_id_alt=data.get("chat_id_alt"),
+            guild_id=data.get("guild_id"),
+            parent_chat_id=data.get("parent_chat_id"),
+            message_id=data.get("message_id"),
         )
     
 
@@ -190,6 +202,31 @@ that requires raw IDs).  Discord is excluded because mentions use ``<@user_id>``
 and the LLM needs the real ID to tag users."""
 
 
+def _discord_tools_loaded() -> bool:
+    """True iff the agent will actually have Discord tools this session.
+
+    Two conditions must hold:
+      1. The `discord` or `discord_admin` toolset is enabled for the
+         Discord platform via `hermes tools` (opt-in, default OFF).
+      2. `DISCORD_BOT_TOKEN` is set — the tool's `check_fn` gates on it
+         at registry time, so the toolset being enabled in config is not
+         enough if the token isn't configured.
+
+    Returns False (safe default — keeps the stale-API disclaimer) on any
+    error so a bad config can't silently promise tools the agent lacks.
+    """
+    if not (os.environ.get("DISCORD_BOT_TOKEN") or "").strip():
+        return False
+    try:
+        from hermes_cli.config import load_config
+        from hermes_cli.tools_config import _get_platform_tools
+        cfg = load_config()
+        enabled = _get_platform_tools(cfg, "discord", include_default_mcp_servers=False)
+        return "discord" in enabled or "discord_admin" in enabled
+    except Exception:
+        return False
+
+
 def build_session_context_prompt(
     context: SessionContext,
     *,
@@ -273,18 +310,38 @@ def build_session_context_prompt(
             "**Platform notes:** You are running inside Slack. "
             "You do NOT have access to Slack-specific APIs — you cannot search "
             "channel history, pin/unpin messages, manage channels, or list users. "
-            "Do not promise to perform these actions. If the user asks, explain "
-            "that you can only read messages sent directly to you and respond."
+            "Do not promise to perform these actions. The gateway may inline the "
+            "current message's Slack block/attachment payload when available, but "
+            "you still cannot call Slack APIs yourself."
         )
     elif context.source.platform == Platform.DISCORD:
-        lines.append("")
-        lines.append(
-            "**Platform notes:** You are running inside Discord. "
-            "You do NOT have access to Discord-specific APIs — you cannot search "
-            "channel history, pin messages, manage roles, or list server members. "
-            "Do not promise to perform these actions. If the user asks, explain "
-            "that you can only read messages sent directly to you and respond."
-        )
+        # Inject the Discord IDs block only when the agent actually has
+        # Discord tools loaded this session — i.e. the user opted into
+        # `discord` / `discord_admin` via `hermes tools` AND the bot
+        # token is configured.  Otherwise keep the stale-API disclaimer
+        # honest so we never promise tools the agent lacks.
+        if _discord_tools_loaded():
+            src = context.source
+            id_lines = ["", "**Discord IDs (for the `discord` / `discord_admin` tools):**"]
+            if src.guild_id:
+                id_lines.append(f"  - Guild: `{src.guild_id}`")
+            if src.thread_id and src.parent_chat_id:
+                id_lines.append(f"  - Parent channel: `{src.parent_chat_id}`")
+                id_lines.append(f"  - Thread: `{src.thread_id}` (use as `channel_id` for fetch_messages etc.)")
+            else:
+                id_lines.append(f"  - Channel: `{src.chat_id}`")
+            if src.message_id:
+                id_lines.append(f"  - Triggering message: `{src.message_id}`")
+            lines.extend(id_lines)
+        else:
+            lines.append("")
+            lines.append(
+                "**Platform notes:** You are running inside Discord. "
+                "You do NOT have access to Discord-specific APIs — you cannot search "
+                "channel history, pin messages, manage roles, or list server members. "
+                "Do not promise to perform these actions. If the user asks, explain "
+                "that you can only read messages sent directly to you and respond."
+            )
     elif context.source.platform == Platform.BLUEBUBBLES:
         lines.append("")
         lines.append(
@@ -297,6 +354,14 @@ def build_session_context_prompt(
             "If the user needs a detailed answer, give the short version first "
             "and offer to elaborate."
         )
+    elif context.source.platform == Platform.YUANBAO:
+        lines.append("")
+        lines.append(
+            "**Platform notes:** You are running inside Yuanbao. "
+            "You CAN send private (DM) messages via the send_message tool. "
+            "Use target='yuanbao:direct:<account_id>' for DM "
+            "and target='yuanbao:group:<group_code>' for group chat."
+        )
 
     # Connected platforms
     platforms_list = ["local (files on this machine)"]
@@ -383,11 +448,11 @@ class SessionEntry:
     auto_reset_reason: Optional[str] = None  # "idle" or "daily"
     reset_had_activity: bool = False  # whether the expired session had any messages
     
-    # Set by the background expiry watcher after it successfully flushes
-    # memories for this session.  Persisted to sessions.json so the flag
-    # survives gateway restarts (the old in-memory _pre_flushed_sessions
-    # set was lost on restart, causing redundant re-flushes).
-    memory_flushed: bool = False
+    # Set by the background expiry watcher after it finalizes an expired
+    # session (invoking on_session_finalize hooks and evicting the cached
+    # agent).  Persisted to sessions.json so the flag survives gateway
+    # restarts — prevents redundant finalization runs.
+    expiry_finalized: bool = False
 
     # When True the next call to get_or_create_session() will auto-reset
     # this session (create a new session_id) so the user starts fresh.
@@ -423,7 +488,7 @@ class SessionEntry:
             "last_prompt_tokens": self.last_prompt_tokens,
             "estimated_cost_usd": self.estimated_cost_usd,
             "cost_status": self.cost_status,
-            "memory_flushed": self.memory_flushed,
+            "expiry_finalized": self.expiry_finalized,
             "suspended": self.suspended,
             "resume_pending": self.resume_pending,
             "resume_reason": self.resume_reason,
@@ -475,7 +540,7 @@ class SessionEntry:
             last_prompt_tokens=data.get("last_prompt_tokens", 0),
             estimated_cost_usd=data.get("estimated_cost_usd", 0.0),
             cost_status=data.get("cost_status", "unknown"),
-            memory_flushed=data.get("memory_flushed", False),
+            expiry_finalized=data.get("expiry_finalized", data.get("memory_flushed", False)),
             suspended=data.get("suspended", False),
             resume_pending=data.get("resume_pending", False),
             resume_reason=data.get("resume_reason"),
@@ -1176,6 +1241,7 @@ class SessionStore:
                     reasoning_content=message.get("reasoning_content") if message.get("role") == "assistant" else None,
                     reasoning_details=message.get("reasoning_details") if message.get("role") == "assistant" else None,
                     codex_reasoning_items=message.get("codex_reasoning_items") if message.get("role") == "assistant" else None,
+                    codex_message_items=message.get("codex_message_items") if message.get("role") == "assistant" else None,
                 )
             except Exception as e:
                 logger.debug("Session DB operation failed: %s", e)
@@ -1208,6 +1274,7 @@ class SessionStore:
                         reasoning_content=msg.get("reasoning_content") if role == "assistant" else None,
                         reasoning_details=msg.get("reasoning_details") if role == "assistant" else None,
                         codex_reasoning_items=msg.get("codex_reasoning_items") if role == "assistant" else None,
+                        codex_message_items=msg.get("codex_message_items") if role == "assistant" else None,
                     )
             except Exception as e:
                 logger.debug("Failed to rewrite transcript in DB: %s", e)

gateway/stream_consumer.py

@@ -44,6 +44,14 @@ class StreamConsumerConfig:
     buffer_threshold: int = 40
     cursor: str = " ▉"
     buffer_only: bool = False
+    # When >0, the final edit for a streamed response is delivered as a
+    # fresh message if the original preview has been visible for at least
+    # this many seconds.  This makes the platform's visible timestamp
+    # reflect completion time instead of first-token time for long-running
+    # responses (e.g. reasoning models that stream slowly).  Ported from
+    # openclaw/openclaw#72038.  Default 0 = always edit in place (legacy
+    # behavior).  The gateway enables this selectively per-platform.
+    fresh_final_after_seconds: float = 0.0
 
 
 class GatewayStreamConsumer:
@@ -91,6 +99,12 @@ class GatewayStreamConsumer:
         self._queue: queue.Queue = queue.Queue()
         self._accumulated = ""
         self._message_id: Optional[str] = None
+        # Wall-clock timestamp (time.monotonic) when ``_message_id`` was
+        # first assigned from a successful first-send.  Used by the
+        # fresh-final logic to detect long-lived previews whose edit
+        # timestamps would be stale by completion time.  Ported from
+        # openclaw/openclaw#72038.
+        self._message_created_ts: Optional[float] = None
         self._already_sent = False
         self._edit_supported = True  # Disabled when progressive edits are no longer usable
         self._last_edit_time = 0.0
@@ -136,6 +150,7 @@ class GatewayStreamConsumer:
         if preserve_no_edit and self._message_id == "__no_edit__":
             return
         self._message_id = None
+        self._message_created_ts = None
         self._accumulated = ""
         self._last_sent_text = ""
         self._fallback_final_send = False
@@ -734,6 +749,81 @@ class GatewayStreamConsumer:
             logger.error("Commentary send error: %s", e)
             return False
 
+    def _should_send_fresh_final(self) -> bool:
+        """Return True when a long-lived preview should be replaced with a
+        fresh final message instead of an edit.
+
+        Conditions:
+        - Fresh-final is enabled (``fresh_final_after_seconds > 0``).
+        - We have a real preview message id (not the ``__no_edit__`` sentinel
+          and not ``None``).
+        - The preview has been visible for at least the configured threshold.
+
+        Ported from openclaw/openclaw#72038.
+        """
+        threshold = getattr(self.cfg, "fresh_final_after_seconds", 0.0) or 0.0
+        if threshold <= 0:
+            return False
+        if not self._message_id or self._message_id == "__no_edit__":
+            return False
+        if self._message_created_ts is None:
+            return False
+        age = time.monotonic() - self._message_created_ts
+        return age >= threshold
+
+    async def _try_fresh_final(self, text: str) -> bool:
+        """Send ``text`` as a brand-new message (best-effort delete the old
+        preview) so the platform's visible timestamp reflects completion
+        time.  Returns True on successful delivery, False on any failure so
+        the caller falls back to the normal edit path.
+
+        Ported from openclaw/openclaw#72038.
+        """
+        old_message_id = self._message_id
+        try:
+            result = await self.adapter.send(
+                chat_id=self.chat_id,
+                content=text,
+                metadata=self.metadata,
+            )
+        except Exception as e:
+            logger.debug("Fresh-final send failed, falling back to edit: %s", e)
+            return False
+        if not getattr(result, "success", False):
+            return False
+        # Successful fresh send — try to delete the stale preview so the
+        # user doesn't see the old edit-stuck message underneath.  Cleanup
+        # is best-effort; platforms that don't implement ``delete_message``
+        # just leave the preview behind (still an acceptable outcome —
+        # the visible final timestamp is the important part).
+        if old_message_id and old_message_id != "__no_edit__":
+            delete_fn = getattr(self.adapter, "delete_message", None)
+            if delete_fn is not None:
+                try:
+                    await delete_fn(self.chat_id, old_message_id)
+                except Exception as e:
+                    logger.debug(
+                        "Fresh-final preview cleanup failed (%s): %s",
+                        old_message_id, e,
+                    )
+        # Adopt the new message id as the current message so subsequent
+        # callers (e.g. overflow split loops, finalize retries) see a
+        # consistent state.
+        new_message_id = getattr(result, "message_id", None)
+        if new_message_id:
+            self._message_id = new_message_id
+            self._message_created_ts = time.monotonic()
+        else:
+            # Send succeeded but platform didn't return an id — treat the
+            # delivery as final-only and fall back to "__no_edit__" so we
+            # don't try to edit something we can't address.
+            self._message_id = "__no_edit__"
+            self._message_created_ts = None
+        self._already_sent = True
+        self._last_sent_text = text
+        self._final_response_sent = True
+        return True
+
     async def _send_or_edit(self, text: str, *, finalize: bool = False) -> bool:
         """Send or edit the streaming message.
 
@@ -786,6 +876,22 @@ class GatewayStreamConsumer:
                         finalize and self._adapter_requires_finalize
                     ):
                         return True
+                    # Fresh-final for long-lived previews: when finalizing
+                    # the last edit in a streaming sequence, if the
+                    # original preview has been visible for at least
+                    # ``fresh_final_after_seconds``, send the completed
+                    # reply as a fresh message so the platform's visible
+                    # timestamp reflects completion time instead of the
+                    # preview creation time.  Best-effort cleanup of the
+                    # old preview follows.  Ported from
+                    # openclaw/openclaw#72038.  Gated by config so the
+                    # legacy edit-in-place path stays the default.
+                    if (
+                        finalize
+                        and self._should_send_fresh_final()
+                        and await self._try_fresh_final(text)
+                    ):
+                        return True
                     # Edit existing message
                     result = await self.adapter.edit_message(
                         chat_id=self.chat_id,
@@ -852,6 +958,10 @@ class GatewayStreamConsumer:
                 if result.success:
                     if result.message_id:
                         self._message_id = result.message_id
+                        # Track when the preview first became visible to
+                        # the user so fresh-final logic can detect stale
+                        # preview timestamps on long-running responses.
+                        self._message_created_ts = time.monotonic()
                     else:
                         self._edit_supported = False
                     self._already_sent = True

gateway/whatsapp_identity.py

@@ -31,8 +31,17 @@ Hermes' own session keys.
 from __future__ import annotations
 
 import json
+import logging
+import re
 from typing import Set
 
+logger = logging.getLogger(__name__)
+
+# WhatsApp JIDs are numeric (or plus-prefixed numeric) with optional
+# ``@``, ``.`` and ``:`` separators. ``\w`` is pinned to ASCII so
+# full-width digits / Unicode word chars can't sneak through.
+_SAFE_IDENTIFIER_RE = re.compile(r"^[A-Za-z0-9@.+\-]+$")
+
 from hermes_constants import get_hermes_home
 
 
@@ -81,6 +90,16 @@ def expand_whatsapp_aliases(identifier: str) -> Set[str]:
         current = queue.pop(0)
         if not current or current in resolved:
             continue
+        # Defense-in-depth: reject identifiers that could sneak path
+        # separators / traversal segments into the ``lid-mapping-{current}``
+        # filename below. The hardcoded ``lid-mapping-`` prefix already
+        # prevents escape via pathlib's component split (an attacker can't
+        # create ``lid-mapping-..`` as a real directory in session_dir), but
+        # this keeps the identifier space to the characters WhatsApp JIDs
+        # actually use and avoids depending on that filesystem-layout
+        # invariant.
+        if not _SAFE_IDENTIFIER_RE.match(current):
+            continue
 
         resolved.add(current)
         for suffix in ("", "_reverse"):
@@ -91,7 +110,8 @@ def expand_whatsapp_aliases(identifier: str) -> Set[str]:
                 mapped = normalize_whatsapp_identifier(
                     json.loads(mapping_path.read_text(encoding="utf-8"))
                 )
-            except Exception:
+            except (OSError, json.JSONDecodeError) as exc:
+                logger.debug("whatsapp_identity: failed to read %s: %s", mapping_path, exc)
                 continue
             if mapped and mapped not in resolved:
                 queue.append(mapped)

hermes_cli/auth.py

@@ -356,6 +356,14 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         api_key_env_vars=(),
         base_url_env_var="BEDROCK_BASE_URL",
     ),
+    "azure-foundry": ProviderConfig(
+        id="azure-foundry",
+        name="Azure Foundry",
+        auth_type="api_key",
+        inference_base_url="",  # User-provided endpoint
+        api_key_env_vars=("AZURE_FOUNDRY_API_KEY",),
+        base_url_env_var="AZURE_FOUNDRY_BASE_URL",
+    ),
 }
 
 
@@ -459,11 +467,27 @@ def _resolve_api_key_provider_secret(
             pass
         return "", ""
 
+    from hermes_cli.config import get_env_value
     for env_var in pconfig.api_key_env_vars:
-        val = os.getenv(env_var, "").strip()
+        # Check both os.environ and ~/.hermes/.env file
+        val = (get_env_value(env_var) or "").strip()
         if has_usable_secret(val):
             return val, env_var
 
+    # Fallback: try credential pool (e.g. zai key stored via auth.json)
+    try:
+        from agent.credential_pool import load_pool
+        pool = load_pool(provider_id)
+        if pool and pool.has_credentials():
+            entry = pool.peek()
+            if entry:
+                key = getattr(entry, "access_token", "") or getattr(entry, "runtime_api_key", "")
+                key = str(key).strip()
+                if has_usable_secret(key):
+                    return key, f"credential_pool:{provider_id}"
+    except Exception:
+        pass
+
     return "", ""
 
 
@@ -4236,10 +4260,10 @@ def _login_nous(args, pconfig: ProviderConfig) -> None:
                 )
 
             from hermes_cli.models import (
-                _PROVIDER_MODELS, get_pricing_for_provider,
+                get_curated_nous_model_ids, get_pricing_for_provider,
                 check_nous_free_tier, partition_nous_models_by_tier,
             )
-            model_ids = _PROVIDER_MODELS.get("nous", [])
+            model_ids = get_curated_nous_model_ids()
 
             print()
             unavailable_models: list = []

hermes_cli/azure_detect.py

@@ -0,0 +1,300 @@
+"""Azure Foundry endpoint auto-detection.
+
+Inspect an Azure AI Foundry / Azure OpenAI endpoint to determine:
+  - API transport (OpenAI-style ``chat_completions`` vs
+    Anthropic-style ``anthropic_messages``)
+  - Available models (best effort — Azure does not expose a deployment
+    listing via the inference API key, but Azure OpenAI v1 endpoints
+    return the resource's model catalog via ``GET /models``)
+  - Context length for each discovered/entered model, via the existing
+    :func:`agent.model_metadata.get_model_context_length` resolver.
+
+Rationale:
+
+Azure has no pure-API-key deployment-listing endpoint — per Microsoft,
+deployment enumeration requires ARM management-plane auth.  Azure
+OpenAI v1 endpoints ``{resource}.openai.azure.com/openai/v1`` do return
+a ``/models`` list, but it reflects the resource's *available* models
+rather than the user's *deployed* deployment names.  In practice it is
+still a useful hint — the user picks a familiar model name and we look
+up its context length from the catalog.
+
+The detector never crashes on errors (every HTTP call is wrapped in a
+broad try/except).  Callers get a :class:`DetectionResult` with whatever
+information could be gathered, and fall back to manual entry for the
+rest.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import re
+from dataclasses import dataclass, field
+from typing import Optional
+from urllib import request as urllib_request
+from urllib.error import HTTPError, URLError
+from urllib.parse import urlparse, urlunparse
+
+logger = logging.getLogger(__name__)
+
+
+# Default Azure OpenAI ``api-version`` to probe with.  The v1 GA endpoint
+# accepts requests without ``api-version`` entirely, so this is only used
+# as a fallback for pre-v1 resources that still require it.
+_AZURE_OPENAI_PROBE_API_VERSIONS = (
+    "2025-04-01-preview",
+    "2024-10-21",  # oldest GA that supports /models
+)
+
+# Default Azure Anthropic ``api-version``.  Matches the value used by
+# ``agent/anthropic_adapter.py`` when building the Anthropic client.
+_AZURE_ANTHROPIC_API_VERSION = "2025-04-15"
+
+
+@dataclass
+class DetectionResult:
+    """Everything auto-detection could gather from a base URL + API key."""
+
+    #: Detected API transport: ``"chat_completions"``,
+    #: ``"anthropic_messages"``, or ``None`` when detection failed.
+    api_mode: Optional[str] = None
+
+    #: Deployment / model IDs returned by ``/models`` (best effort).
+    #: Empty when the endpoint doesn't expose the list with an API key.
+    models: list[str] = field(default_factory=list)
+
+    #: Lowercased host from the base URL (used for display messages).
+    hostname: str = ""
+
+    #: Human-readable reason the detector chose ``api_mode``.  Useful
+    #: for explaining auto-detection to the user in the wizard.
+    reason: str = ""
+
+    #: ``True`` when ``/models`` returned a valid OpenAI-shaped payload.
+    models_probe_ok: bool = False
+
+    #: ``True`` when the URL was determined to be an Anthropic-style
+    #: endpoint (from path suffix or live probe).
+    is_anthropic: bool = False
+
+
+def _http_get_json(url: str, api_key: str, timeout: float = 6.0) -> tuple[int, Optional[dict]]:
+    """GET a URL with ``api-key`` + ``Authorization`` headers.  Return
+    ``(status_code, parsed_json_or_None)``.  Never raises."""
+    req = urllib_request.Request(url, method="GET")
+    # Azure OpenAI uses ``api-key``.  Some Azure deployments (and
+    # Anthropic-style routes) use ``Authorization: Bearer``.  Send both
+    # so we probe once per URL rather than twice.
+    req.add_header("api-key", api_key)
+    req.add_header("Authorization", f"Bearer {api_key}")
+    req.add_header("User-Agent", "hermes-agent/azure-detect")
+    try:
+        with urllib_request.urlopen(req, timeout=timeout) as resp:
+            body = resp.read()
+            try:
+                return resp.status, json.loads(body.decode("utf-8", errors="replace"))
+            except Exception:
+                return resp.status, None
+    except HTTPError as exc:
+        return exc.code, None
+    except (URLError, TimeoutError, OSError) as exc:
+        logger.debug("azure_detect: GET %s failed: %s", url, exc)
+        return 0, None
+    except Exception as exc:  # pragma: no cover — defensive
+        logger.debug("azure_detect: GET %s unexpected error: %s", url, exc)
+        return 0, None
+
+
+def _strip_trailing_v1(url: str) -> str:
+    """Strip trailing ``/v1`` or ``/v1/`` so we can construct sub-paths."""
+    return re.sub(r"/v1/?$", "", url.rstrip("/"))
+
+
+def _looks_like_anthropic_path(url: str) -> bool:
+    """Return True when the URL's path ends in ``/anthropic`` or
+    contains a ``/anthropic/`` segment.  Used by Azure Foundry
+    resources that route Claude traffic through a dedicated path."""
+    try:
+        parsed = urlparse(url)
+        path = (parsed.path or "").lower().rstrip("/")
+        return path.endswith("/anthropic") or "/anthropic/" in path + "/"
+    except Exception:
+        return False
+
+
+def _extract_model_ids(payload: dict) -> list[str]:
+    """Extract a list of model IDs from an OpenAI-shaped ``/models``
+    response.  Returns ``[]`` on any shape mismatch."""
+    data = payload.get("data") if isinstance(payload, dict) else None
+    if not isinstance(data, list):
+        return []
+    ids: list[str] = []
+    for item in data:
+        if not isinstance(item, dict):
+            continue
+        # OpenAI shape: {"id": "gpt-5.4", "object": "model", ...}
+        mid = item.get("id") or item.get("model") or item.get("name")
+        if isinstance(mid, str) and mid:
+            ids.append(mid)
+    return ids
+
+
+def _probe_openai_models(base_url: str, api_key: str) -> tuple[bool, list[str]]:
+    """Probe ``<base>/models`` for an OpenAI-shaped response.
+
+    Returns ``(ok, models)``.  ``ok`` is True iff the endpoint accepted
+    us as an OpenAI-style caller (200 OK + OpenAI-shaped JSON body).
+    """
+    base_url = base_url.rstrip("/")
+
+    # Azure OpenAI v1: {resource}.openai.azure.com/openai/v1 — no
+    # api-version required for GA paths, so probe without first.
+    candidates = [f"{base_url}/models"]
+    # Fallback: explicit api-version for pre-v1 resources
+    for v in _AZURE_OPENAI_PROBE_API_VERSIONS:
+        candidates.append(f"{base_url}/models?api-version={v}")
+
+    for url in candidates:
+        status, body = _http_get_json(url, api_key)
+        if status == 200 and body is not None:
+            ids = _extract_model_ids(body)
+            if ids:
+                logger.info(
+                    "azure_detect: /models probe OK at %s (%d models)",
+                    url, len(ids),
+                )
+                return True, ids
+            # 200 + empty list still counts as "OpenAI shape, no models
+            # listed" — let the user proceed with manual entry.
+            if isinstance(body, dict) and "data" in body:
+                return True, []
+    return False, []
+
+
+def _probe_anthropic_messages(base_url: str, api_key: str) -> bool:
+    """Send a zero-token request to ``<base>/v1/messages`` and check
+    whether the endpoint at least *recognises* the Anthropic Messages
+    shape (any 4xx that mentions ``messages`` or ``model``, or a 400
+    ``invalid_request`` with an Anthropic error shape).  Never completes
+    a real chat.
+    """
+    base = _strip_trailing_v1(base_url)
+    url = f"{base}/v1/messages?api-version={_AZURE_ANTHROPIC_API_VERSION}"
+    payload = json.dumps({
+        "model": "probe",
+        "max_tokens": 1,
+        "messages": [{"role": "user", "content": "ping"}],
+    }).encode("utf-8")
+    req = urllib_request.Request(url, method="POST", data=payload)
+    req.add_header("api-key", api_key)
+    req.add_header("Authorization", f"Bearer {api_key}")
+    req.add_header("anthropic-version", "2023-06-01")
+    req.add_header("content-type", "application/json")
+    req.add_header("User-Agent", "hermes-agent/azure-detect")
+    try:
+        with urllib_request.urlopen(req, timeout=6.0) as resp:
+            # Should never 200 — "probe" isn't a real deployment.  But
+            # if it does, the endpoint definitely speaks Anthropic.
+            return resp.status < 500
+    except HTTPError as exc:
+        # 4xx with an Anthropic-shaped error body = Anthropic endpoint.
+        try:
+            body = exc.read().decode("utf-8", errors="replace")
+            lowered = body.lower()
+            if "anthropic" in lowered or '"type"' in lowered and '"error"' in lowered:
+                return True
+            # Pre-Azure-v1 Azure Foundry returns a plain 404 for
+            # Anthropic-style calls on non-Anthropic deployments.  A
+            # 400 "model not found" IS Anthropic though.
+            if exc.code == 400 and ("messages" in lowered or "model" in lowered):
+                return True
+            return False
+        except Exception:
+            return False
+    except (URLError, TimeoutError, OSError):
+        return False
+    except Exception:  # pragma: no cover
+        return False
+
+
+def detect(base_url: str, api_key: str) -> DetectionResult:
+    """Inspect an Azure endpoint and describe its transport + models.
+
+    Call this from the wizard before asking the user to pick an API
+    mode manually.  The caller should treat the returned
+    :class:`DetectionResult` as *advisory* — if ``api_mode`` is None,
+    fall back to asking the user.
+    """
+    result = DetectionResult()
+
+    try:
+        parsed = urlparse(base_url)
+        result.hostname = (parsed.hostname or "").lower()
+    except Exception:
+        result.hostname = ""
+
+    # 1. Path sniff.  Azure Foundry exposes Anthropic-style deployments
+    #    under a dedicated ``/anthropic`` path.
+    if _looks_like_anthropic_path(base_url):
+        result.is_anthropic = True
+        result.api_mode = "anthropic_messages"
+        result.reason = "URL path ends in /anthropic → Anthropic Messages API"
+        return result
+
+    # 2. Try the OpenAI-style /models probe.  If this works, the
+    #    endpoint definitely speaks OpenAI wire.
+    ok, models = _probe_openai_models(base_url, api_key)
+    if ok:
+        result.models_probe_ok = True
+        result.models = models
+        result.api_mode = "chat_completions"
+        result.reason = (
+            f"GET /models returned {len(models)} model(s) — OpenAI-style endpoint"
+            if models
+            else "GET /models returned an OpenAI-shaped empty list — OpenAI-style endpoint"
+        )
+        return result
+
+    # 3. Fallback: probe the Anthropic Messages shape.  Slower and more
+    #    intrusive than /models, so only run it when the OpenAI probe
+    #    failed.
+    if _probe_anthropic_messages(base_url, api_key):
+        result.is_anthropic = True
+        result.api_mode = "anthropic_messages"
+        result.reason = "Endpoint accepts Anthropic Messages shape"
+        return result
+
+    # Nothing matched.  Caller falls back to manual selection.
+    result.reason = (
+        "Could not probe endpoint (private network, missing model list, or "
+        "non-standard path) — falling back to manual API-mode selection"
+    )
+    return result
+
+
+def lookup_context_length(model: str, base_url: str, api_key: str) -> Optional[int]:
+    """Thin wrapper around :func:`agent.model_metadata.get_model_context_length`
+    that returns ``None`` when only the fallback default (128k) would
+    fire, so the wizard can distinguish "we actually know this" from
+    "we guessed."""
+    try:
+        from agent.model_metadata import (
+            DEFAULT_FALLBACK_CONTEXT,
+            get_model_context_length,
+        )
+    except Exception:
+        return None
+
+    try:
+        n = get_model_context_length(model, base_url=base_url, api_key=api_key)
+    except Exception as exc:
+        logger.debug("azure_detect: context length lookup failed: %s", exc)
+        return None
+
+    if isinstance(n, int) and n > 0 and n != DEFAULT_FALLBACK_CONTEXT:
+        return n
+    return None
+
+
+__all__ = ["DetectionResult", "detect", "lookup_context_length"]

hermes_cli/commands.py

@@ -84,9 +84,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
     CommandDef("deny", "Deny a pending dangerous command", "Session",
                gateway_only=True),
     CommandDef("background", "Run a prompt in the background", "Session",
-               aliases=("bg",), args_hint="<prompt>"),
-    CommandDef("btw", "Ephemeral side question using session context (no tools, not persisted)", "Session",
-               args_hint="<question>"),
+               aliases=("bg", "btw"), args_hint="<prompt>"),
     CommandDef("agents", "Show active agents and running tasks", "Session",
                aliases=("tasks",)),
     CommandDef("queue", "Queue a prompt for the next turn (doesn't interrupt)", "Session",
@@ -103,7 +101,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
     # Configuration
     CommandDef("config", "Show current configuration", "Configuration",
                cli_only=True),
-    CommandDef("model", "Switch model for this session", "Configuration", args_hint="[model] [--provider name] [--global]"),
+    CommandDef("model", "Switch model for this session", "Configuration",
+               aliases=("provider",), args_hint="[model] [--provider name] [--global]"),
     CommandDef("gquota", "Show Google Gemini Code Assist quota usage", "Info",
                cli_only=True),
 
@@ -127,8 +126,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
     CommandDef("voice", "Toggle voice mode", "Configuration",
                args_hint="[on|off|tts|status]", subcommands=("on", "off", "tts", "status")),
     CommandDef("busy", "Control what Enter does while Hermes is working", "Configuration",
-               cli_only=True, args_hint="[queue|interrupt|status]",
-               subcommands=("queue", "interrupt", "status")),
+               cli_only=True, args_hint="[queue|steer|interrupt|status]",
+               subcommands=("queue", "steer", "interrupt", "status")),
 
     # Tools & Skills
     CommandDef("tools", "Manage tools: /tools [list|disable|enable] [name...]", "Tools & Skills",
@@ -807,6 +806,114 @@ def discord_skill_commands_by_category(
     return trimmed_categories, uncategorized, hidden
 
 
+# ---------------------------------------------------------------------------
+# Slack native slash commands
+# ---------------------------------------------------------------------------
+
+# Slack slash command name constraints: lowercase a-z, 0-9, hyphens,
+# underscores. Max 32 chars. Slack app manifest accepts up to 50 slash
+# commands per app.
+_SLACK_MAX_SLASH_COMMANDS = 50
+_SLACK_NAME_LIMIT = 32
+_SLACK_INVALID_CHARS = re.compile(r"[^a-z0-9_\-]")
+
+
+def _sanitize_slack_name(raw: str) -> str:
+    """Convert a command name to a valid Slack slash command name.
+
+    Slack allows lowercase a-z, digits, hyphens, and underscores. Max 32
+    chars. Uppercase is lowercased; invalid chars are stripped.
+    """
+    name = raw.lower()
+    name = _SLACK_INVALID_CHARS.sub("", name)
+    name = name.strip("-_")
+    return name[:_SLACK_NAME_LIMIT]
+
+
+def slack_native_slashes() -> list[tuple[str, str, str]]:
+    """Return (slash_name, description, usage_hint) triples for Slack.
+
+    Every gateway-available command in ``COMMAND_REGISTRY`` is surfaced as
+    a standalone Slack slash command (e.g. ``/btw``, ``/stop``, ``/model``),
+    matching Discord's and Telegram's model where every command is a
+    first-class slash and not a ``/hermes <verb>`` subcommand.
+
+    Both canonical names and aliases are included so users can type any
+    documented form (e.g. ``/background``, ``/bg``, and ``/btw`` all work).
+    Plugin-registered slash commands are included too.
+
+    Results are clamped to Slack's 50-command limit with duplicate-name
+    avoidance. ``/hermes`` is always reserved as the first entry so the
+    legacy ``/hermes <subcommand>`` form keeps working for anything that
+    gets dropped by the clamp or for free-form questions.
+    """
+    overrides = _resolve_config_gates()
+    entries: list[tuple[str, str, str]] = []
+    seen: set[str] = set()
+
+    # Reserve /hermes as the catch-all top-level command.
+    entries.append(("hermes", "Talk to Hermes or run a subcommand", "[subcommand] [args]"))
+    seen.add("hermes")
+
+    def _add(name: str, desc: str, hint: str) -> None:
+        slack_name = _sanitize_slack_name(name)
+        if not slack_name or slack_name in seen:
+            return
+        if len(entries) >= _SLACK_MAX_SLASH_COMMANDS:
+            return
+        # Slack description cap is 2000 chars; keep it short.
+        entries.append((slack_name, desc[:140], hint[:100]))
+        seen.add(slack_name)
+
+    # First pass: canonical names (so they win slots if we hit the cap).
+    for cmd in COMMAND_REGISTRY:
+        if not _is_gateway_available(cmd, overrides):
+            continue
+        _add(cmd.name, cmd.description, cmd.args_hint or "")
+
+    # Second pass: aliases.
+    for cmd in COMMAND_REGISTRY:
+        if not _is_gateway_available(cmd, overrides):
+            continue
+        for alias in cmd.aliases:
+            # Skip aliases that only differ from canonical by case/punctuation
+            # normalization (already covered by _add dedup).
+            _add(alias, f"Alias for /{cmd.name} — {cmd.description}", cmd.args_hint or "")
+
+    # Third pass: plugin commands.
+    for name, description, args_hint in _iter_plugin_command_entries():
+        _add(name, description, args_hint or "")
+
+    return entries
+
+
+def slack_app_manifest(request_url: str = "https://hermes-agent.local/slack/commands") -> dict[str, Any]:
+    """Generate a Slack app manifest with all gateway commands as slashes.
+
+    ``request_url`` is required by Slack's manifest schema for every slash
+    command, but in Socket Mode (which we use) Slack ignores it and routes
+    the command event through the WebSocket. A placeholder URL is fine.
+
+    The returned dict is the ``features.slash_commands`` portion only —
+    callers compose it into a full manifest (or merge into an existing
+    one). Keeping it narrow avoids coupling us to the rest of the manifest
+    schema (display_information, oauth_config, settings, etc.) which users
+    set up once in the Slack UI and rarely change.
+    """
+    slashes = []
+    for name, desc, usage in slack_native_slashes():
+        entry = {
+            "command": f"/{name}",
+            "description": desc or f"Run /{name}",
+            "should_escape": False,
+            "url": request_url,
+        }
+        if usage:
+            entry["usage_hint"] = usage
+        slashes.append(entry)
+    return {"features": {"slash_commands": slashes}}
+
+
 def slack_subcommand_map() -> dict[str, str]:
     """Return subcommand -> /command mapping for Slack /hermes handler.
 

hermes_cli/config.py

@@ -465,6 +465,7 @@ DEFAULT_CONFIG = {
         "command_timeout": 30,  # Timeout for browser commands in seconds (screenshot, navigate, etc.)
         "record_sessions": False,  # Auto-record browser sessions as WebM videos
         "allow_private_urls": False,  # Allow navigating to private/internal IPs (localhost, 192.168.x.x, etc.)
+        "auto_local_for_private_urls": True,  # When a cloud provider is set, auto-spawn local Chromium for LAN/localhost URLs instead of sending them to the cloud
         "cdp_url": "",  # Optional persistent CDP endpoint for attaching to an existing Chromium/Chrome
         # CDP supervisor — dialog + frame detection via a persistent WebSocket.
         # Active only when a CDP-capable backend is attached (Browserbase or
@@ -486,6 +487,19 @@ DEFAULT_CONFIG = {
     "checkpoints": {
         "enabled": True,
         "max_snapshots": 50,  # Max checkpoints to keep per directory
+        # Auto-maintenance: shadow repos accumulate forever under
+        # ~/.hermes/checkpoints/ (one per cd'd working directory). Field
+        # reports put the typical offender at 1000+ repos / ~12 GB. When
+        # auto_prune is on, hermes sweeps at startup (at most once per
+        # min_interval_hours) and deletes:
+        #   * orphan repos: HERMES_WORKDIR no longer exists on disk
+        #   * stale repos:  newest mtime older than retention_days
+        # Opt-in so users who rely on /rollback against long-ago sessions
+        # never lose data silently.
+        "auto_prune": False,
+        "retention_days": 7,
+        "delete_orphans": True,
+        "min_interval_hours": 24,
     },
 
     # Maximum characters returned by a single read_file call.  Reads that
@@ -612,14 +626,6 @@ DEFAULT_CONFIG = {
             "timeout": 30,
             "extra_body": {},
         },
-        "flush_memories": {
-            "provider": "auto",
-            "model": "",
-            "base_url": "",
-            "api_key": "",
-            "timeout": 30,
-            "extra_body": {},
-        },
         "title_generation": {
             "provider": "auto",
             "model": "",
@@ -634,7 +640,7 @@ DEFAULT_CONFIG = {
         "compact": False,
         "personality": "kawaii",
         "resume_display": "full",
-        "busy_input_mode": "interrupt",
+        "busy_input_mode": "interrupt",  # interrupt | queue | steer
         "bell_on_complete": False,
         "show_reasoning": False,
         "streaming": False,
@@ -848,7 +854,7 @@ DEFAULT_CONFIG = {
         "auto_thread": True,           # Auto-create threads on @mention in channels (like Slack)
         "reactions": True,             # Add 👀/✅/❌ reactions to messages during processing
         "channel_prompts": {},         # Per-channel ephemeral system prompts (forum parents apply to child threads)
-        # discord_server tool: restrict which actions the agent may call.
+        # discord / discord_admin tools: restrict which actions the agent may call.
         # Default (empty) = all actions allowed (subject to bot privileged intents).
         # Accepts comma-separated string ("list_guilds,list_channels,fetch_messages")
         # or YAML list. Unknown names are dropped with a warning at load time.
@@ -967,6 +973,27 @@ DEFAULT_CONFIG = {
         "backup_count": 3,     # Number of rotated backup files to keep
     },
 
+    # Remotely-hosted model catalog manifest.  When enabled, the CLI fetches
+    # curated model lists for OpenRouter and Nous Portal from this URL,
+    # falling back to the in-repo snapshot on network failure.  Lets us
+    # update model picker lists without shipping a hermes-agent release.
+    # The default URL is served by the docs site GitHub Pages deploy.
+    "model_catalog": {
+        "enabled": True,
+        "url": "https://hermes-agent.nousresearch.com/docs/api/model-catalog.json",
+        # Disk cache TTL in hours.  Beyond this, the CLI refetches on the
+        # next /model or `hermes model` invocation; network failures
+        # silently fall back to the stale cache.
+        "ttl_hours": 24,
+        # Optional per-provider override URLs for third parties that want
+        # to self-host their own curation list using the same schema.
+        # Example:
+        #   providers:
+        #     openrouter:
+        #       url: https://example.com/my-curation.json
+        "providers": {},
+    },
+
     # Network settings — workarounds for connectivity issues.
     "network": {
         # Force IPv4 connections.  On servers with broken or unreachable IPv6,
@@ -1003,6 +1030,13 @@ DEFAULT_CONFIG = {
         "min_interval_hours": 24,
     },
 
+    # Contextual first-touch onboarding hints (see agent/onboarding.py).
+    # Each hint is shown once per install and then latched here so it
+    # never fires again.  Users can wipe the section to re-see all hints.
+    "onboarding": {
+        "seen": {},
+    },
+
     # Config schema version - bump this when adding new required fields
     "_config_version": 22,
 }
@@ -1379,6 +1413,21 @@ OPTIONAL_ENV_VARS = {
         "category": "provider",
         "advanced": True,
     },
+    "AZURE_FOUNDRY_API_KEY": {
+        "description": "Azure Foundry API key for custom Azure endpoints",
+        "prompt": "Azure Foundry API Key",
+        "url": "https://ai.azure.com/",
+        "password": True,
+        "category": "provider",
+    },
+    "AZURE_FOUNDRY_BASE_URL": {
+        "description": "Azure Foundry base URL (set via 'hermes model' for endpoint-specific config)",
+        "prompt": "Azure Foundry base URL",
+        "url": None,
+        "password": False,
+        "category": "provider",
+        "advanced": True,
+    },
 
     # ── Tool API keys ──
     "EXA_API_KEY": {
@@ -1546,6 +1595,44 @@ OPTIONAL_ENV_VARS = {
         "category": "tool",
     },
 
+    # ── Bundled skills (opt-in: only needed if the user uses that skill) ──
+    # These use category="skill" (distinct from "tool") so the sandbox
+    # env blocklist in tools/environments/local.py does NOT rewrite them —
+    # skills legitimately need these passed through to curl via
+    # tools/env_passthrough.py when the user's skill calls out.
+    "NOTION_API_KEY": {
+        "description": "Notion integration token (used by the `notion` skill)",
+        "prompt": "Notion API key",
+        "url": "https://www.notion.so/my-integrations",
+        "password": True,
+        "category": "skill",
+        "advanced": True,
+    },
+    "LINEAR_API_KEY": {
+        "description": "Linear personal API key (used by the `linear` skill)",
+        "prompt": "Linear API key",
+        "url": "https://linear.app/settings/api",
+        "password": True,
+        "category": "skill",
+        "advanced": True,
+    },
+    "AIRTABLE_API_KEY": {
+        "description": "Airtable personal access token (used by the `airtable` skill)",
+        "prompt": "Airtable API key",
+        "url": "https://airtable.com/create/tokens",
+        "password": True,
+        "category": "skill",
+        "advanced": True,
+    },
+    "TENOR_API_KEY": {
+        "description": "Tenor API key for GIF search (used by the `gif-search` skill)",
+        "prompt": "Tenor API key",
+        "url": "https://developers.google.com/tenor/guides/quickstart",
+        "password": True,
+        "category": "skill",
+        "advanced": True,
+    },
+
     # ── Honcho ──
     "HONCHO_API_KEY": {
         "description": "Honcho API key for AI-native persistent memory",
@@ -2214,6 +2301,71 @@ def get_compatible_custom_providers(
     return compatible
 
 
+def get_custom_provider_context_length(
+    model: str,
+    base_url: str,
+    custom_providers: Optional[List[Dict[str, Any]]] = None,
+    config: Optional[Dict[str, Any]] = None,
+) -> Optional[int]:
+    """Look up a per-model ``context_length`` override from ``custom_providers``.
+
+    Matches any entry whose ``base_url`` equals ``base_url`` (trailing-slash
+    insensitive) and returns ``custom_providers[i].models.<model>.context_length``
+    if present and valid.  Returns ``None`` when no override applies.
+
+    This is the single source of truth for custom-provider context overrides,
+    used by:
+      * ``AIAgent.__init__`` (startup resolution)
+      * ``AIAgent.switch_model`` (mid-session ``/model`` switch)
+      * ``hermes_cli.model_switch.resolve_display_context_length`` (``/model`` confirmation display)
+      * ``gateway.run._format_session_info`` (``/info`` display)
+      * ``agent.model_metadata.get_model_context_length`` (when custom_providers is threaded through)
+
+    Before this helper existed, the lookup was duplicated in ``run_agent.py``'s
+    startup path only; every other path (notably ``/model`` switch) fell back
+    to the 128K default.  See #15779.
+    """
+    if not model or not base_url:
+        return None
+    if custom_providers is None:
+        try:
+            custom_providers = get_compatible_custom_providers(config)
+        except Exception:
+            if config is None:
+                return None
+            raw = config.get("custom_providers")
+            custom_providers = raw if isinstance(raw, list) else []
+    if not isinstance(custom_providers, list):
+        return None
+
+    target_url = (base_url or "").rstrip("/")
+    if not target_url:
+        return None
+
+    for entry in custom_providers:
+        if not isinstance(entry, dict):
+            continue
+        entry_url = (entry.get("base_url") or "").rstrip("/")
+        if not entry_url or entry_url != target_url:
+            continue
+        models = entry.get("models")
+        if not isinstance(models, dict):
+            continue
+        model_cfg = models.get(model)
+        if not isinstance(model_cfg, dict):
+            continue
+        raw_ctx = model_cfg.get("context_length")
+        if raw_ctx is None:
+            continue
+        try:
+            ctx = int(raw_ctx)
+        except (TypeError, ValueError):
+            continue
+        if ctx > 0:
+            return ctx
+    return None
+
+
 def check_config_version() -> Tuple[int, int]:
     """
     Check config version.

hermes_cli/doctor.py

@@ -320,7 +320,11 @@ def run_doctor(args):
                     known_providers.add("custom:" + name.lower().replace(" ", "-"))
 
             canonical_provider = provider
-            if provider and _resolve_provider_full is not None and provider != "auto":
+            if (
+                provider
+                and _resolve_provider_full is not None
+                and provider not in ("auto", "custom")
+            ):
                 provider_def = _resolve_provider_full(provider, user_providers, custom_providers)
                 canonical_provider = provider_def.id if provider_def is not None else None
 

hermes_cli/fallback_cmd.py

@@ -0,0 +1,361 @@
+"""
+hermes fallback — manage the fallback provider chain.
+
+Fallback providers are tried in order when the primary model fails with
+rate-limit, overload, or connection errors. See:
+https://hermes-agent.nousresearch.com/docs/user-guide/features/fallback-providers
+
+Subcommands:
+  hermes fallback [list]   Show the current fallback chain (default when no subcommand)
+  hermes fallback add      Pick provider + model via the same picker as `hermes model`,
+                           then append the selection to the chain
+  hermes fallback remove   Pick an entry to delete from the chain
+  hermes fallback clear    Remove all fallback entries
+
+Storage: ``fallback_providers`` in ``~/.hermes/config.yaml`` (top-level, list of
+``{provider, model, base_url?, api_mode?}`` dicts).  The legacy single-dict
+``fallback_model`` format is migrated to the new list format on first add.
+"""
+from __future__ import annotations
+
+import copy
+from typing import Any, Dict, List, Optional
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _read_chain(config: Dict[str, Any]) -> List[Dict[str, Any]]:
+    """Return the normalized fallback chain as a list of dicts.
+
+    Accepts both the new list format (``fallback_providers``) and the legacy
+    single-dict format (``fallback_model``).  The returned list is always a
+    fresh copy — callers can mutate without touching the config dict.
+    """
+    chain = config.get("fallback_providers") or []
+    if isinstance(chain, list):
+        result = [dict(e) for e in chain if isinstance(e, dict) and e.get("provider") and e.get("model")]
+        if result:
+            return result
+    legacy = config.get("fallback_model")
+    if isinstance(legacy, dict) and legacy.get("provider") and legacy.get("model"):
+        return [dict(legacy)]
+    if isinstance(legacy, list):
+        return [dict(e) for e in legacy if isinstance(e, dict) and e.get("provider") and e.get("model")]
+    return []
+
+
+def _write_chain(config: Dict[str, Any], chain: List[Dict[str, Any]]) -> None:
+    """Persist the chain to ``fallback_providers`` and clear legacy key."""
+    config["fallback_providers"] = chain
+    # Drop the legacy single-dict key on write so there's only one source of truth.
+    if "fallback_model" in config:
+        config.pop("fallback_model", None)
+
+
+def _format_entry(entry: Dict[str, Any]) -> str:
+    """One-line human-readable rendering of a fallback entry."""
+    provider = entry.get("provider", "?")
+    model = entry.get("model", "?")
+    base = entry.get("base_url")
+    suffix = f"  [{base}]" if base else ""
+    return f"{model}  (via {provider}){suffix}"
+
+
+def _extract_fallback_from_model_cfg(model_cfg: Any) -> Optional[Dict[str, Any]]:
+    """Pull the ``{provider, model, base_url?, api_mode?}`` dict from a ``config["model"]`` snapshot."""
+    if not isinstance(model_cfg, dict):
+        return None
+    provider = (model_cfg.get("provider") or "").strip()
+    # The picker writes the selected model to ``model.default``.
+    model = (model_cfg.get("default") or model_cfg.get("model") or "").strip()
+    if not provider or not model:
+        return None
+    entry: Dict[str, Any] = {"provider": provider, "model": model}
+    base_url = (model_cfg.get("base_url") or "").strip()
+    if base_url:
+        entry["base_url"] = base_url
+    api_mode = (model_cfg.get("api_mode") or "").strip()
+    if api_mode:
+        entry["api_mode"] = api_mode
+    return entry
+
+
+def _snapshot_auth_active_provider() -> Any:
+    """Return the current ``active_provider`` in auth.json, or a sentinel if unavailable."""
+    try:
+        from hermes_cli.auth import _load_auth_store
+        store = _load_auth_store()
+        return store.get("active_provider")
+    except Exception:
+        return None
+
+
+def _restore_auth_active_provider(value: Any) -> None:
+    """Write back a previously snapshotted ``active_provider`` value."""
+    try:
+        from hermes_cli.auth import _auth_store_lock, _load_auth_store, _save_auth_store
+        with _auth_store_lock():
+            store = _load_auth_store()
+            store["active_provider"] = value
+            _save_auth_store(store)
+    except Exception:
+        # Best-effort — if auth.json can't be restored, the user's primary
+        # provider may have been deactivated by the picker.  They can re-run
+        # `hermes model` to fix it.  Don't fail the fallback add.
+        pass
+
+
+# ---------------------------------------------------------------------------
+# Subcommand handlers
+# ---------------------------------------------------------------------------
+
+def cmd_fallback_list(args) -> None:  # noqa: ARG001
+    """Print the current fallback chain."""
+    from hermes_cli.config import load_config
+
+    config = load_config()
+    chain = _read_chain(config)
+
+    print()
+    if not chain:
+        print("  No fallback providers configured.")
+        print()
+        print("  Add one with:  hermes fallback add")
+        print()
+        return
+
+    primary = _describe_primary(config)
+    if primary:
+        print(f"  Primary:   {primary}")
+        print()
+    print(f"  Fallback chain ({len(chain)} {'entry' if len(chain) == 1 else 'entries'}):")
+    for i, entry in enumerate(chain, 1):
+        print(f"    {i}. {_format_entry(entry)}")
+    print()
+    print("  Tried in order when the primary fails (rate-limit, 5xx, connection errors).")
+    print("  Docs: https://hermes-agent.nousresearch.com/docs/user-guide/features/fallback-providers")
+    print()
+
+
+def _describe_primary(config: Dict[str, Any]) -> Optional[str]:
+    """One-line description of the primary model for display purposes."""
+    model_cfg = config.get("model")
+    if isinstance(model_cfg, dict):
+        provider = (model_cfg.get("provider") or "?").strip() or "?"
+        model = (model_cfg.get("default") or model_cfg.get("model") or "?").strip() or "?"
+        return f"{model}  (via {provider})"
+    if isinstance(model_cfg, str) and model_cfg.strip():
+        return model_cfg.strip()
+    return None
+
+
+def cmd_fallback_add(args) -> None:
+    """Launch the same picker as `hermes model`, then append the selection to the chain."""
+    from hermes_cli.main import _require_tty, select_provider_and_model
+    from hermes_cli.config import load_config, save_config
+
+    _require_tty("fallback add")
+
+    # Snapshot BEFORE the picker runs so we can distinguish "user actually
+    # picked something" from "user cancelled" by comparing before/after.
+    before_cfg = load_config()
+    model_before = copy.deepcopy(before_cfg.get("model"))
+    active_provider_before = _snapshot_auth_active_provider()
+
+    print()
+    print("  Adding a fallback provider.  The picker below is the same one used by")
+    print("  `hermes model` — select the provider + model you want as a fallback.")
+    print()
+
+    try:
+        select_provider_and_model(args=args)
+    except SystemExit:
+        # Some provider flows exit on auth failure — restore state and re-raise.
+        _restore_model_cfg(model_before)
+        _restore_auth_active_provider(active_provider_before)
+        raise
+
+    # Read the post-picker state to see what the user selected.
+    after_cfg = load_config()
+    model_after = after_cfg.get("model")
+
+    new_entry = _extract_fallback_from_model_cfg(model_after)
+    if not new_entry:
+        # Picker didn't complete (user cancelled or flow bailed).  Nothing to do.
+        _restore_model_cfg(model_before)
+        _restore_auth_active_provider(active_provider_before)
+        print()
+        print("  No fallback added.")
+        return
+
+    # Picker picked the same thing that's already the primary → nothing changed,
+    # and there's nothing useful to add as a fallback to itself.
+    primary_entry = _extract_fallback_from_model_cfg(model_before)
+    if primary_entry and primary_entry["provider"] == new_entry["provider"] \
+            and primary_entry["model"] == new_entry["model"]:
+        _restore_model_cfg(model_before)
+        _restore_auth_active_provider(active_provider_before)
+        print()
+        print(f"  Selected model matches the current primary ({_format_entry(new_entry)}).")
+        print("  A provider cannot be a fallback for itself — no change.")
+        return
+
+    # Reload the config with the primary restored, then append the new entry
+    # to ``fallback_providers``.  We deliberately re-load (rather than mutating
+    # ``after_cfg``) because the picker may have touched other top-level keys
+    # (custom_providers, providers credentials) that we want to keep.
+    _restore_model_cfg(model_before)
+    _restore_auth_active_provider(active_provider_before)
+
+    final_cfg = load_config()
+    chain = _read_chain(final_cfg)
+
+    # Reject exact-duplicate fallback entries.
+    for existing in chain:
+        if existing.get("provider") == new_entry["provider"] \
+                and existing.get("model") == new_entry["model"]:
+            print()
+            print(f"  {_format_entry(new_entry)} is already in the fallback chain — skipped.")
+            return
+
+    chain.append(new_entry)
+    _write_chain(final_cfg, chain)
+    save_config(final_cfg)
+
+    print()
+    print(f"  Added fallback: {_format_entry(new_entry)}")
+    print(f"  Chain is now {len(chain)} {'entry' if len(chain) == 1 else 'entries'} long.")
+    print()
+    print("  Run `hermes fallback list` to view, or `hermes fallback remove` to delete.")
+
+
+def _restore_model_cfg(model_before: Any) -> None:
+    """Restore ``config["model"]`` to a previously-captured snapshot."""
+    from hermes_cli.config import load_config, save_config
+
+    cfg = load_config()
+    if model_before is None:
+        cfg.pop("model", None)
+    else:
+        cfg["model"] = copy.deepcopy(model_before)
+    save_config(cfg)
+
+
+def cmd_fallback_remove(args) -> None:  # noqa: ARG001
+    """Pick an entry from the chain and remove it."""
+    from hermes_cli.config import load_config, save_config
+
+    config = load_config()
+    chain = _read_chain(config)
+
+    if not chain:
+        print()
+        print("  No fallback providers configured — nothing to remove.")
+        print()
+        return
+
+    choices = [_format_entry(e) for e in chain]
+    choices.append("Cancel")
+
+    try:
+        from hermes_cli.setup import _curses_prompt_choice
+        idx = _curses_prompt_choice("Select a fallback to remove:", choices, 0)
+    except Exception:
+        idx = _numbered_pick("Select a fallback to remove:", choices)
+
+    if idx is None or idx < 0 or idx >= len(chain):
+        print()
+        print("  Cancelled — no change.")
+        return
+
+    removed = chain.pop(idx)
+    _write_chain(config, chain)
+    save_config(config)
+
+    print()
+    print(f"  Removed fallback: {_format_entry(removed)}")
+    if chain:
+        print(f"  Chain is now {len(chain)} {'entry' if len(chain) == 1 else 'entries'} long.")
+    else:
+        print("  Fallback chain is now empty.")
+    print()
+
+
+def cmd_fallback_clear(args) -> None:  # noqa: ARG001
+    """Remove all fallback entries (with confirmation)."""
+    from hermes_cli.config import load_config, save_config
+
+    config = load_config()
+    chain = _read_chain(config)
+
+    if not chain:
+        print()
+        print("  No fallback providers configured — nothing to clear.")
+        print()
+        return
+
+    print()
+    print(f"  Current fallback chain ({len(chain)} {'entry' if len(chain) == 1 else 'entries'}):")
+    for i, entry in enumerate(chain, 1):
+        print(f"    {i}. {_format_entry(entry)}")
+    print()
+    try:
+        resp = input("  Clear all entries? [y/N]: ").strip().lower()
+    except (KeyboardInterrupt, EOFError):
+        print()
+        print("  Cancelled.")
+        return
+    if resp not in ("y", "yes"):
+        print("  Cancelled — no change.")
+        return
+
+    _write_chain(config, [])
+    save_config(config)
+    print()
+    print("  Fallback chain cleared.")
+    print()
+
+
+def _numbered_pick(question: str, choices: List[str]) -> Optional[int]:
+    """Fallback numbered-list picker when curses is unavailable."""
+    print(question)
+    for i, c in enumerate(choices, 1):
+        print(f"  {i}. {c}")
+    print()
+    while True:
+        try:
+            val = input(f"Choice [1-{len(choices)}]: ").strip()
+            if not val:
+                return None
+            idx = int(val) - 1
+            if 0 <= idx < len(choices):
+                return idx
+            print(f"Please enter 1-{len(choices)}")
+        except ValueError:
+            print("Please enter a number")
+        except (KeyboardInterrupt, EOFError):
+            print()
+            return None
+
+
+# ---------------------------------------------------------------------------
+# Dispatch
+# ---------------------------------------------------------------------------
+
+def cmd_fallback(args) -> None:
+    """Top-level dispatcher for ``hermes fallback [subcommand]``."""
+    sub = getattr(args, "fallback_command", None)
+    if sub in (None, "", "list", "ls"):
+        cmd_fallback_list(args)
+    elif sub == "add":
+        cmd_fallback_add(args)
+    elif sub in ("remove", "rm"):
+        cmd_fallback_remove(args)
+    elif sub == "clear":
+        cmd_fallback_clear(args)
+    else:
+        print(f"Unknown fallback subcommand: {sub}")
+        print("Use one of: list, add, remove, clear")
+        raise SystemExit(2)

hermes_cli/gateway.py

@@ -2724,6 +2724,24 @@ _PLATFORMS = [
              "help": "OpenID to deliver cron results and notifications to."},
         ],
     },
+    {
+        "key": "yuanbao",
+        "label": "Yuanbao",
+        "emoji": "💎",
+        "token_var": "YUANBAO_APP_ID",
+        "setup_instructions": [
+            "1. Download the Yuanbao app from https://yuanbao.tencent.com/",
+            "2. In the app, go to PAI → My Bot and create a new bot",
+            "3. After the bot is created, copy the App ID and App Secret",
+            "4. Enter them below and Hermes will connect automatically over WebSocket",
+        ],
+        "vars": [
+            {"name": "YUANBAO_APP_ID", "prompt": "App ID", "password": False,
+             "help": "The App ID from your Yuanbao IM Bot credentials."},
+            {"name": "YUANBAO_APP_SECRET", "prompt": "App Secret", "password": True,
+             "help": "The App Secret (used for HMAC signing) from your Yuanbao IM Bot."},
+        ],
+    },
 ]
 
 
@@ -3108,6 +3126,12 @@ def _setup_wecom():
     print_success("💬 WeCom configured!")
 
 
+def _setup_yuanbao():
+    """Configure Yuanbao via the standard platform setup."""
+    yuanbao_platform = next(p for p in _PLATFORMS if p["key"] == "yuanbao")
+    _setup_standard_platform(yuanbao_platform)
+
+
 def _is_service_installed() -> bool:
     """Check if the gateway is installed as a system service."""
     if supports_systemd_services():

hermes_cli/hooks.py

@@ -125,6 +125,7 @@ _DEFAULT_PAYLOADS = {
         "task_id": "test-task",
         "tool_call_id": "test-call",
         "result": '{"output": "hello"}',
+        "duration_ms": 42,
     },
     "pre_llm_call": {
         "session_id": "test-session",

hermes_cli/model_catalog.py

@@ -0,0 +1,329 @@
+"""Remote model catalog fetcher.
+
+The Hermes docs site hosts a JSON manifest of curated models for providers
+we want to update without shipping a release (currently OpenRouter and
+Nous Portal). This module fetches, validates, and caches that manifest,
+falling back to the in-repo hardcoded lists when the network is unavailable.
+
+Pipeline
+--------
+1. ``get_catalog()`` — returns a parsed manifest dict.
+   - Checks in-process cache (invalidated by TTL).
+   - Reads disk cache at ``~/.hermes/cache/model_catalog.json``.
+   - Fetches the master URL if disk cache is stale or missing.
+   - On any fetch failure, keeps using the stale cache (or empty dict).
+
+2. ``get_curated_openrouter_models()`` / ``get_curated_nous_models()`` —
+   thin accessors returning the shapes existing callers expect. Each
+   falls back to the in-repo hardcoded list on any lookup failure.
+
+Schema (version 1)
+------------------
+::
+
+    {
+      "version": 1,
+      "updated_at": "2026-04-25T22:00:00Z",
+      "metadata": {...},                # free-form
+      "providers": {
+        "openrouter": {
+          "metadata": {...},            # free-form
+          "models": [
+            {"id": "vendor/model", "description": "recommended",
+             "metadata": {...}}          # free-form, model-level
+          ]
+        },
+        "nous": {...}
+      }
+    }
+
+Unknown fields are ignored — extra metadata can be added at either level
+without bumping ``version``. ``version`` bumps are reserved for
+breaking changes (renaming ``providers``, changing ``models`` shape).
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import time
+import urllib.error
+import urllib.request
+from pathlib import Path
+from typing import Any
+
+from hermes_cli import __version__ as _HERMES_VERSION
+
+logger = logging.getLogger(__name__)
+
+# ---------------------------------------------------------------------------
+# Constants
+# ---------------------------------------------------------------------------
+
+DEFAULT_CATALOG_URL = (
+    "https://hermes-agent.nousresearch.com/docs/api/model-catalog.json"
+)
+DEFAULT_TTL_HOURS = 24
+DEFAULT_FETCH_TIMEOUT = 8.0
+SUPPORTED_SCHEMA_VERSION = 1
+
+_HERMES_USER_AGENT = f"hermes-cli/{_HERMES_VERSION}"
+
+# In-process cache to avoid repeated disk + parse work across multiple
+# calls within the same session. Invalidated by TTL against the disk file's
+# mtime, so calling code never has to think about this.
+_catalog_cache: dict[str, Any] | None = None
+_catalog_cache_source_mtime: float = 0.0
+
+
+# ---------------------------------------------------------------------------
+# Config
+# ---------------------------------------------------------------------------
+
+
+def _load_catalog_config() -> dict[str, Any]:
+    """Load the ``model_catalog`` config block with defaults filled in."""
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config() or {}
+    except Exception:
+        cfg = {}
+
+    raw = cfg.get("model_catalog")
+    if not isinstance(raw, dict):
+        raw = {}
+
+    return {
+        "enabled": bool(raw.get("enabled", True)),
+        "url": str(raw.get("url") or DEFAULT_CATALOG_URL),
+        "ttl_hours": float(raw.get("ttl_hours") or DEFAULT_TTL_HOURS),
+        "providers": raw.get("providers") if isinstance(raw.get("providers"), dict) else {},
+    }
+
+
+def _cache_path() -> Path:
+    """Return the disk cache path. Import lazily so tests can monkeypatch home."""
+    from hermes_constants import get_hermes_home
+    return get_hermes_home() / "cache" / "model_catalog.json"
+
+
+# ---------------------------------------------------------------------------
+# Fetch + validate + cache
+# ---------------------------------------------------------------------------
+
+
+def _fetch_manifest(url: str, timeout: float) -> dict[str, Any] | None:
+    """HTTP GET the manifest URL and return a parsed dict, or None on failure."""
+    try:
+        req = urllib.request.Request(
+            url,
+            headers={
+                "Accept": "application/json",
+                "User-Agent": _HERMES_USER_AGENT,
+            },
+        )
+        with urllib.request.urlopen(req, timeout=timeout) as resp:
+            data = json.loads(resp.read().decode())
+    except (urllib.error.URLError, TimeoutError, json.JSONDecodeError, OSError) as exc:
+        logger.info("model catalog fetch failed (%s): %s", url, exc)
+        return None
+    except Exception as exc:  # pragma: no cover — defensive
+        logger.info("model catalog fetch errored (%s): %s", url, exc)
+        return None
+
+    if not _validate_manifest(data):
+        logger.info("model catalog at %s failed schema validation", url)
+        return None
+
+    return data
+
+
+def _validate_manifest(data: Any) -> bool:
+    """Return True when ``data`` matches the minimum manifest shape."""
+    if not isinstance(data, dict):
+        return False
+    version = data.get("version")
+    if not isinstance(version, int) or version > SUPPORTED_SCHEMA_VERSION:
+        # Future schema version we don't understand — refuse rather than
+        # guess. Older schemas (version < 1) aren't supported either.
+        return False
+    providers = data.get("providers")
+    if not isinstance(providers, dict):
+        return False
+    for pname, pblock in providers.items():
+        if not isinstance(pname, str) or not isinstance(pblock, dict):
+            return False
+        models = pblock.get("models")
+        if not isinstance(models, list):
+            return False
+        for m in models:
+            if not isinstance(m, dict):
+                return False
+            if not isinstance(m.get("id"), str) or not m["id"].strip():
+                return False
+    return True
+
+
+def _read_disk_cache() -> tuple[dict[str, Any] | None, float]:
+    """Return ``(data_or_none, mtime)``. mtime is 0 if file is missing."""
+    path = _cache_path()
+    try:
+        mtime = path.stat().st_mtime
+    except (OSError, FileNotFoundError):
+        return (None, 0.0)
+    try:
+        with open(path) as fh:
+            data = json.load(fh)
+    except (OSError, json.JSONDecodeError):
+        return (None, 0.0)
+    if not _validate_manifest(data):
+        return (None, 0.0)
+    return (data, mtime)
+
+
+def _write_disk_cache(data: dict[str, Any]) -> None:
+    path = _cache_path()
+    try:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        tmp = path.with_suffix(path.suffix + ".tmp")
+        with open(tmp, "w") as fh:
+            json.dump(data, fh, indent=2)
+            fh.write("\n")
+        os.replace(tmp, path)
+    except OSError as exc:
+        logger.info("model catalog cache write failed: %s", exc)
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+
+def get_catalog(*, force_refresh: bool = False) -> dict[str, Any]:
+    """Return the parsed model catalog manifest, or an empty dict on failure.
+
+    Callers should treat a missing provider/model as "use the in-repo fallback"
+    — never raise from this function so the CLI keeps working offline.
+    """
+    global _catalog_cache, _catalog_cache_source_mtime
+
+    cfg = _load_catalog_config()
+    if not cfg["enabled"]:
+        return {}
+
+    ttl_seconds = max(0.0, cfg["ttl_hours"] * 3600.0)
+
+    disk_data, disk_mtime = _read_disk_cache()
+    now = time.time()
+    disk_fresh = disk_data is not None and (now - disk_mtime) < ttl_seconds
+
+    # In-process cache hit: disk hasn't changed since we loaded it and still fresh.
+    if (
+        not force_refresh
+        and _catalog_cache is not None
+        and disk_data is not None
+        and disk_mtime == _catalog_cache_source_mtime
+        and disk_fresh
+    ):
+        return _catalog_cache
+
+    # Disk is fresh enough — use it without a network hit.
+    if not force_refresh and disk_fresh and disk_data is not None:
+        _catalog_cache = disk_data
+        _catalog_cache_source_mtime = disk_mtime
+        return disk_data
+
+    # Need to (re)fetch. If it fails, fall back to any stale disk copy.
+    fetched = _fetch_manifest(cfg["url"], DEFAULT_FETCH_TIMEOUT)
+    if fetched is not None:
+        _write_disk_cache(fetched)
+        new_disk_data, new_mtime = _read_disk_cache()
+        if new_disk_data is not None:
+            _catalog_cache = new_disk_data
+            _catalog_cache_source_mtime = new_mtime
+            return new_disk_data
+        _catalog_cache = fetched
+        _catalog_cache_source_mtime = now
+        return fetched
+
+    if disk_data is not None:
+        _catalog_cache = disk_data
+        _catalog_cache_source_mtime = disk_mtime
+        return disk_data
+
+    return {}
+
+
+def _fetch_provider_override(provider: str) -> dict[str, Any] | None:
+    """If ``model_catalog.providers.<name>.url`` is set, fetch that instead."""
+    cfg = _load_catalog_config()
+    if not cfg["enabled"]:
+        return None
+    provider_cfg = cfg["providers"].get(provider)
+    if not isinstance(provider_cfg, dict):
+        return None
+    override_url = provider_cfg.get("url")
+    if not isinstance(override_url, str) or not override_url.strip():
+        return None
+    # Override fetches skip the disk cache because they're usually
+    # third-party self-hosted. Re-request on every call but with a short
+    # timeout so they don't block the picker.
+    return _fetch_manifest(override_url.strip(), DEFAULT_FETCH_TIMEOUT)
+
+
+def _get_provider_block(provider: str) -> dict[str, Any] | None:
+    """Return the provider's manifest block, respecting per-provider overrides."""
+    override = _fetch_provider_override(provider)
+    if override is not None:
+        block = override.get("providers", {}).get(provider)
+        if isinstance(block, dict):
+            return block
+
+    catalog = get_catalog()
+    if not catalog:
+        return None
+    block = catalog.get("providers", {}).get(provider)
+    return block if isinstance(block, dict) else None
+
+
+def get_curated_openrouter_models() -> list[tuple[str, str]] | None:
+    """Return OpenRouter's curated ``[(id, description), ...]`` from the manifest.
+
+    Returns ``None`` when the manifest is unavailable, so callers can fall
+    back to their hardcoded list.
+    """
+    block = _get_provider_block("openrouter")
+    if not block:
+        return None
+    out: list[tuple[str, str]] = []
+    for m in block.get("models", []):
+        mid = str(m.get("id") or "").strip()
+        if not mid:
+            continue
+        desc = str(m.get("description") or "")
+        out.append((mid, desc))
+    return out or None
+
+
+def get_curated_nous_models() -> list[str] | None:
+    """Return Nous Portal's curated list of model ids from the manifest.
+
+    Returns ``None`` when the manifest is unavailable.
+    """
+    block = _get_provider_block("nous")
+    if not block:
+        return None
+    out: list[str] = []
+    for m in block.get("models", []):
+        mid = str(m.get("id") or "").strip()
+        if mid:
+            out.append(mid)
+    return out or None
+
+
+def reset_cache() -> None:
+    """Clear the in-process cache. Used by tests and ``hermes model --refresh``."""
+    global _catalog_cache, _catalog_cache_source_mtime
+    _catalog_cache = None
+    _catalog_cache_source_mtime = 0.0

hermes_cli/model_switch.py

@@ -533,6 +533,7 @@ def resolve_display_context_length(
     base_url: str = "",
     api_key: str = "",
     model_info: Optional[ModelInfo] = None,
+    custom_providers: list | None = None,
 ) -> Optional[int]:
     """Resolve the context length to show in /model output.
 
@@ -543,6 +544,11 @@ def resolve_display_context_length(
     about Codex OAuth, Copilot, Nous, and falls back to models.dev for the
     rest.
 
+    When ``custom_providers`` is provided, per-model ``context_length``
+    overrides from ``custom_providers[].models.<id>.context_length`` are
+    honored — this closes #15779 where ``/model`` switch ignored user-set
+    overrides.
+
     Prefer the provider-aware value; fall back to ``model_info.context_window``
     only if the resolver returns nothing.
     """
@@ -553,6 +559,7 @@ def resolve_display_context_length(
             base_url=base_url or "",
             api_key=api_key or "",
             provider=provider or None,
+            custom_providers=custom_providers,
         )
         if ctx:
             return int(ctx)
@@ -831,9 +838,14 @@ def switch_model(
                 requested=current_provider,
                 target_model=new_model,
             )
-            api_key = runtime.get("api_key", "")
-            base_url = runtime.get("base_url", "")
-            api_mode = runtime.get("api_mode", "")
+            # If resolution fell through to "custom" (e.g. named custom provider like
+            # "ollama-launch" that resolve_runtime_provider doesn't know), keep existing
+            # credentials. Otherwise use the resolved values (picks up credential rotation,
+            # base_url adjustments for OpenCode, etc.).
+            if runtime.get("provider") != "custom":
+                api_key = runtime.get("api_key", "")
+                base_url = runtime.get("base_url", "")
+                api_mode = runtime.get("api_mode", "")
         except Exception:
             pass
 
@@ -867,16 +879,31 @@ def switch_model(
             "message": f"Could not validate `{new_model}`: {e}",
         }
 
+    # Override rejection if model is in the user's saved provider config.
+    # API /v1/models may not list cloud/aliased models even though the server supports them.
     if not validation.get("accepted"):
-        msg = validation.get("message", "Invalid model")
-        return ModelSwitchResult(
-            success=False,
-            new_model=new_model,
-            target_provider=target_provider,
-            provider_label=provider_label,
-            is_global=is_global,
-            error_message=msg,
-        )
+        override = False
+        if user_providers:
+            for up in user_providers:
+                if isinstance(up, dict) and up.get("provider") == target_provider:
+                    cfg_models = up.get("models", [])
+                    if new_model in cfg_models or any(
+                        m.get("name") == new_model for m in cfg_models if isinstance(m, dict)
+                    ):
+                        override = True
+                        break
+        if override:
+            validation = {"accepted": True, "persist": True, "recognized": False, "message": validation.get("message", "")}
+        else:
+            msg = validation.get("message", "Invalid model")
+            return ModelSwitchResult(
+                success=False,
+                new_model=new_model,
+                target_provider=target_provider,
+                provider_label=provider_label,
+                is_global=is_global,
+                error_message=msg,
+            )
 
     # Apply auto-correction if validation found a closer match
     if validation.get("corrected_model"):

hermes_cli/models.py

@@ -33,8 +33,6 @@ COPILOT_REASONING_EFFORTS_O_SERIES = ["low", "medium", "high"]
 # (model_id, display description shown in menus)
 OPENROUTER_MODELS: list[tuple[str, str]] = [
     ("moonshotai/kimi-k2.6",            "recommended"),
-    ("deepseek/deepseek-v4-pro",        ""),
-    ("deepseek/deepseek-v4-flash",      ""),
     ("anthropic/claude-opus-4.7",       ""),
     ("anthropic/claude-opus-4.6",       ""),
     ("anthropic/claude-sonnet-4.6",     ""),
@@ -111,8 +109,6 @@ def _codex_curated_models() -> list[str]:
 _PROVIDER_MODELS: dict[str, list[str]] = {
     "nous": [
         "moonshotai/kimi-k2.6",
-        "deepseek/deepseek-v4-pro",
-        "deepseek/deepseek-v4-flash",
         "xiaomi/mimo-v2.5-pro",
         "xiaomi/mimo-v2.5",
         "anthropic/claude-opus-4.7",
@@ -383,6 +379,9 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "us.meta.llama4-maverick-17b-instruct-v1:0",
         "us.meta.llama4-scout-17b-instruct-v1:0",
     ],
+    # Azure Foundry: user-provided endpoint and model.
+    # Empty list because models depend on the endpoint configuration.
+    "azure-foundry": [],
 }
 
 # Vercel AI Gateway: derive the bare-model-id catalog from the curated
@@ -740,6 +739,7 @@ CANONICAL_PROVIDERS: list[ProviderEntry] = [
     ProviderEntry("opencode-zen",   "OpenCode Zen",             "OpenCode Zen (35+ curated models, pay-as-you-go)"),
     ProviderEntry("opencode-go",    "OpenCode Go",              "OpenCode Go (open models, $10/month subscription)"),
     ProviderEntry("bedrock",        "AWS Bedrock",              "AWS Bedrock (Claude, Nova, Llama, DeepSeek — IAM or API key)"),
+    ProviderEntry("azure-foundry",  "Azure Foundry",            "Azure Foundry (OpenAI-style or Anthropic-style endpoint — your Azure AI deployment)"),
 ]
 
 # Derived dicts — used throughout the codebase
@@ -872,7 +872,16 @@ def fetch_openrouter_models(
     if _openrouter_catalog_cache is not None and not force_refresh:
         return list(_openrouter_catalog_cache)
 
-    fallback = list(OPENROUTER_MODELS)
+    # Prefer the remotely-hosted catalog manifest; fall back to the in-repo
+    # snapshot when the manifest is unreachable. Both are curated lists that
+    # drive the picker; the OpenRouter live /v1/models filter (tool support,
+    # free pricing) is applied on top either way.
+    try:
+        from hermes_cli.model_catalog import get_curated_openrouter_models
+        remote = get_curated_openrouter_models()
+    except Exception:
+        remote = None
+    fallback = list(remote) if remote else list(OPENROUTER_MODELS)
     preferred_ids = [mid for mid, _ in fallback]
 
     try:
@@ -925,6 +934,24 @@ def model_ids(*, force_refresh: bool = False) -> list[str]:
     return [mid for mid, _ in fetch_openrouter_models(force_refresh=force_refresh)]
 
 
+def get_curated_nous_model_ids() -> list[str]:
+    """Return the curated Nous Portal model-id list.
+
+    Prefers the remotely-hosted catalog manifest (published under
+    ``website/static/api/model-catalog.json``); falls back to the in-repo
+    snapshot in ``_PROVIDER_MODELS["nous"]`` when the manifest is
+    unreachable. Always returns a list (never None).
+    """
+    try:
+        from hermes_cli.model_catalog import get_curated_nous_models
+        remote = get_curated_nous_models()
+    except Exception:
+        remote = None
+    if remote:
+        return list(remote)
+    return list(_PROVIDER_MODELS.get("nous", []))
+
+
 def _ai_gateway_model_is_free(pricing: Any) -> bool:
     """Return True if an AI Gateway model has $0 input AND output pricing."""
     if not isinstance(pricing, dict):
@@ -1379,27 +1406,93 @@ def curated_models_for_provider(
     return [(m, "") for m in models]
 
 
-def detect_provider_for_model(
+def _provider_keys(provider: str) -> set[str]:
+    key = (provider or "").strip().lower()
+    normalized = normalize_provider(provider)
+    return {k for k in (key, normalized) if k}
+
+
+def _model_in_provider_catalog(name_lower: str, providers: set[str]) -> bool:
+    return any(
+        name_lower == model.lower()
+        for provider in providers
+        for model in _PROVIDER_MODELS.get(provider, [])
+    )
+
+
+_AGGREGATOR_PROVIDERS = frozenset(
+    {"nous", "openrouter", "ai-gateway", "copilot", "kilocode"}
+)
+
+
+def _resolve_static_model_alias(
+    name_lower: str,
+    current_keys: set[str],
+) -> Optional[tuple[str, str]]:
+    """Resolve short aliases (e.g. sonnet/opus) using static catalogs only."""
+    try:
+        from hermes_cli.model_switch import MODEL_ALIASES
+    except Exception:
+        return None
+
+    identity = MODEL_ALIASES.get(name_lower)
+    if identity is None:
+        return None
+
+    vendor = identity.vendor
+    family = identity.family
+
+    def _match(provider: str) -> Optional[str]:
+        models = _PROVIDER_MODELS.get(provider, [])
+        if not models:
+            return None
+        prefix = (
+            f"{vendor}/{family}"
+            if provider in _AGGREGATOR_PROVIDERS
+            else family
+        ).lower()
+        for model in models:
+            if model.lower().startswith(prefix):
+                return model
+        return None
+
+    for provider in current_keys:
+        if matched := _match(provider):
+            return provider, matched
+
+    for provider in _PROVIDER_MODELS:
+        if provider in current_keys or provider in _AGGREGATOR_PROVIDERS:
+            continue
+        if matched := _match(provider):
+            return provider, matched
+
+    for provider in _AGGREGATOR_PROVIDERS:
+        if provider in current_keys and (matched := _match(provider)):
+            return provider, matched
+
+    return None
+
+
+def detect_static_provider_for_model(
     model_name: str,
     current_provider: str,
 ) -> Optional[tuple[str, str]]:
-    """Auto-detect the best provider for a model name.
+    """Auto-detect a provider from static catalogs only.
 
-    Returns ``(provider_id, model_name)`` — the model name may be remapped
-    (e.g. bare ``deepseek-chat`` → ``deepseek/deepseek-chat`` for OpenRouter).
+    Returns ``(provider_id, model_name)``. The model name may be remapped
+    when a static alias or bare provider name resolves to a catalog default.
     Returns ``None`` when no confident match is found.
-
-    Priority:
-    0. Bare provider name → switch to that provider's default model
-    1. Direct provider with credentials (highest)
-    2. Direct provider without credentials → remap to OpenRouter slug
-    3. OpenRouter catalog match
     """
     name = (model_name or "").strip()
     if not name:
         return None
 
     name_lower = name.lower()
+    current_keys = _provider_keys(current_provider)
+
+    alias_match = _resolve_static_model_alias(name_lower, current_keys)
+    if alias_match:
+        return alias_match
 
     # --- Step 0: bare provider name typed as model ---
     # If someone types `/model nous` or `/model anthropic`, treat it as a
@@ -1412,64 +1505,49 @@ def detect_provider_for_model(
         if (
             resolved_provider in _PROVIDER_LABELS
             and default_models
-            and resolved_provider != normalize_provider(current_provider)
+            and resolved_provider not in current_keys
         ):
             return (resolved_provider, default_models[0])
 
     # Aggregators list other providers' models — never auto-switch TO them
-    _AGGREGATORS = {"nous", "openrouter", "ai-gateway", "copilot", "kilocode"}
-
     # If the model belongs to the current provider's catalog, don't suggest switching
-    current_models = _PROVIDER_MODELS.get(current_provider, [])
-    if any(name_lower == m.lower() for m in current_models):
+    if _model_in_provider_catalog(name_lower, current_keys):
         return None
 
     # --- Step 1: check static provider catalogs for a direct match ---
-    direct_match: Optional[str] = None
     for pid, models in _PROVIDER_MODELS.items():
-        if pid == current_provider or pid in _AGGREGATORS:
+        if pid in current_keys or pid in _AGGREGATOR_PROVIDERS:
             continue
         if any(name_lower == m.lower() for m in models):
-            direct_match = pid
-            break
+            return (pid, name)
 
-    if direct_match:
-        # Check if we have credentials for this provider — env vars,
-        # credential pool, or auth store entries.
-        has_creds = False
-        try:
-            from hermes_cli.auth import PROVIDER_REGISTRY
-            pconfig = PROVIDER_REGISTRY.get(direct_match)
-            if pconfig:
-                for env_var in pconfig.api_key_env_vars:
-                    if os.getenv(env_var, "").strip():
-                        has_creds = True
-                        break
-        except Exception:
-            pass
-        # Also check credential pool and auth store — covers OAuth,
-        # Claude Code tokens, and other non-env-var credentials (#10300).
-        if not has_creds:
-            try:
-                from agent.credential_pool import load_pool
-                pool = load_pool(direct_match)
-                if pool.has_credentials():
-                    has_creds = True
-            except Exception:
-                pass
-        if not has_creds:
-            try:
-                from hermes_cli.auth import _load_auth_store
-                store = _load_auth_store()
-                if direct_match in store.get("providers", {}) or direct_match in store.get("credential_pool", {}):
-                    has_creds = True
-            except Exception:
-                pass
+    return None
 
-        # Always return the direct provider match.  If credentials are
-        # missing, the client init will give a clear error rather than
-        # silently routing through the wrong provider (#10300).
-        return (direct_match, name)
+
+def detect_provider_for_model(
+    model_name: str,
+    current_provider: str,
+) -> Optional[tuple[str, str]]:
+    """Auto-detect the best provider for a model name.
+
+    Returns ``(provider_id, model_name)`` — the model name may be remapped
+    (e.g. bare ``deepseek-chat`` → ``deepseek/deepseek-chat`` for OpenRouter).
+    Returns ``None`` when no confident match is found.
+
+    Priority:
+    0. Bare provider name → switch to that provider's default model
+    1. Direct provider static catalog match
+    2. OpenRouter catalog match
+    """
+    name = (model_name or "").strip()
+    if not name:
+        return None
+
+    static_match = detect_static_provider_for_model(name, current_provider)
+    if static_match:
+        return static_match
+    if _model_in_provider_catalog(name.lower(), _provider_keys(current_provider)):
+        return None
 
     # --- Step 2: check OpenRouter catalog ---
     # First try exact match (handles provider/model format)
@@ -2571,8 +2649,8 @@ def validate_requested_model(
                 )
 
             return {
-                "accepted": False,
-                "persist": False,
+                "accepted": True,
+                "persist": True,
                 "recognized": False,
                 "message": message,
             }

hermes_cli/nous_subscription.py

@@ -9,6 +9,7 @@ from typing import Dict, Iterable, Optional, Set
 from hermes_cli.auth import get_nous_auth_status
 from hermes_cli.config import get_env_value, load_config
 from tools.managed_tool_gateway import is_managed_tool_gateway_ready
+from utils import is_truthy_value
 from tools.tool_backend_helpers import (
     fal_key_is_configured,
     has_direct_modal_credentials,
@@ -25,6 +26,13 @@ _DEFAULT_PLATFORM_TOOLSETS = {
 }
 
 
+def _uses_gateway(section: object) -> bool:
+    """Return True when a config section explicitly opts into the gateway."""
+    if not isinstance(section, dict):
+        return False
+    return is_truthy_value(section.get("use_gateway"), default=False)
+
+
 @dataclass(frozen=True)
 class NousFeatureState:
     key: str
@@ -262,11 +270,11 @@ def get_nous_subscription_features(
     # use_gateway flags — when True, the user explicitly opted into the
     # Tool Gateway via `hermes model`, so direct credentials should NOT
     # prevent gateway routing.
-    web_use_gateway = bool(web_cfg.get("use_gateway"))
-    tts_use_gateway = bool(tts_cfg.get("use_gateway"))
-    browser_use_gateway = bool(browser_cfg.get("use_gateway"))
+    web_use_gateway = _uses_gateway(web_cfg)
+    tts_use_gateway = _uses_gateway(tts_cfg)
+    browser_use_gateway = _uses_gateway(browser_cfg)
     image_gen_cfg = config.get("image_gen") if isinstance(config.get("image_gen"), dict) else {}
-    image_use_gateway = bool(image_gen_cfg.get("use_gateway"))
+    image_use_gateway = _uses_gateway(image_gen_cfg)
 
     direct_exa = bool(get_env_value("EXA_API_KEY"))
     direct_firecrawl = bool(get_env_value("FIRECRAWL_API_KEY") or get_env_value("FIRECRAWL_API_URL"))
@@ -601,10 +609,10 @@ def get_gateway_eligible_tools(
     # no direct keys exist — we only skip the prompt for tools where
     # use_gateway was explicitly set.
     opted_in = {
-        "web": bool((config.get("web") if isinstance(config.get("web"), dict) else {}).get("use_gateway")),
-        "image_gen": bool((config.get("image_gen") if isinstance(config.get("image_gen"), dict) else {}).get("use_gateway")),
-        "tts": bool((config.get("tts") if isinstance(config.get("tts"), dict) else {}).get("use_gateway")),
-        "browser": bool((config.get("browser") if isinstance(config.get("browser"), dict) else {}).get("use_gateway")),
+        "web": _uses_gateway(config.get("web")),
+        "image_gen": _uses_gateway(config.get("image_gen")),
+        "tts": _uses_gateway(config.get("tts")),
+        "browser": _uses_gateway(config.get("browser")),
     }
 
     unconfigured: list[str] = []

hermes_cli/oneshot.py

@@ -0,0 +1,202 @@
+"""Oneshot (-z) mode: send a prompt, get the final content block, exit.
+
+Bypasses cli.py entirely.  No banner, no spinner, no session_id line,
+no stderr chatter.  Just the agent's final text to stdout.
+
+Toolsets = whatever the user has configured for "cli" in `hermes tools`.
+Rules / memory / AGENTS.md / preloaded skills = same as a normal chat turn.
+Approvals = auto-bypassed (HERMES_YOLO_MODE=1 is set for the call).
+Working directory = the user's CWD (AGENTS.md etc. resolve from there as usual).
+
+Model / provider selection mirrors `hermes chat`:
+    - Both optional. If omitted, use the user's configured default.
+    - If both given, pair them exactly as given.
+    - If only --model given, auto-detect the provider that serves it.
+    - If only --provider given, error out (ambiguous — caller must pick a model).
+
+Env var fallbacks (used when the corresponding arg is not passed):
+    - HERMES_INFERENCE_MODEL
+    - HERMES_INFERENCE_PROVIDER  (already read by resolve_runtime_provider)
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import sys
+from contextlib import redirect_stderr, redirect_stdout
+from typing import Optional
+
+
+def run_oneshot(
+    prompt: str,
+    model: Optional[str] = None,
+    provider: Optional[str] = None,
+) -> int:
+    """Execute a single prompt and print only the final content block.
+
+    Args:
+        prompt: The user message to send.
+        model: Optional model override. Falls back to HERMES_INFERENCE_MODEL
+            env var, then config.yaml's model.default / model.model.
+        provider: Optional provider override. Falls back to
+            HERMES_INFERENCE_PROVIDER env var, then config.yaml's model.provider,
+            then "auto".
+
+    Returns the exit code.  Caller should sys.exit() with the return.
+    """
+    # Silence every stdlib logger for the duration.  AIAgent, tools, and
+    # provider adapters all log to stderr through the root logger; file
+    # handlers added by setup_logging() keep working (they're attached to
+    # the root logger's handler list, not affected by level), but no
+    # bytes reach the terminal.
+    logging.disable(logging.CRITICAL)
+
+    # --provider without --model is ambiguous: carrying the user's configured
+    # model across to a different provider is usually wrong (that provider may
+    # not host it), and silently picking the provider's catalog default hides
+    # the mismatch.  Require the caller to be explicit.  Validate BEFORE the
+    # stderr redirect so the message actually reaches the terminal.
+    env_model_early = os.getenv("HERMES_INFERENCE_MODEL", "").strip()
+    if provider and not ((model or "").strip() or env_model_early):
+        sys.stderr.write(
+            "hermes -z: --provider requires --model (or HERMES_INFERENCE_MODEL). "
+            "Pass both explicitly, or neither to use your configured defaults.\n"
+        )
+        return 2
+
+    # Auto-approve any shell / tool approvals.  Non-interactive by
+    # definition — a prompt would hang forever.
+    os.environ["HERMES_YOLO_MODE"] = "1"
+    os.environ["HERMES_ACCEPT_HOOKS"] = "1"
+
+    # Redirect stderr AND stdout to devnull for the entire call tree.
+    # We'll print the final response to the real stdout at the end.
+    real_stdout = sys.stdout
+    devnull = open(os.devnull, "w")
+
+    try:
+        with redirect_stdout(devnull), redirect_stderr(devnull):
+            response = _run_agent(prompt, model=model, provider=provider)
+    finally:
+        try:
+            devnull.close()
+        except Exception:
+            pass
+
+    if response:
+        real_stdout.write(response)
+        if not response.endswith("\n"):
+            real_stdout.write("\n")
+        real_stdout.flush()
+    return 0
+
+
+def _run_agent(
+    prompt: str,
+    model: Optional[str] = None,
+    provider: Optional[str] = None,
+) -> str:
+    """Build an AIAgent exactly like a normal CLI chat turn would, then
+    run a single conversation.  Returns the final response string."""
+    # Imports are local so they don't run when hermes is invoked for
+    # other commands (keeps top-level CLI startup cheap).
+    from hermes_cli.config import load_config
+    from hermes_cli.models import detect_provider_for_model
+    from hermes_cli.runtime_provider import resolve_runtime_provider
+    from hermes_cli.tools_config import _get_platform_tools
+    from run_agent import AIAgent
+
+    cfg = load_config()
+
+    # Resolve effective model: explicit arg → env var → config.
+    model_cfg = cfg.get("model") or {}
+    if isinstance(model_cfg, str):
+        cfg_model = model_cfg
+    else:
+        cfg_model = model_cfg.get("default") or model_cfg.get("model") or ""
+
+    env_model = os.getenv("HERMES_INFERENCE_MODEL", "").strip()
+    effective_model = (model or "").strip() or env_model or cfg_model
+
+    # Resolve effective provider: explicit arg → (auto-detect from model if
+    # model was explicit) → env / config (handled inside resolve_runtime_provider).
+    #
+    # When --model is given without --provider, auto-detect the provider that
+    # serves that model — same semantic as `/model <name>` in an interactive
+    # session.  Without this, resolve_runtime_provider() would fall back to
+    # the user's configured default provider, which may not host the model
+    # the caller just asked for.
+    effective_provider = (provider or "").strip() or None
+    if effective_provider is None and (model or env_model):
+        # Only auto-detect when the model was explicitly requested via arg or
+        # env var (not when it came from config — that's the "use my defaults"
+        # path and the configured provider is already correct).
+        explicit_model = (model or "").strip() or env_model
+        if explicit_model:
+            cfg_provider = ""
+            if isinstance(model_cfg, dict):
+                cfg_provider = str(model_cfg.get("provider") or "").strip().lower()
+            current_provider = (
+                cfg_provider
+                or os.getenv("HERMES_INFERENCE_PROVIDER", "").strip().lower()
+                or "auto"
+            )
+            detected = detect_provider_for_model(explicit_model, current_provider)
+            if detected:
+                effective_provider, effective_model = detected
+
+    runtime = resolve_runtime_provider(
+        requested=effective_provider,
+        target_model=effective_model or None,
+    )
+
+    # Pull in whatever toolsets the user has enabled for "cli".
+    # sorted() gives stable ordering; set→list for AIAgent's signature.
+    toolsets_list = sorted(_get_platform_tools(cfg, "cli"))
+
+    agent = AIAgent(
+        api_key=runtime.get("api_key"),
+        base_url=runtime.get("base_url"),
+        provider=runtime.get("provider"),
+        api_mode=runtime.get("api_mode"),
+        model=effective_model,
+        enabled_toolsets=toolsets_list,
+        quiet_mode=True,
+        platform="cli",
+        credential_pool=runtime.get("credential_pool"),
+        # Interactive callbacks are intentionally NOT wired beyond this
+        # one.  In oneshot mode there's no user sitting at a terminal:
+        #   - clarify  → returns a synthetic "pick a default" instruction
+        #                so the agent continues instead of stalling on
+        #                the tool's built-in "not available" error
+        #   - sudo password prompt → terminal_tool gates on
+        #                HERMES_INTERACTIVE which we never set
+        #   - shell-hook approval → auto-approved via HERMES_ACCEPT_HOOKS=1
+        #                (set above); also falls back to deny on non-tty
+        #   - dangerous-command approval → bypassed via HERMES_YOLO_MODE=1
+        #   - skill secret capture → returns gracefully when no callback set
+        clarify_callback=_oneshot_clarify_callback,
+    )
+
+    # Belt-and-braces: make sure AIAgent doesn't invoke any streaming
+    # display callbacks that would bypass our stdout capture.
+    agent.suppress_status_output = True
+    agent.stream_delta_callback = None
+    agent.tool_gen_callback = None
+
+    return agent.chat(prompt) or ""
+
+
+def _oneshot_clarify_callback(question: str, choices=None) -> str:
+    """Clarify is disabled in oneshot mode — tell the agent to pick a
+    default and proceed instead of stalling or erroring."""
+    if choices:
+        return (
+            f"[oneshot mode: no user available. Pick the best option from "
+            f"{choices} using your own judgment and continue.]"
+        )
+    return (
+        "[oneshot mode: no user available. Make the most reasonable "
+        "assumption you can and continue.]"
+    )

hermes_cli/platforms.py

@@ -36,6 +36,7 @@ PLATFORMS: OrderedDict[str, PlatformInfo] = OrderedDict([
     ("wecom_callback", PlatformInfo(label="💬 WeCom Callback",  default_toolset="hermes-wecom-callback")),
     ("weixin",         PlatformInfo(label="💬 Weixin",          default_toolset="hermes-weixin")),
     ("qqbot",          PlatformInfo(label="💬 QQBot",           default_toolset="hermes-qqbot")),
+    ("yuanbao",        PlatformInfo(label="🤖 Yuanbao",         default_toolset="hermes-yuanbao")),
     ("webhook",        PlatformInfo(label="🔗 Webhook",         default_toolset="hermes-webhook")),
     ("api_server",     PlatformInfo(label="🌐 API Server",      default_toolset="hermes-api-server")),
     ("cron",           PlatformInfo(label="⏰ Cron",            default_toolset="hermes-cron")),

hermes_cli/providers.py

@@ -167,6 +167,12 @@ HERMES_OVERLAYS: Dict[str, HermesOverlay] = {
         transport="openai_chat",
         base_url_env_var="OLLAMA_BASE_URL",
     ),
+    # Azure Foundry: supports both OpenAI-style and Anthropic-style endpoints.
+    # The transport is determined at runtime from config.yaml model.api_mode.
+    "azure-foundry": HermesOverlay(
+        transport="openai_chat",  # default; overridden by api_mode in config
+        base_url_env_var="AZURE_FOUNDRY_BASE_URL",
+    ),
 }
 
 

hermes_cli/runtime_provider.py

@@ -221,6 +221,19 @@ def _resolve_runtime_from_pool_entry(
     elif provider == "copilot":
         api_mode = _copilot_runtime_api_mode(model_cfg, getattr(entry, "runtime_api_key", ""))
         base_url = base_url or PROVIDER_REGISTRY["copilot"].inference_base_url
+    elif provider == "azure-foundry":
+        # Azure Foundry: read api_mode and base_url from config
+        cfg_provider = str(model_cfg.get("provider") or "").strip().lower()
+        if cfg_provider == "azure-foundry":
+            cfg_base_url = str(model_cfg.get("base_url") or "").strip().rstrip("/")
+            if cfg_base_url:
+                base_url = cfg_base_url
+            configured_mode = _parse_api_mode(model_cfg.get("api_mode"))
+            if configured_mode:
+                api_mode = configured_mode
+        # For Anthropic-style endpoints, strip /v1 suffix
+        if api_mode == "anthropic_messages":
+            base_url = re.sub(r"/v1/?$", "", base_url)
     else:
         configured_provider = str(model_cfg.get("provider") or "").strip().lower()
         # Honour model.base_url from config.yaml when the configured provider
@@ -589,6 +602,71 @@ def _resolve_openrouter_runtime(
     }
 
 
+def _resolve_azure_foundry_runtime(
+    *,
+    requested_provider: str,
+    model_cfg: Dict[str, Any],
+    explicit_api_key: Optional[str] = None,
+    explicit_base_url: Optional[str] = None,
+) -> Dict[str, Any]:
+    """Resolve an Azure Foundry runtime entry.
+
+    Reads ``model.base_url`` + ``model.api_mode`` from config.yaml (or
+    explicit overrides), pulls the API key from ``.env`` / env var, and
+    strips a trailing ``/v1`` for Anthropic-style endpoints because the
+    Anthropic SDK appends ``/v1/messages`` internally.
+
+    Raises :class:`AuthError` when required values are missing.
+    """
+    explicit_api_key = str(explicit_api_key or "").strip()
+    explicit_base_url_clean = str(explicit_base_url or "").strip().rstrip("/")
+
+    cfg_provider = str(model_cfg.get("provider") or "").strip().lower()
+    cfg_base_url = ""
+    cfg_api_mode = "chat_completions"
+    if cfg_provider == "azure-foundry":
+        cfg_base_url = str(model_cfg.get("base_url") or "").strip().rstrip("/")
+        cfg_api_mode = _parse_api_mode(model_cfg.get("api_mode")) or "chat_completions"
+
+    env_base_url = os.getenv("AZURE_FOUNDRY_BASE_URL", "").strip().rstrip("/")
+    base_url = explicit_base_url_clean or cfg_base_url or env_base_url
+    if not base_url:
+        raise AuthError(
+            "Azure Foundry requires a base URL. Set it via 'hermes model' or "
+            "the AZURE_FOUNDRY_BASE_URL environment variable."
+        )
+
+    api_key = explicit_api_key
+    if not api_key:
+        try:
+            from hermes_cli.config import get_env_value
+            api_key = get_env_value("AZURE_FOUNDRY_API_KEY") or ""
+        except Exception:
+            api_key = ""
+    if not api_key:
+        api_key = os.getenv("AZURE_FOUNDRY_API_KEY", "").strip()
+    if not api_key:
+        raise AuthError(
+            "Azure Foundry requires an API key. Set AZURE_FOUNDRY_API_KEY in "
+            "~/.hermes/.env or run 'hermes model' to configure."
+        )
+
+    # Anthropic SDK appends /v1/messages itself, so strip any trailing /v1
+    # we inherited from the configured base_url to avoid double-/v1 paths.
+    if cfg_api_mode == "anthropic_messages":
+        base_url = re.sub(r"/v1/?$", "", base_url)
+
+    source = "explicit" if (explicit_api_key or explicit_base_url) else "config"
+    return {
+        "provider": "azure-foundry",
+        "api_mode": cfg_api_mode,
+        "base_url": base_url,
+        "api_key": api_key,
+        "source": source,
+        "requested_provider": requested_provider,
+    }
+
+
 def _resolve_explicit_runtime(
     *,
     provider: str,
@@ -678,6 +756,15 @@ def _resolve_explicit_runtime(
             "requested_provider": requested_provider,
         }
 
+    # Azure Foundry: user-configured endpoint with selectable API mode
+    if provider == "azure-foundry":
+        return _resolve_azure_foundry_runtime(
+            requested_provider=requested_provider,
+            model_cfg=model_cfg,
+            explicit_api_key=explicit_api_key,
+            explicit_base_url=explicit_base_url,
+        )
+
     pconfig = PROVIDER_REGISTRY.get(provider)
     if pconfig and pconfig.auth_type == "api_key":
         env_url = ""
@@ -746,6 +833,40 @@ def resolve_runtime_provider(
     """
     requested_provider = resolve_requested_provider(requested)
 
+    # Azure Anthropic short-circuit: when explicitly targeting an Azure endpoint
+    # with provider="anthropic", bypass _resolve_named_custom_runtime (which would
+    # return provider="custom" with chat_completions api_mode and no valid key).
+    # Instead, use the Azure key directly with anthropic_messages api_mode.
+    _eff_base = (explicit_base_url or "").strip()
+    if requested_provider == "anthropic" and "azure.com" in _eff_base:
+        _azure_key = (
+            (explicit_api_key or "").strip()
+            or os.getenv("AZURE_ANTHROPIC_KEY", "").strip()
+            or os.getenv("ANTHROPIC_API_KEY", "").strip()
+        )
+        return {
+            "provider": "anthropic",
+            "api_mode": "anthropic_messages",
+            "base_url": _eff_base.rstrip("/"),
+            "api_key": _azure_key,
+            "source": "azure-explicit",
+            "requested_provider": requested_provider,
+        }
+
+    # Azure Foundry: user-configured endpoint with selectable API mode
+    # (OpenAI-style chat_completions or Anthropic-style anthropic_messages).
+    # Resolve before the custom-runtime / pool / generic paths so Azure
+    # config is always picked up from model.base_url + model.api_mode,
+    # regardless of whether the caller passed explicit_* args.
+    if requested_provider == "azure-foundry":
+        azure_runtime = _resolve_azure_foundry_runtime(
+            requested_provider=requested_provider,
+            model_cfg=_get_model_config(),
+            explicit_api_key=explicit_api_key,
+            explicit_base_url=explicit_base_url,
+        )
+        return azure_runtime
+
     custom_runtime = _resolve_named_custom_runtime(
         requested_provider=requested_provider,
         explicit_api_key=explicit_api_key,
@@ -924,13 +1045,6 @@ def resolve_runtime_provider(
 
     # Anthropic (native Messages API)
     if provider == "anthropic":
-        from agent.anthropic_adapter import resolve_anthropic_token
-        token = resolve_anthropic_token()
-        if not token:
-            raise AuthError(
-                "No Anthropic credentials found. Set ANTHROPIC_TOKEN or ANTHROPIC_API_KEY, "
-                "run 'claude setup-token', or authenticate with 'claude /login'."
-            )
         # Allow base URL override from config.yaml model.base_url, but only
         # when the configured provider is anthropic — otherwise a non-Anthropic
         # base_url (e.g. Codex endpoint) would leak into Anthropic requests.
@@ -939,6 +1053,33 @@ def resolve_runtime_provider(
         if cfg_provider == "anthropic":
             cfg_base_url = (model_cfg.get("base_url") or "").strip().rstrip("/")
         base_url = cfg_base_url or "https://api.anthropic.com"
+
+        # For Azure AI Foundry endpoints, use ANTHROPIC_API_KEY directly —
+        # Claude Code OAuth tokens (sk-ant-oat01) are not accepted by Azure.
+        # Azure keys don't start with "sk-ant-" so resolve_anthropic_token()
+        # would find the Claude Code OAuth token first (priority 3) and return
+        # that instead, causing 401s. Detect Azure endpoints and use the env
+        # key directly to bypass the OAuth priority chain.
+        _is_azure_endpoint = "azure.com" in base_url.lower() or (
+            cfg_base_url and "azure.com" in cfg_base_url.lower()
+        )
+        if _is_azure_endpoint:
+            token = (
+                os.getenv("AZURE_ANTHROPIC_KEY", "").strip()
+                or os.getenv("ANTHROPIC_API_KEY", "").strip()
+            )
+            if not token:
+                raise AuthError(
+                    "No Azure Anthropic API key found. Set AZURE_ANTHROPIC_KEY or ANTHROPIC_API_KEY."
+                )
+        else:
+            from agent.anthropic_adapter import resolve_anthropic_token
+            token = resolve_anthropic_token()
+            if not token:
+                raise AuthError(
+                    "No Anthropic credentials found. Set ANTHROPIC_TOKEN or ANTHROPIC_API_KEY, "
+                    "run 'claude setup-token', or authenticate with 'claude /login'."
+                )
         return {
             "provider": "anthropic",
             "api_mode": "anthropic_messages",

hermes_cli/setup.py

@@ -1856,27 +1856,32 @@ def _setup_slack():
     if existing:
         print_info("Slack: already configured")
         if not prompt_yes_no("Reconfigure Slack?", False):
+            # Even without reconfiguring, offer to refresh the manifest so
+            # new commands (e.g. /btw, /stop, ...) get registered in Slack.
+            if prompt_yes_no(
+                "Regenerate the Slack app manifest with the latest command "
+                "list? (recommended after `hermes update`)",
+                True,
+            ):
+                _write_slack_manifest_and_instruct()
             return
 
     print_info("Steps to create a Slack app:")
-    print_info("   1. Go to https://api.slack.com/apps → Create New App (from scratch)")
+    print_info("   1. Go to https://api.slack.com/apps → Create New App")
+    print_info("      Pick 'From an app manifest' — we'll generate one for you below.")
     print_info("   2. Enable Socket Mode: Settings → Socket Mode → Enable")
     print_info("      • Create an App-Level Token with 'connections:write' scope")
-    print_info("   3. Add Bot Token Scopes: Features → OAuth & Permissions")
-    print_info("      Required scopes: chat:write, app_mentions:read,")
-    print_info("      channels:history, channels:read, im:history,")
-    print_info("      im:read, im:write, users:read, files:read, files:write")
-    print_info("      Optional for private channels: groups:history")
-    print_info("   4. Subscribe to Events: Features → Event Subscriptions → Enable")
-    print_info("      Required events: message.im, message.channels, app_mention")
-    print_info("      Optional for private channels: message.groups")
-    print_warning("   ⚠ Without message.channels the bot will ONLY work in DMs,")
-    print_warning("     not public channels.")
-    print_info("   5. Install to Workspace: Settings → Install App")
-    print_info("   6. Reinstall the app after any scope or event changes")
-    print_info("   7. After installing, invite the bot to channels: /invite @YourBot")
+    print_info("   3. Install to Workspace: Settings → Install App")
+    print_info("   4. After installing, invite the bot to channels: /invite @YourBot")
     print()
     print_info("   Full guide: https://hermes-agent.nousresearch.com/docs/user-guide/messaging/slack/")
+    print()
+
+    # Generate and write manifest up-front so the user can paste it into
+    # the "Create from manifest" flow instead of clicking through scopes /
+    # events / slash commands one at a time.
+    _write_slack_manifest_and_instruct()
+
     print()
     bot_token = prompt("Slack Bot Token (xoxb-...)", password=True)
     if not bot_token:
@@ -1902,6 +1907,49 @@ def _setup_slack():
         print_info("   Set SLACK_ALLOW_ALL_USERS=true or GATEWAY_ALLOW_ALL_USERS=true only if you intentionally want open workspace access.")
 
 
+def _write_slack_manifest_and_instruct():
+    """Generate the Slack manifest, write it under HERMES_HOME, and print
+    paste-into-Slack instructions.
+
+    Exposed as its own helper so both the initial setup flow and the
+    "reconfigure? → no" branch can refresh the manifest without the user
+    re-entering tokens. Failures are non-fatal — if the manifest write
+    fails for any reason, we print a warning and skip rather than abort
+    the whole Slack setup.
+    """
+    try:
+        from hermes_cli.slack_cli import _build_full_manifest
+        from hermes_constants import get_hermes_home
+
+        manifest = _build_full_manifest(
+            bot_name="Hermes",
+            bot_description="Your Hermes agent on Slack",
+        )
+        target = Path(get_hermes_home()) / "slack-manifest.json"
+        target.parent.mkdir(parents=True, exist_ok=True)
+        import json as _json
+        target.write_text(
+            _json.dumps(manifest, indent=2, ensure_ascii=False) + "\n",
+            encoding="utf-8",
+        )
+        print_success(f"Slack app manifest written to: {target}")
+        print_info(
+            "   Paste it into https://api.slack.com/apps → your app → Features "
+            "→ App Manifest → Edit, then Save.  Slack will prompt to "
+            "reinstall if scopes or slash commands changed."
+        )
+        print_info(
+            "   Re-run `hermes slack manifest --write` anytime to refresh after "
+            "Hermes adds new commands."
+        )
+    except Exception as exc:  # pragma: no cover - best-effort UX helper
+        print_warning(f"Couldn't write Slack manifest: {exc}")
+        print_info(
+            "   You can generate it manually later with: "
+            "hermes slack manifest --write"
+        )
+
+
 def _setup_matrix():
     """Configure Matrix credentials."""
     print_header("Matrix")
@@ -2085,6 +2133,12 @@ def _setup_feishu():
     _gateway_setup_feishu()
 
 
+def _setup_yuanbao():
+    """Configure Yuanbao via gateway setup."""
+    from hermes_cli.gateway import _setup_yuanbao as _gateway_setup_yuanbao
+    _gateway_setup_yuanbao()
+
+
 def _setup_wecom():
     """Configure WeCom (Enterprise WeChat) via gateway setup."""
     from hermes_cli.gateway import _setup_wecom as _gateway_setup_wecom
@@ -2229,6 +2283,7 @@ _GATEWAY_PLATFORMS = [
     ("WhatsApp", "WHATSAPP_ENABLED", _setup_whatsapp),
     ("DingTalk", "DINGTALK_CLIENT_ID", _setup_dingtalk),
     ("Feishu / Lark", "FEISHU_APP_ID", _setup_feishu),
+    ("Yuanbao", "YUANBAO_APP_ID", _setup_yuanbao),
     ("WeCom (Enterprise WeChat)", "WECOM_BOT_ID", _setup_wecom),
     ("WeCom Callback (Self-Built App)", "WECOM_CALLBACK_CORP_ID", _setup_wecom_callback),
     ("Weixin (WeChat)", "WEIXIN_ACCOUNT_ID", _setup_weixin),
@@ -2863,17 +2918,6 @@ SETUP_SECTIONS = [
     ("agent", "Agent Settings", setup_agent_settings),
 ]
 
-# The returning-user menu intentionally omits standalone TTS because model setup
-# already includes TTS selection and tools setup covers the rest of the provider
-# configuration. Keep this list in the same order as the visible menu entries.
-RETURNING_USER_MENU_SECTION_KEYS = [
-    "model",
-    "terminal",
-    "gateway",
-    "tools",
-    "agent",
-]
-
 
 def run_setup_wizard(args):
     """Run the interactive setup wizard.
@@ -2898,6 +2942,9 @@ def run_setup_wizard(args):
         save_config(copy.deepcopy(DEFAULT_CONFIG))
         print_success("Configuration reset to defaults.")
 
+    reconfigure_requested = bool(getattr(args, "reconfigure", False))
+    quick_requested = bool(getattr(args, "quick", False))
+
     config = load_config()
     hermes_home = get_hermes_home()
 
@@ -2989,50 +3036,36 @@ def run_setup_wizard(args):
     migration_ran = False
 
     if is_existing:
-        # ── Returning User Menu ──
-        print()
-        print_header("Welcome Back!")
-        print_success("You already have Hermes configured.")
-        print()
-
-        menu_choices = [
-            "Quick Setup - configure missing items only",
-            "Full Setup - reconfigure everything",
-            "Model & Provider",
-            "Terminal Backend",
-            "Messaging Platforms (Gateway)",
-            "Tools",
-            "Agent Settings",
-            "Exit",
-        ]
-        choice = prompt_choice("What would you like to do?", menu_choices, 0)
-
-        if choice == 0:
-            # Quick setup
+        # Existing install — default is the full-wizard reconfigure flow.
+        # Every prompt shows the current value as its default, so pressing
+        # Enter keeps it.  Opt into `--quick` for the narrow "just fill in
+        # missing items" flow (useful after a partial OpenClaw migration
+        # or when a required API key got cleared).
+        if quick_requested:
             _run_quick_setup(config, hermes_home)
             return
-        elif choice == 1:
-            # Full setup — fall through to run all sections
-            pass
-        elif choice == 7:
-            print_info("Exiting. Run 'hermes setup' again when ready.")
-            return
-        elif 2 <= choice <= 6:
-            # Individual section — map by key, not by position.
-            # SETUP_SECTIONS includes TTS but the returning-user menu skips it,
-            # so positional indexing (choice - 2) would dispatch the wrong section.
-            section_key = RETURNING_USER_MENU_SECTION_KEYS[choice - 2]
-            section = next((s for s in SETUP_SECTIONS if s[0] == section_key), None)
-            if section:
-                _, label, func = section
-                func(config)
-                save_config(config)
-                _print_setup_summary(config, hermes_home)
-            return
+
+        print()
+        print_header("Reconfigure")
+        print_success("You already have Hermes configured.")
+        print_info("Running the full wizard — each prompt shows your current value.")
+        print_info("Press Enter to keep it, or type a new value to change it.")
+        print_info("")
+        print_info("Tip: jump straight to a section with 'hermes setup model|terminal|")
+        print_info("     gateway|tools|agent', or fill only missing items with --quick.")
+        # Fall through to the "Full Setup — run all sections" block below.
+        # --reconfigure is now the default on existing installs; the flag
+        # is preserved for backwards compatibility but is a no-op here.
     else:
         # ── First-Time Setup ──
         print()
 
+        # --reconfigure / --quick on a fresh install are meaningless — fall
+        # through to the normal first-time flow.
+        if reconfigure_requested or quick_requested:
+            print_info("No existing configuration found — running first-time setup.")
+            print()
+
         # Offer OpenClaw migration before configuration begins
         migration_ran = _offer_openclaw_migration(hermes_home)
         if migration_ran:

hermes_cli/skills_hub.py

@@ -11,9 +11,10 @@ handler are thin wrappers that parse args and delegate.
 """
 
 import json
+import re
 import shutil
 from pathlib import Path
-from typing import Any, Dict, Optional
+from typing import Any, Dict, List, Optional
 
 from rich.console import Console
 from rich.panel import Panel
@@ -141,6 +142,103 @@ def _derive_category_from_install_path(install_path: str) -> str:
     return "" if parent == "." else parent
 
 
+# ---------------------------------------------------------------------------
+# Interactive name/category resolution for URL-installed skills
+# ---------------------------------------------------------------------------
+
+_VALID_NAME_RE = re.compile(r"^[a-z][a-z0-9_-]*$")
+_VALID_CATEGORY_RE = re.compile(r"^[a-z][a-z0-9_/-]*$")
+
+
+def _is_valid_installed_skill_name(name: str) -> bool:
+    """Accept identifier-shaped names, reject empty / sentinel-y values."""
+    if not isinstance(name, str):
+        return False
+    candidate = name.strip().lower()
+    if not candidate or candidate in {"skill", "readme", "index", "unnamed-skill"}:
+        return False
+    return bool(_VALID_NAME_RE.match(candidate))
+
+
+def _existing_categories() -> List[str]:
+    """Return sorted subdirectory names under ``~/.hermes/skills/`` that look
+    like category buckets (contain at least one ``SKILL.md`` somewhere below).
+
+    Used to suggest reusable categories when interactively installing from a
+    URL. Hidden dirs (``.hub``, ``.trash``) are skipped.
+    """
+    from tools.skills_hub import SKILLS_DIR
+    out: List[str] = []
+    try:
+        for entry in SKILLS_DIR.iterdir():
+            if not entry.is_dir() or entry.name.startswith("."):
+                continue
+            # Only count as a category if it contains skills, not if it IS a skill.
+            # Heuristic: if ``<entry>/SKILL.md`` exists, it's a skill at the
+            # top level (no category); otherwise treat as a category bucket.
+            if (entry / "SKILL.md").exists():
+                continue
+            # Has at least one nested SKILL.md?
+            try:
+                if any(entry.rglob("SKILL.md")):
+                    out.append(entry.name)
+            except OSError:
+                continue
+    except (FileNotFoundError, OSError):
+        return []
+    return sorted(set(out))
+
+
+def _prompt_for_skill_name(c: Console, url: str, default: str = "") -> Optional[str]:
+    """Prompt interactively for a skill name. Returns None on cancel/EOF."""
+    c.print()
+    c.print(
+        f"[yellow]The SKILL.md at {url} doesn't declare a `name:` in its "
+        f"frontmatter,[/]\n[yellow]and the URL path doesn't produce a valid "
+        f"identifier either.[/]"
+    )
+    default_hint = f" [{default}]" if default else ""
+    c.print(
+        f"[bold]Enter a skill name{default_hint}:[/] "
+        f"[dim](lowercase letters, digits, hyphens, underscores; starts with a letter)[/]"
+    )
+    try:
+        answer = input("Name: ").strip()
+    except (EOFError, KeyboardInterrupt):
+        return None
+    if not answer and default:
+        answer = default
+    if not _is_valid_installed_skill_name(answer):
+        c.print(f"[bold red]Invalid name:[/] {answer!r}. Aborting install.\n")
+        return None
+    return answer
+
+
+def _prompt_for_category(c: Console, existing: List[str]) -> str:
+    """Prompt interactively for a category. Empty/None input means flat install."""
+    c.print()
+    if existing:
+        c.print(
+            "[bold]Pick a category[/] "
+            "[dim](reuse an existing bucket, type a new one, or press Enter to install flat)[/]"
+        )
+        c.print(f"[dim]Existing: {', '.join(existing)}[/]")
+    else:
+        c.print(
+            "[bold]Category[/] [dim](optional — press Enter to install flat at ~/.hermes/skills/<name>/)[/]"
+        )
+    try:
+        answer = input("Category: ").strip()
+    except (EOFError, KeyboardInterrupt):
+        return ""
+    if not answer:
+        return ""
+    if not _VALID_CATEGORY_RE.match(answer):
+        c.print(f"[dim]Invalid category {answer!r} — installing flat.[/]")
+        return ""
+    return answer
+
+
 def do_search(query: str, source: str = "all", limit: int = 10,
               console: Optional[Console] = None) -> None:
     """Search registries and display results as a Rich table."""
@@ -309,8 +407,17 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
 
 def do_install(identifier: str, category: str = "", force: bool = False,
                console: Optional[Console] = None, skip_confirm: bool = False,
-               invalidate_cache: bool = True) -> None:
-    """Fetch, quarantine, scan, confirm, and install a skill."""
+               invalidate_cache: bool = True,
+               name_override: str = "") -> None:
+    """Fetch, quarantine, scan, confirm, and install a skill.
+
+    ``name_override`` lets non-interactive callers (slash commands, gateway,
+    scripts) supply a skill name when the upstream SKILL.md lacks a valid
+    ``name:`` frontmatter field. On interactive TTY surfaces, a missing name
+    triggers a prompt instead; ``skip_confirm=True`` means "non-interactive"
+    (so pair it with ``name_override`` when installing from a URL that has
+    no frontmatter).
+    """
     from tools.skills_hub import (
         GitHubAuth, create_source_router, ensure_hub_dirs,
         quarantine_bundle, install_from_quarantine, HubLockFile,
@@ -354,6 +461,58 @@ def do_install(identifier: str, category: str = "", force: bool = False,
             c.print()
         return
 
+    # URL-sourced skills may arrive with an empty name when SKILL.md has no
+    # ``name:`` in frontmatter AND the URL path doesn't yield a valid
+    # identifier. Resolve by (1) --name override, (2) interactive prompt on
+    # a TTY, (3) refuse with an actionable error on non-interactive surfaces.
+    bundle_meta = getattr(bundle, "metadata", {}) or {}
+    if bundle.source == "url" and (not bundle.name or bundle_meta.get("awaiting_name")):
+        if name_override and _is_valid_installed_skill_name(name_override):
+            bundle.name = name_override.strip()
+            bundle_meta["awaiting_name"] = False
+        elif name_override:
+            c.print(
+                f"[bold red]Invalid --name:[/] {name_override!r}. "
+                "Must be a lowercase identifier (letters, digits, hyphens, "
+                "underscores; starts with a letter).\n"
+            )
+            return
+        elif skip_confirm:
+            # Non-interactive surface (slash command / TUI / gateway). Can't
+            # prompt — emit an actionable error.
+            url = bundle_meta.get("url") or identifier
+            c.print(
+                f"[bold red]Cannot install from URL:[/] {url}\n"
+                "[yellow]The SKILL.md has no `name:` in its frontmatter, "
+                "and the URL path doesn't produce a valid identifier.[/]\n\n"
+                "Retry with an explicit name:\n"
+                f"  [bold]/skills install {url} --name <your-name>[/]\n"
+                f"  [bold]hermes skills install {url} --name <your-name>[/]\n\n"
+                "[dim]Or ask the SKILL.md's author to add a `name:` field to "
+                "its YAML frontmatter.[/]\n"
+            )
+            return
+        else:
+            # Interactive TTY — prompt.
+            url = bundle_meta.get("url") or identifier
+            chosen = _prompt_for_skill_name(c, url)
+            if not chosen:
+                c.print("[dim]Installation cancelled.[/]\n")
+                return
+            bundle.name = chosen
+            bundle_meta["awaiting_name"] = False
+        # Keep SkillMeta in sync so downstream "already installed" checks,
+        # audit logs, and display all see the final name.
+        if meta is not None:
+            meta.name = bundle.name
+            meta.path = bundle.name
+
+    # URL-sourced skills: offer to pick a category interactively when the
+    # caller didn't specify one (TTY only — non-interactive installs fall
+    # through to flat install, matching all other sources).
+    if bundle.source == "url" and not category and not skip_confirm:
+        category = _prompt_for_category(c, _existing_categories())
+
     # Auto-detect category for official skills (e.g. "official/autonomous-ai-agents/blackbox")
     if bundle.source == "official" and not category:
         id_parts = bundle.identifier.split("/")  # ["official", "category", "skill"]
@@ -599,11 +758,24 @@ def inspect_skill(identifier: str) -> Optional[dict]:
     return out
 
 
-def do_list(source_filter: str = "all", console: Optional[Console] = None) -> None:
-    """List installed skills, distinguishing hub, builtin, and local skills."""
+def do_list(source_filter: str = "all",
+            enabled_only: bool = False,
+            console: Optional[Console] = None) -> None:
+    """List installed skills, distinguishing hub, builtin, and local skills.
+
+    Args:
+        source_filter: ``all`` | ``hub`` | ``builtin`` | ``local``.
+        enabled_only: If True, hide disabled skills from the output.
+
+    Enabled/disabled state is resolved against the currently active profile's
+    config — ``hermes -p <profile> skills list`` reads that profile's
+    ``skills.disabled`` list because ``-p`` swaps ``HERMES_HOME`` at process
+    start.  No explicit profile flag needed here.
+    """
     from tools.skills_hub import HubLockFile, ensure_hub_dirs
     from tools.skills_sync import _read_manifest
     from tools.skills_tool import _find_all_skills
+    from agent.skill_utils import get_disabled_skill_names
 
     c = console or _console
     ensure_hub_dirs()
@@ -611,17 +783,26 @@ def do_list(source_filter: str = "all", console: Optional[Console] = None) -> No
     hub_installed = {e["name"]: e for e in lock.list_installed()}
     builtin_names = set(_read_manifest())
 
-    all_skills = _find_all_skills()
+    # Pull ALL skills (including disabled ones) so we can annotate status.
+    all_skills = _find_all_skills(skip_disabled=True)
+    disabled_names = get_disabled_skill_names()
 
-    table = Table(title="Installed Skills")
+    title = "Installed Skills"
+    if enabled_only:
+        title += " (enabled only)"
+
+    table = Table(title=title)
     table.add_column("Name", style="bold cyan")
     table.add_column("Category", style="dim")
     table.add_column("Source", style="dim")
     table.add_column("Trust", style="dim")
+    table.add_column("Status", style="dim")
 
     hub_count = 0
     builtin_count = 0
     local_count = 0
+    enabled_count = 0
+    disabled_count = 0
 
     for skill in sorted(all_skills, key=lambda s: (s.get("category") or "", s["name"])):
         name = skill["name"]
@@ -632,29 +813,48 @@ def do_list(source_filter: str = "all", console: Optional[Console] = None) -> No
             source_type = "hub"
             source_display = hub_entry.get("source", "hub")
             trust = hub_entry.get("trust_level", "community")
-            hub_count += 1
         elif name in builtin_names:
             source_type = "builtin"
             source_display = "builtin"
             trust = "builtin"
-            builtin_count += 1
         else:
             source_type = "local"
             source_display = "local"
             trust = "local"
-            local_count += 1
 
         if source_filter != "all" and source_filter != source_type:
             continue
 
+        is_enabled = name not in disabled_names
+        if enabled_only and not is_enabled:
+            continue
+
+        if source_type == "hub":
+            hub_count += 1
+        elif source_type == "builtin":
+            builtin_count += 1
+        else:
+            local_count += 1
+
+        if is_enabled:
+            enabled_count += 1
+            status_cell = "[bold green]enabled[/]"
+        else:
+            disabled_count += 1
+            status_cell = "[dim red]disabled[/]"
+
         trust_style = {"builtin": "bright_cyan", "trusted": "green", "community": "yellow", "local": "dim"}.get(trust, "dim")
         trust_label = "official" if source_display == "official" else trust
-        table.add_row(name, category, source_display, f"[{trust_style}]{trust_label}[/]")
+        table.add_row(name, category, source_display, f"[{trust_style}]{trust_label}[/]", status_cell)
 
     c.print(table)
-    c.print(
-        f"[dim]{hub_count} hub-installed, {builtin_count} builtin, {local_count} local[/]\n"
-    )
+    summary = f"[dim]{hub_count} hub-installed, {builtin_count} builtin, {local_count} local"
+    if enabled_only:
+        summary += f" — {enabled_count} enabled shown"
+    else:
+        summary += f" — {enabled_count} enabled, {disabled_count} disabled"
+    summary += "[/]\n"
+    c.print(summary)
 
 
 def do_check(name: Optional[str] = None, console: Optional[Console] = None) -> None:
@@ -1123,11 +1323,15 @@ def skills_command(args) -> None:
         do_search(args.query, source=args.source, limit=args.limit)
     elif action == "install":
         do_install(args.identifier, category=args.category, force=args.force,
-                   skip_confirm=getattr(args, "yes", False))
+                   skip_confirm=getattr(args, "yes", False),
+                   name_override=getattr(args, "name", "") or "")
     elif action == "inspect":
         do_inspect(args.identifier)
     elif action == "list":
-        do_list(source_filter=args.source)
+        do_list(
+            source_filter=args.source,
+            enabled_only=getattr(args, "enabled_only", False),
+        )
     elif action == "check":
         do_check(name=getattr(args, "name", None))
     elif action == "update":
@@ -1177,6 +1381,7 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
         /skills search kubernetes
         /skills install openai/skills/skill-creator
         /skills install openai/skills/skill-creator --force
+        /skills install https://example.com/path/SKILL.md
         /skills inspect openai/skills/skill-creator
         /skills list
         /skills list --source hub
@@ -1253,10 +1458,11 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
 
     elif action == "install":
         if not args:
-            c.print("[bold red]Usage:[/] /skills install <identifier> [--category <cat>] [--force] [--now]\n")
+            c.print("[bold red]Usage:[/] /skills install <identifier-or-url> [--name <name>] [--category <cat>] [--force] [--now]\n")
             return
         identifier = args[0]
         category = ""
+        name_override = ""
         # Slash commands run inside prompt_toolkit where input() hangs.
         # Always skip confirmation — the user typing the command is implicit consent.
         skip_confirm = True
@@ -1267,9 +1473,11 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
         for i, a in enumerate(args):
             if a == "--category" and i + 1 < len(args):
                 category = args[i + 1]
+            elif a == "--name" and i + 1 < len(args):
+                name_override = args[i + 1]
         do_install(identifier, category=category, force=force,
                    skip_confirm=skip_confirm, invalidate_cache=invalidate_cache,
-                   console=c)
+                   name_override=name_override, console=c)
 
     elif action == "inspect":
         if not args:
@@ -1279,11 +1487,12 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
 
     elif action == "list":
         source_filter = "all"
+        enabled_only = "--enabled-only" in args or "--enabled" in args
         if "--source" in args:
             idx = args.index("--source")
             if idx + 1 < len(args):
                 source_filter = args[idx + 1]
-        do_list(source_filter=source_filter, console=c)
+        do_list(source_filter=source_filter, enabled_only=enabled_only, console=c)
 
     elif action == "check":
         name = args[0] if args else None
@@ -1371,7 +1580,8 @@ def _print_skills_help(console: Console) -> None:
         "  [cyan]search[/] <query>              Search registries for skills\n"
         "  [cyan]install[/] <identifier>        Install a skill (with security scan)\n"
         "  [cyan]inspect[/] <identifier>        Preview a skill without installing\n"
-        "  [cyan]list[/] [--source hub|builtin|local] List installed skills\n"
+        "  [cyan]list[/] [--source hub|builtin|local] [--enabled-only]\n"
+        "       List installed skills; --enabled-only filters to the active profile's live set\n"
         "  [cyan]check[/] [name]                Check hub skills for upstream updates\n"
         "  [cyan]update[/] [name]               Update hub skills with upstream changes\n"
         "  [cyan]audit[/] [name]                Re-scan hub skills for security\n"

hermes_cli/slack_cli.py

@@ -0,0 +1,152 @@
+"""``hermes slack ...`` CLI subcommands.
+
+Today only ``hermes slack manifest`` is implemented — it generates the
+Slack app manifest JSON for registering every gateway command as a native
+Slack slash (``/btw``, ``/stop``, ``/model``, …) so users get the same
+first-class slash UX Discord and Telegram already have.
+
+Typical workflow::
+
+    $ hermes slack manifest > slack-manifest.json
+    # or:
+    $ hermes slack manifest --write
+
+Then paste the printed JSON into the Slack app config (Features → App
+Manifest → Edit) and click Save. Slack diffs the manifest and prompts
+for reinstall when scopes/commands change.
+"""
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+
+def _build_full_manifest(bot_name: str, bot_description: str) -> dict:
+    """Build a full Slack manifest merging display info + our slash list.
+
+    The slash-command list is always generated from ``COMMAND_REGISTRY`` so
+    it stays in sync with the rest of Hermes. Other manifest sections
+    (display info, OAuth scopes, socket mode) are set to sensible defaults
+    for a Hermes deployment — users can tweak them in the Slack UI after
+    pasting.
+    """
+    from hermes_cli.commands import slack_app_manifest
+
+    partial = slack_app_manifest()
+    slashes = partial["features"]["slash_commands"]
+
+    return {
+        "_metadata": {
+            "major_version": 1,
+            "minor_version": 1,
+        },
+        "display_information": {
+            "name": bot_name[:35],
+            "description": (bot_description or "Your Hermes agent on Slack")[:140],
+            "background_color": "#1a1a2e",
+        },
+        "features": {
+            "bot_user": {
+                "display_name": bot_name[:80],
+                "always_online": True,
+            },
+            "slash_commands": slashes,
+            "assistant_view": {
+                "assistant_description": "Chat with Hermes in threads and DMs.",
+            },
+        },
+        "oauth_config": {
+            "scopes": {
+                "bot": [
+                    "app_mentions:read",
+                    "assistant:write",
+                    "channels:history",
+                    "channels:read",
+                    "chat:write",
+                    "commands",
+                    "files:read",
+                    "files:write",
+                    "groups:history",
+                    "im:history",
+                    "im:read",
+                    "im:write",
+                    "users:read",
+                ],
+            },
+        },
+        "settings": {
+            "event_subscriptions": {
+                "bot_events": [
+                    "app_mention",
+                    "assistant_thread_context_changed",
+                    "assistant_thread_started",
+                    "message.channels",
+                    "message.groups",
+                    "message.im",
+                ],
+            },
+            "interactivity": {
+                "is_enabled": True,
+            },
+            "org_deploy_enabled": False,
+            "socket_mode_enabled": True,
+            "token_rotation_enabled": False,
+        },
+    }
+
+
+def slack_manifest_command(args) -> int:
+    """Print or write a Slack app manifest JSON.
+
+    Flags (all parsed in ``hermes_cli/main.py``):
+      --write [PATH]  Write to file instead of stdout (default path:
+                      ``$HERMES_HOME/slack-manifest.json``)
+      --name NAME     Override the bot display name (default: "Hermes")
+      --description DESC  Override the bot description
+      --slashes-only  Emit only the ``features.slash_commands`` array (for
+                      merging into an existing manifest manually)
+    """
+    name = getattr(args, "name", None) or "Hermes"
+    description = getattr(args, "description", None) or "Your Hermes agent on Slack"
+
+    if getattr(args, "slashes_only", False):
+        from hermes_cli.commands import slack_app_manifest
+
+        manifest = slack_app_manifest()["features"]["slash_commands"]
+    else:
+        manifest = _build_full_manifest(name, description)
+
+    payload = json.dumps(manifest, indent=2, ensure_ascii=False) + "\n"
+
+    write_target = getattr(args, "write", None)
+    if write_target is not None:
+        if isinstance(write_target, bool) and write_target:
+            # --write with no value → default location
+            try:
+                from hermes_constants import get_hermes_home
+
+                target = Path(get_hermes_home()) / "slack-manifest.json"
+            except Exception:
+                target = Path.home() / ".hermes" / "slack-manifest.json"
+        else:
+            target = Path(write_target).expanduser()
+        target.parent.mkdir(parents=True, exist_ok=True)
+        target.write_text(payload, encoding="utf-8")
+        print(f"Slack manifest written to: {target}", file=sys.stderr)
+        print(
+            "\nNext steps:\n"
+            "  1. Open https://api.slack.com/apps and pick your Hermes app\n"
+            "     (or create a new one: Create New App → From an app manifest).\n"
+            f"  2. Features → App Manifest → paste the contents of\n"
+            f"     {target}\n"
+            "  3. Save; Slack will prompt to reinstall the app if scopes or\n"
+            "     slash commands changed.\n"
+            "  4. Make sure Socket Mode is enabled and you have a bot token\n"
+            "     (xoxb-...) and app token (xapp-...) configured via\n"
+            "     `hermes setup`.\n",
+            file=sys.stderr,
+        )
+    else:
+        sys.stdout.write(payload)
+    return 0

hermes_cli/status.py

@@ -326,7 +326,8 @@ def show_status(args):
         "WeCom Callback": ("WECOM_CALLBACK_CORP_ID", None),
         "Weixin": ("WEIXIN_ACCOUNT_ID", "WEIXIN_HOME_CHANNEL"),
         "BlueBubbles": ("BLUEBUBBLES_SERVER_URL", "BLUEBUBBLES_HOME_CHANNEL"),
-        "QQBot": ("QQ_APP_ID", "QQBOT_HOME_CHANNEL"),
+        "QQBot": ("QQ_APP_ID", "QQ_HOME_CHANNEL"),
+        "Yuanbao": ("YUANBAO_APP_ID", "YUANBAO_HOME_CHANNEL"),
     }
     
     for name, (token_var, home_var) in platforms.items():

hermes_cli/timeouts.py

@@ -20,10 +20,10 @@ def get_provider_request_timeout(
 
     try:
         from hermes_cli.config import load_config
-    except ImportError:
+        config = load_config()
+    except Exception:
         return None
 
-    config = load_config()
     providers = config.get("providers", {}) if isinstance(config, dict) else {}
     provider_config = (
         providers.get(provider_id, {}) if isinstance(providers, dict) else {}
@@ -49,10 +49,10 @@ def get_provider_stale_timeout(
 
     try:
         from hermes_cli.config import load_config
-    except ImportError:
+        config = load_config()
+    except Exception:
         return None
 
-    config = load_config()
     providers = config.get("providers", {}) if isinstance(config, dict) else {}
     provider_config = (
         providers.get(provider_id, {}) if isinstance(providers, dict) else {}

hermes_cli/tips.py

@@ -10,8 +10,7 @@ import random
 
 TIPS = [
     # --- Slash Commands ---
-    "/btw <question> asks a quick side question without tools or history — great for clarifications.",
-    "/background <prompt> runs a task in a separate session while your current one stays free.",
+    "/background <prompt> (alias /bg or /btw) runs a task in a separate session while your current one stays free.",
     "/branch forks the current session so you can explore a different direction without losing progress.",
     "/compress manually compresses conversation context when things get long.",
     "/rollback lists filesystem checkpoints — restore files the agent modified to any prior state.",
@@ -107,7 +106,7 @@ TIPS = [
     "Set display.streaming: true to see tokens appear in real time as the model generates.",
     "Set display.show_reasoning: true to watch the model's chain-of-thought reasoning.",
     "Set display.compact: true to reduce whitespace in output for denser information.",
-    "Set display.busy_input_mode: queue to queue messages instead of interrupting the agent.",
+    "Set display.busy_input_mode: queue to queue messages instead of interrupting the agent, or steer to inject them mid-run via /steer.",
     "Set display.resume_display: minimal to skip the full conversation recap on session resume.",
     "Set compression.threshold: 0.50 to control when auto-compression fires (default: 50% of context).",
     "Set agent.max_turns: 200 to let the agent take more tool-calling steps per turn.",

hermes_cli/tools_config.py

@@ -11,6 +11,7 @@ the `platform_toolsets` key.
 
 import json as _json
 import logging
+import os
 import sys
 from pathlib import Path
 from typing import Dict, List, Optional, Set
@@ -25,7 +26,7 @@ from hermes_cli.nous_subscription import (
     get_nous_subscription_features,
 )
 from tools.tool_backend_helpers import fal_key_is_configured, managed_nous_tools_enabled
-from utils import base_url_hostname
+from utils import base_url_hostname, is_truthy_value
 
 logger = logging.getLogger(__name__)
 
@@ -68,25 +69,59 @@ CONFIGURABLE_TOOLSETS = [
     ("rl",              "🧪 RL Training",               "Tinker-Atropos training tools"),
     ("homeassistant",    "🏠 Home Assistant",           "smart home device control"),
     ("spotify",          "🎵 Spotify",                  "playback, search, playlists, library"),
+    ("discord",         "💬 Discord (read/participate)", "fetch messages, search members, create thread"),
+    ("discord_admin",   "🛡️  Discord Server Admin",    "list channels/roles, pin, assign roles"),
+    ("yuanbao",          "🤖 Yuanbao",                  "group info, member queries, DM"),
 ]
 
 # Toolsets that are OFF by default for new installs.
 # They're still in _HERMES_CORE_TOOLS (available at runtime if enabled),
 # but the setup checklist won't pre-select them for first-time users.
-_DEFAULT_OFF_TOOLSETS = {"moa", "homeassistant", "rl", "spotify"}
+_DEFAULT_OFF_TOOLSETS = {"moa", "homeassistant", "rl", "spotify", "discord", "discord_admin"}
+
+# Platform-scoped toolsets: only appear in the `hermes tools` checklist for
+# these platforms, and only resolve/save for these platforms.  A toolset
+# absent from this map is available on every platform (current behaviour).
+#
+# Use this for tools whose APIs only make sense on one platform (Discord
+# server admin, Slack workspace admin, etc.).  Keeps every other platform's
+# checklist from filling up with irrelevant toggles.
+_TOOLSET_PLATFORM_RESTRICTIONS: Dict[str, Set[str]] = {
+    "discord": {"discord"},
+    "discord_admin": {"discord"},
+}
+
+
+def _toolset_allowed_for_platform(ts_key: str, platform: str) -> bool:
+    """Return True if ``ts_key`` is configurable on ``platform``.
+
+    Toolsets without a restriction entry are allowed everywhere (the default).
+    """
+    allowed = _TOOLSET_PLATFORM_RESTRICTIONS.get(ts_key)
+    return allowed is None or platform in allowed
 
 
 def _get_effective_configurable_toolsets():
     """Return CONFIGURABLE_TOOLSETS + any plugin-provided toolsets.
 
     Plugin toolsets are appended at the end so they appear after the
-    built-in toolsets in the TUI checklist.
+    built-in toolsets in the TUI checklist. A plugin whose toolset key
+    already appears in ``CONFIGURABLE_TOOLSETS`` is skipped — bundled
+    plugins (e.g. ``plugins/spotify``) share their toolset key with the
+    built-in entry, and we want the built-in label/description to win.
+    Without the dedupe, ``hermes tools`` → "reconfigure existing" would
+    list the same toolset twice.
     """
     result = list(CONFIGURABLE_TOOLSETS)
+    seen = {ts_key for ts_key, _, _ in result}
     try:
         from hermes_cli.plugins import discover_plugins, get_plugin_toolsets
         discover_plugins()  # idempotent — ensures plugins are loaded
-        result.extend(get_plugin_toolsets())
+        for entry in get_plugin_toolsets():
+            if entry[0] in seen:
+                continue
+            seen.add(entry[0])
+            result.append(entry)
     except Exception:
         pass
     return result
@@ -591,7 +626,7 @@ def _get_platform_tools(
     include_default_mcp_servers: bool = True,
 ) -> Set[str]:
     """Resolve which individual toolset names are enabled for a platform."""
-    from toolsets import resolve_toolset
+    from toolsets import resolve_toolset, TOOLSETS
 
     platform_toolsets = config.get("platform_toolsets") or {}
     toolset_names = platform_toolsets.get(platform)
@@ -605,6 +640,8 @@ def _get_platform_tools(
     toolset_names = [str(ts) for ts in toolset_names]
 
     configurable_keys = {ts_key for ts_key, _, _ in CONFIGURABLE_TOOLSETS}
+    plugin_ts_keys = _get_plugin_toolset_keys()
+    platform_default_keys = {p["default_toolset"] for p in PLATFORMS.values()}
 
     # If the saved list contains any configurable keys directly, the user
     # has explicitly configured this platform — use direct membership.
@@ -614,7 +651,10 @@ def _get_platform_tools(
     has_explicit_config = any(ts in configurable_keys for ts in toolset_names)
 
     if has_explicit_config:
-        enabled_toolsets = {ts for ts in toolset_names if ts in configurable_keys}
+        enabled_toolsets = {
+            ts for ts in toolset_names
+            if ts in configurable_keys and _toolset_allowed_for_platform(ts, platform)
+        }
     else:
         # No explicit config — fall back to resolving composite toolset names
         # (e.g. "hermes-cli") to individual tool names and reverse-mapping.
@@ -624,14 +664,61 @@ def _get_platform_tools(
 
         enabled_toolsets = set()
         for ts_key, _, _ in CONFIGURABLE_TOOLSETS:
+            if not _toolset_allowed_for_platform(ts_key, platform):
+                continue
             ts_tools = set(resolve_toolset(ts_key))
             if ts_tools and ts_tools.issubset(all_tool_names):
                 enabled_toolsets.add(ts_key)
+
         default_off = set(_DEFAULT_OFF_TOOLSETS)
-        if platform in default_off:
+        # Legacy safety: if the platform's own name matches a default-off
+        # toolset (e.g. `homeassistant` platform + `homeassistant` toolset),
+        # keep that toolset enabled on first install.  Skip this dodge for
+        # platform-restricted toolsets — those are always opt-in even on
+        # their own platform (e.g. `discord` + `discord` should stay OFF).
+        if platform in default_off and platform not in _TOOLSET_PLATFORM_RESTRICTIONS:
             default_off.remove(platform)
+        # Home Assistant is already runtime-gated by its check_fn (requires
+        # HASS_TOKEN to register any tools). When a user has configured
+        # HASS_TOKEN, they've explicitly opted in — don't also strip it via
+        # _DEFAULT_OFF_TOOLSETS, which would silently drop HA from platforms
+        # (e.g. cron) that run through _get_platform_tools without an
+        # explicit saved toolset list. Without this, Norbert's HA cron jobs
+        # regressed after #14798 made cron honor per-platform tool config.
+        if "homeassistant" in default_off and os.getenv("HASS_TOKEN"):
+            default_off.remove("homeassistant")
         enabled_toolsets -= default_off
 
+    # Recover non-configurable platform toolsets (e.g. discord, feishu_doc,
+    # feishu_drive).  These are part of the platform's default composite but
+    # absent from CONFIGURABLE_TOOLSETS, so they can't appear in the TUI
+    # checklist or in a user-saved config.  Must run in BOTH branches —
+    # otherwise saving via `hermes tools` (which flips has_explicit_config
+    # to True) silently drops them.
+    platform_tool_universe = set(resolve_toolset(PLATFORMS[platform]["default_toolset"]))
+    configurable_tool_universe = set()
+    for ck in configurable_keys:
+        configurable_tool_universe.update(resolve_toolset(ck))
+    claimed = set()
+    for ts_key in enabled_toolsets:
+        claimed.update(resolve_toolset(ts_key))
+    skip = configurable_keys | plugin_ts_keys | platform_default_keys
+    skip |= {k for k in TOOLSETS if k.startswith("hermes-")}
+    skip |= set(_DEFAULT_OFF_TOOLSETS) - {platform}
+    for ts_key, ts_def in TOOLSETS.items():
+        if ts_key in skip:
+            continue
+        if ts_def.get("includes"):
+            continue
+        ts_tools = set(resolve_toolset(ts_key))
+        if not ts_tools or not ts_tools.issubset(platform_tool_universe):
+            continue
+        if ts_tools.issubset(configurable_tool_universe):
+            continue
+        if not ts_tools.issubset(claimed):
+            enabled_toolsets.add(ts_key)
+            claimed.update(ts_tools)
+
     # Plugin toolsets: enabled by default unless explicitly disabled, or
     # unless the toolset is in _DEFAULT_OFF_TOOLSETS (e.g. spotify —
     # shipped as a bundled plugin but user must opt in via `hermes tools`
@@ -639,7 +726,6 @@ def _get_platform_tools(
     # A plugin toolset is "known" for a platform once `hermes tools`
     # has been saved for that platform (tracked via known_plugin_toolsets).
     # Unknown plugins default to enabled; known-but-absent = disabled.
-    plugin_ts_keys = _get_plugin_toolset_keys()
     if plugin_ts_keys:
         known_map = config.get("known_plugin_toolsets", {})
         known_for_platform = set(known_map.get(platform, []))
@@ -657,7 +743,6 @@ def _get_platform_tools(
 
     # Preserve any explicit non-configurable toolset entries (for example,
     # custom toolsets or MCP server names saved in platform_toolsets).
-    platform_default_keys = {p["default_toolset"] for p in PLATFORMS.values()}
     explicit_passthrough = {
         ts
         for ts in toolset_names
@@ -703,6 +788,14 @@ def _save_platform_tools(config: dict, platform: str, enabled_toolset_keys: Set[
     """
     config.setdefault("platform_toolsets", {})
 
+    # Drop platform-scoped toolsets that don't apply here.  Prevents the
+    # "Configure all platforms" checklist (or a hand-edited config.yaml)
+    # from turning on, say, the `discord` toolset for Telegram.
+    enabled_toolset_keys = {
+        ts for ts in enabled_toolset_keys
+        if _toolset_allowed_for_platform(ts, platform)
+    }
+
     # Get the set of all configurable toolset keys (built-in + plugin)
     configurable_keys = {ts_key for ts_key, _, _ in CONFIGURABLE_TOOLSETS}
     plugin_keys = _get_plugin_toolset_keys()
@@ -717,6 +810,7 @@ def _save_platform_tools(config: dict, platform: str, enabled_toolset_keys: Set[
     existing_toolsets = config.get("platform_toolsets", {}).get(platform, [])
     if not isinstance(existing_toolsets, list):
         existing_toolsets = []
+    existing_toolsets = [str(ts) for ts in existing_toolsets]
 
     # Preserve any entries that are NOT configurable toolsets and NOT platform
     # defaults (i.e. only MCP server names should be preserved)
@@ -724,6 +818,11 @@ def _save_platform_tools(config: dict, platform: str, enabled_toolset_keys: Set[
         entry for entry in existing_toolsets
         if entry not in configurable_keys and entry not in platform_default_keys
     }
+    # Opening `hermes tools` is the user's opt-in to reconfigure tools, so treat
+    # saving from the picker as consent to clear the "no_mcp" sentinel. The
+    # picker has no checkbox for no_mcp, so without this users who once set it
+    # by hand could never re-enable MCP servers through the UI.
+    preserved_entries.discard("no_mcp")
 
     # Merge preserved entries with new enabled toolsets
     config["platform_toolsets"][platform] = sorted(enabled_toolset_keys | preserved_entries)
@@ -831,7 +930,7 @@ def _estimate_tool_tokens() -> Dict[str, int]:
     return _tool_token_cache
 
 
-def _prompt_toolset_checklist(platform_label: str, enabled: Set[str]) -> Set[str]:
+def _prompt_toolset_checklist(platform_label: str, enabled: Set[str], platform: str = "cli") -> Set[str]:
     """Multi-select checklist of toolsets. Returns set of selected toolset keys."""
     from hermes_cli.curses_ui import curses_checklist
     from toolsets import resolve_toolset
@@ -839,7 +938,12 @@ def _prompt_toolset_checklist(platform_label: str, enabled: Set[str]) -> Set[str
     # Pre-compute per-tool token counts (cached after first call).
     tool_tokens = _estimate_tool_tokens()
 
-    effective = _get_effective_configurable_toolsets()
+    effective_all = _get_effective_configurable_toolsets()
+    # Drop platform-scoped toolsets that don't apply to this platform.
+    effective = [
+        (k, l, d) for (k, l, d) in effective_all
+        if _toolset_allowed_for_platform(k, platform)
+    ]
 
     labels = []
     for ts_key, ts_label, ts_desc in effective:
@@ -1084,7 +1188,7 @@ def _is_provider_active(provider: dict, config: dict) -> bool:
                 configured_provider = image_cfg.get("provider")
                 if configured_provider not in (None, "", "fal"):
                     return False
-                if image_cfg.get("use_gateway") is False:
+                if image_cfg.get("use_gateway") is not None and not is_truthy_value(image_cfg.get("use_gateway"), default=False):
                     return False
             return feature.managed_by_nous
         if provider.get("tts_provider"):
@@ -1116,7 +1220,7 @@ def _is_provider_active(provider: dict, config: dict) -> bool:
         return (
             provider["imagegen_backend"] == "fal"
             and configured_provider in (None, "", "fal")
-            and not image_cfg.get("use_gateway")
+            and not is_truthy_value(image_cfg.get("use_gateway"), default=False)
         )
     return False
 
@@ -1753,7 +1857,7 @@ def tools_command(args=None, first_install: bool = False, config: dict = None):
             checklist_preselected = current_enabled - _DEFAULT_OFF_TOOLSETS
 
             # Show checklist
-            new_enabled = _prompt_toolset_checklist(pinfo["label"], checklist_preselected)
+            new_enabled = _prompt_toolset_checklist(pinfo["label"], checklist_preselected, pkey)
 
             added = new_enabled - current_enabled
             removed = current_enabled - new_enabled
@@ -2109,7 +2213,11 @@ def _apply_mcp_change(config: dict, targets: List[str], action: str) -> Set[str]
 
 def _print_tools_list(enabled_toolsets: set, mcp_servers: dict, platform: str = "cli"):
     """Print a summary of enabled/disabled toolsets and MCP tool filters."""
-    effective = _get_effective_configurable_toolsets()
+    effective_all = _get_effective_configurable_toolsets()
+    effective = [
+        (k, l, d) for (k, l, d) in effective_all
+        if _toolset_allowed_for_platform(k, platform)
+    ]
     builtin_keys = {ts_key for ts_key, _, _ in CONFIGURABLE_TOOLSETS}
 
     print(f"Built-in toolsets ({platform}):")
@@ -2175,6 +2283,20 @@ def tools_disable_enable_command(args):
             _print_error(f"Unknown toolset '{name}'")
         toolset_targets = [t for t in toolset_targets if t in valid_toolsets]
 
+    # Reject platform-scoped toolsets on platforms that don't allow them.
+    restricted_targets = [
+        t for t in toolset_targets
+        if not _toolset_allowed_for_platform(t, platform)
+    ]
+    if restricted_targets:
+        for name in restricted_targets:
+            allowed = sorted(_TOOLSET_PLATFORM_RESTRICTIONS.get(name) or set())
+            _print_error(
+                f"Toolset '{name}' is not available on platform '{platform}' "
+                f"(only: {', '.join(allowed)})"
+            )
+        toolset_targets = [t for t in toolset_targets if t not in restricted_targets]
+
     if toolset_targets:
         _apply_toolset_change(config, platform, toolset_targets, action)
 

hermes_cli/web_server.py

@@ -287,7 +287,7 @@ _SCHEMA_OVERRIDES: Dict[str, Dict[str, Any]] = {
     "display.busy_input_mode": {
         "type": "select",
         "description": "Input behavior while agent is running",
-        "options": ["interrupt", "queue"],
+        "options": ["interrupt", "queue", "steer"],
     },
     "memory.provider": {
         "type": "select",
@@ -2327,16 +2327,14 @@ def _resolve_chat_argv(
     from hermes_cli.main import PROJECT_ROOT, _make_tui_argv
 
     argv, cwd = _make_tui_argv(PROJECT_ROOT / "ui-tui", tui_dev=False)
-    env: Optional[dict] = None
+    env = os.environ.copy()
+    env.setdefault("NODE_ENV", "production")
 
-    if resume or sidecar_url:
-        env = os.environ.copy()
+    if resume:
+        env["HERMES_TUI_RESUME"] = resume
 
-        if resume:
-            env["HERMES_TUI_RESUME"] = resume
-
-        if sidecar_url:
-            env["HERMES_TUI_SIDECAR_URL"] = sidecar_url
+    if sidecar_url:
+        env["HERMES_TUI_SIDECAR_URL"] = sidecar_url
 
     return list(argv), str(cwd) if cwd else None, env
 

hermes_logging.py

@@ -195,10 +195,6 @@ def setup_logging(
         The ``logs/`` directory where files are written.
     """
     global _logging_initialized
-    if _logging_initialized and not force:
-        home = hermes_home or get_hermes_home()
-        return home / "logs"
-
     home = hermes_home or get_hermes_home()
     log_dir = home / "logs"
     log_dir.mkdir(parents=True, exist_ok=True)
@@ -248,6 +244,9 @@ def setup_logging(
             log_filter=_ComponentFilter(COMPONENT_PREFIXES["gateway"]),
         )
 
+    if _logging_initialized and not force:
+        return log_dir
+
     # Ensure root logger level is low enough for the handlers to fire.
     if root.level == logging.NOTSET or root.level > level:
         root.setLevel(level)

hermes_state.py

@@ -31,7 +31,7 @@ T = TypeVar("T")
 
 DEFAULT_DB_PATH = get_hermes_home() / "state.db"
 
-SCHEMA_VERSION = 8
+SCHEMA_VERSION = 9
 
 SCHEMA_SQL = """
 CREATE TABLE IF NOT EXISTS schema_version (
@@ -83,7 +83,8 @@ CREATE TABLE IF NOT EXISTS messages (
     reasoning TEXT,
     reasoning_content TEXT,
     reasoning_details TEXT,
-    codex_reasoning_items TEXT
+    codex_reasoning_items TEXT,
+    codex_message_items TEXT
 );
 
 CREATE TABLE IF NOT EXISTS state_meta (
@@ -356,6 +357,15 @@ class SessionDB:
                 except sqlite3.OperationalError:
                     pass  # Column already exists
                 cursor.execute("UPDATE schema_version SET version = 8")
+            if current_version < 9:
+                # v9: preserve replayable Codex assistant message ids/phases so
+                # follow-up turns can rebuild Responses API message items instead
+                # of flattening everything to plain assistant text.
+                try:
+                    cursor.execute('ALTER TABLE messages ADD COLUMN "codex_message_items" TEXT')
+                except sqlite3.OperationalError:
+                    pass  # Column already exists
+                cursor.execute("UPDATE schema_version SET version = 9")
 
         # Unique title index — always ensure it exists (safe to run after migrations
         # since the title column is guaranteed to exist at this point)
@@ -822,7 +832,18 @@ class SessionDB:
         params = []
 
         if not include_children:
-            where_clauses.append("s.parent_session_id IS NULL")
+            # Show root sessions and branch sessions (whose parent ended with
+            # end_reason='branched' before the child was created), while still
+            # hiding sub-agent runs and compression continuations (which also
+            # carry a parent_session_id but were spawned while the parent was
+            # still live — i.e., started_at < parent.ended_at).
+            where_clauses.append(
+                "(s.parent_session_id IS NULL"
+                " OR EXISTS (SELECT 1 FROM sessions p"
+                "            WHERE p.id = s.parent_session_id"
+                "            AND p.end_reason = 'branched'"
+                "            AND s.started_at >= p.ended_at))"
+            )
 
         if source:
             where_clauses.append("s.source = ?")
@@ -956,6 +977,7 @@ class SessionDB:
         reasoning_content: str = None,
         reasoning_details: Any = None,
         codex_reasoning_items: Any = None,
+        codex_message_items: Any = None,
     ) -> int:
         """
         Append a message to a session. Returns the message row ID.
@@ -972,6 +994,10 @@ class SessionDB:
             json.dumps(codex_reasoning_items)
             if codex_reasoning_items else None
         )
+        codex_message_items_json = (
+            json.dumps(codex_message_items)
+            if codex_message_items else None
+        )
         tool_calls_json = json.dumps(tool_calls) if tool_calls else None
 
         # Pre-compute tool call count
@@ -983,8 +1009,9 @@ class SessionDB:
             cursor = conn.execute(
                 """INSERT INTO messages (session_id, role, content, tool_call_id,
                    tool_calls, tool_name, timestamp, token_count, finish_reason,
-                   reasoning, reasoning_content, reasoning_details, codex_reasoning_items)
-                   VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+                   reasoning, reasoning_content, reasoning_details, codex_reasoning_items,
+                   codex_message_items)
+                   VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
                 (
                     session_id,
                     role,
@@ -999,6 +1026,7 @@ class SessionDB:
                     reasoning_content,
                     reasoning_details_json,
                     codex_items_json,
+                    codex_message_items_json,
                 ),
             )
             msg_id = cursor.lastrowid
@@ -1104,19 +1132,27 @@ 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(
+            placeholders = ",".join("?" for _ in session_ids)
+            rows = self._conn.execute(
                 "SELECT role, content, tool_call_id, tool_calls, tool_name, "
-                "reasoning, reasoning_content, reasoning_details, codex_reasoning_items "
-                "FROM messages WHERE session_id = ? ORDER BY timestamp, id",
-                (session_id,),
-            )
-            rows = cursor.fetchall()
+                "reasoning, reasoning_content, reasoning_details, codex_reasoning_items, "
+                "codex_message_items "
+                f"FROM messages WHERE session_id IN ({placeholders}) ORDER BY timestamp, id",
+                tuple(session_ids),
+            ).fetchall()
+
         messages = []
         for row in rows:
             msg = {"role": row["role"], "content": row["content"]}
@@ -1150,9 +1186,53 @@ class SessionDB:
                     except (json.JSONDecodeError, TypeError):
                         logger.warning("Failed to deserialize codex_reasoning_items, falling back to None")
                         msg["codex_reasoning_items"] = None
+                if row["codex_message_items"]:
+                    try:
+                        msg["codex_message_items"] = json.loads(row["codex_message_items"])
+                    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
     # =========================================================================
@@ -1477,12 +1557,45 @@ class SessionDB:
             )
         self._execute_write(_do)
 
-    def delete_session(self, session_id: str) -> bool:
+    @staticmethod
+    def _remove_session_files(sessions_dir: Optional[Path], session_id: str) -> None:
+        """Remove on-disk transcript files for a session.
+
+        Cleans up ``{session_id}.json``, ``{session_id}.jsonl``, and any
+        ``request_dump_{session_id}_*.json`` files left by the gateway.
+        Silently skips files that don't exist and swallows OSError so a
+        filesystem hiccup never blocks a DB operation.
+        """
+        if sessions_dir is None:
+            return
+        for suffix in (".json", ".jsonl"):
+            p = sessions_dir / f"{session_id}{suffix}"
+            try:
+                p.unlink(missing_ok=True)
+            except OSError:
+                pass
+        # request_dump files use session_id as a prefix component
+        try:
+            for p in sessions_dir.glob(f"request_dump_{session_id}_*.json"):
+                try:
+                    p.unlink(missing_ok=True)
+                except OSError:
+                    pass
+        except OSError:
+            pass
+
+    def delete_session(
+        self,
+        session_id: str,
+        sessions_dir: Optional[Path] = None,
+    ) -> bool:
         """Delete a session and all its messages.
 
         Child sessions are orphaned (parent_session_id set to NULL) rather
         than cascade-deleted, so they remain accessible independently.
-        Returns True if the session was found and deleted.
+        When *sessions_dir* is provided, also removes on-disk transcript
+        files (``.json`` / ``.jsonl`` / ``request_dump_*``) for the deleted
+        session. Returns True if the session was found and deleted.
         """
         def _do(conn):
             cursor = conn.execute(
@@ -1499,16 +1612,29 @@ class SessionDB:
             conn.execute("DELETE FROM messages WHERE session_id = ?", (session_id,))
             conn.execute("DELETE FROM sessions WHERE id = ?", (session_id,))
             return True
-        return self._execute_write(_do)
 
-    def prune_sessions(self, older_than_days: int = 90, source: str = None) -> int:
+        deleted = self._execute_write(_do)
+        if deleted:
+            self._remove_session_files(sessions_dir, session_id)
+        return deleted
+
+    def prune_sessions(
+        self,
+        older_than_days: int = 90,
+        source: str = None,
+        sessions_dir: Optional[Path] = None,
+    ) -> int:
         """Delete sessions older than N days. Returns count of deleted sessions.
 
         Only prunes ended sessions (not active ones).  Child sessions outside
         the prune window are orphaned (parent_session_id set to NULL) rather
-        than cascade-deleted.
+        than cascade-deleted.  When *sessions_dir* is provided, also removes
+        on-disk transcript files (``.json`` / ``.jsonl`` /
+        ``request_dump_*``) for every pruned session, outside the DB
+        transaction.
         """
         cutoff = time.time() - (older_than_days * 86400)
+        removed_ids: list[str] = []
 
         def _do(conn):
             if source:
@@ -1538,9 +1664,14 @@ class SessionDB:
             for sid in session_ids:
                 conn.execute("DELETE FROM messages WHERE session_id = ?", (sid,))
                 conn.execute("DELETE FROM sessions WHERE id = ?", (sid,))
+                removed_ids.append(sid)
             return len(session_ids)
 
-        return self._execute_write(_do)
+        count = self._execute_write(_do)
+        # Clean up on-disk files outside the DB transaction
+        for sid in removed_ids:
+            self._remove_session_files(sessions_dir, sid)
+        return count
 
     # ── Meta key/value (for scheduler bookkeeping) ──
 
@@ -1594,6 +1725,7 @@ class SessionDB:
         retention_days: int = 90,
         min_interval_hours: int = 24,
         vacuum: bool = True,
+        sessions_dir: Optional[Path] = None,
     ) -> Dict[str, Any]:
         """Idempotent auto-maintenance: prune old sessions + optional VACUUM.
 
@@ -1601,6 +1733,10 @@ class SessionDB:
         within ``min_interval_hours`` no-op. Designed to be called once at
         startup from long-lived entrypoints (CLI, gateway, cron scheduler).
 
+        When *sessions_dir* is provided, on-disk transcript files
+        (``.json`` / ``.jsonl`` / ``request_dump_*``) for pruned sessions
+        are removed as part of the same sweep (issue #3015).
+
         Never raises. On any failure, logs a warning and returns a dict
         with ``"error"`` set.
 
@@ -1624,7 +1760,10 @@ class SessionDB:
                 except (TypeError, ValueError):
                     pass  # corrupt meta; treat as no prior run
 
-            pruned = self.prune_sessions(older_than_days=retention_days)
+            pruned = self.prune_sessions(
+                older_than_days=retention_days,
+                sessions_dir=sessions_dir,
+            )
             result["pruned"] = pruned
 
             # Only VACUUM if we actually freed rows — VACUUM on a tight DB

model_tools.py

@@ -24,6 +24,7 @@ import json
 import asyncio
 import logging
 import threading
+import time
 from typing import Dict, Any, List, Optional, Tuple
 
 from tools.registry import discover_builtin_tools, registry
@@ -288,30 +289,34 @@ def get_tool_definitions(
                 filtered_tools[i] = {"type": "function", "function": dynamic_schema}
                 break
 
-    # Rebuild discord_server schema based on the bot's privileged intents
-    # (detected from GET /applications/@me) and the user's action allowlist
-    # in config.  Hides actions the bot's intents don't support so the
-    # model never attempts them, and annotates fetch_messages when the
+    # Rebuild discord / discord_admin schemas based on the bot's privileged
+    # intents (detected from GET /applications/@me) and the user's action
+    # allowlist in config.  Hides actions the bot's intents don't support so
+    # the model never attempts them, and annotates fetch_messages when the
     # MESSAGE_CONTENT intent is missing.
-    if "discord_server" in available_tool_names:
-        try:
-            from tools.discord_tool import get_dynamic_schema
-            dynamic = get_dynamic_schema()
-        except Exception:  # pragma: no cover — defensive, fall back to static
-            dynamic = None
-        if dynamic is None:
-            # Tool filtered out entirely (empty allowlist or detection disabled
-            # the only remaining actions).  Drop it from the schema list.
-            filtered_tools = [
-                t for t in filtered_tools
-                if t.get("function", {}).get("name") != "discord_server"
-            ]
-            available_tool_names.discard("discord_server")
-        else:
-            for i, td in enumerate(filtered_tools):
-                if td.get("function", {}).get("name") == "discord_server":
-                    filtered_tools[i] = {"type": "function", "function": dynamic}
-                    break
+    _discord_schema_fns = {
+        "discord": "get_dynamic_schema_core",
+        "discord_admin": "get_dynamic_schema_admin",
+    }
+    for discord_tool_name in _discord_schema_fns:
+        if discord_tool_name in available_tool_names:
+            try:
+                from tools import discord_tool as _dt
+                schema_fn = getattr(_dt, _discord_schema_fns[discord_tool_name])
+                dynamic = schema_fn()
+            except Exception:
+                dynamic = None
+            if dynamic is None:
+                filtered_tools = [
+                    t for t in filtered_tools
+                    if t.get("function", {}).get("name") != discord_tool_name
+                ]
+                available_tool_names.discard(discord_tool_name)
+            else:
+                for i, td in enumerate(filtered_tools):
+                    if td.get("function", {}).get("name") == discord_tool_name:
+                        filtered_tools[i] = {"type": "function", "function": dynamic}
+                        break
 
     # Strip web tool cross-references from browser_navigate description when
     # web_search / web_extract are not available.  The static schema says
@@ -563,6 +568,14 @@ def handle_function_call(
             except Exception:
                 pass  # file_tools may not be loaded yet
 
+        # Measure tool dispatch latency so post_tool_call and
+        # transform_tool_result hooks can observe per-tool duration.
+        # Inspired by Claude Code 2.1.119, which added ``duration_ms`` to
+        # PostToolUse hook inputs so plugin authors can build latency
+        # dashboards, budget alerts, and regression canaries without having
+        # to wrap every tool manually.  We use monotonic() so the value is
+        # unaffected by wall-clock adjustments during the call.
+        _dispatch_start = time.monotonic()
         if function_name == "execute_code":
             # Prefer the caller-provided list so subagents can't overwrite
             # the parent's tool set via the process-global.
@@ -578,6 +591,7 @@ def handle_function_call(
                 task_id=task_id,
                 user_task=user_task,
             )
+        duration_ms = int((time.monotonic() - _dispatch_start) * 1000)
 
         try:
             from hermes_cli.plugins import invoke_hook
@@ -589,6 +603,7 @@ def handle_function_call(
                 task_id=task_id or "",
                 session_id=session_id or "",
                 tool_call_id=tool_call_id or "",
+                duration_ms=duration_ms,
             )
         except Exception:
             pass
@@ -609,6 +624,7 @@ def handle_function_call(
                 task_id=task_id or "",
                 session_id=session_id or "",
                 tool_call_id=tool_call_id or "",
+                duration_ms=duration_ms,
             )
             for hook_result in hook_results:
                 if isinstance(hook_result, str):

nix/tui.nix

@@ -4,7 +4,7 @@ let
   src = ../ui-tui;
   npmDeps = pkgs.fetchNpmDeps {
     inherit src;
-    hash = "sha256-RU4qSHgJPMyfRSEJDzkG4+MReDZDc6QbTD2wisa5QE0=";
+    hash = "sha256-Chz+NW9NXqboXHOa6PKwf5bhAkkcFtKNhvKWwg2XSPc=";
   };
 
   npm = hermesNpmLib.mkNpmPassthru { folder = "ui-tui"; attr = "tui"; pname = "hermes-tui"; };

optional-skills/migration/openclaw-migration/scripts/openclaw_to_hermes.py

@@ -380,6 +380,10 @@ def backup_existing(path: Path, backup_root: Path) -> Optional[Path]:
 # Replace OpenClaw brand names with Hermes in migrated text so that
 # memory entries, user profiles, SOUL.md, and workspace instructions
 # read as self-referential to the new agent identity.
+#
+# Case-preserving: ``OpenClaw`` → ``Hermes`` (prose), but lowercase matches
+# like ``openclaw`` → ``hermes`` (so filesystem paths like ``~/.openclaw``
+# become ``~/.hermes`` — the real Hermes home — not the broken ``~/.Hermes``).
 _REBRAND_PATTERNS: List[Tuple[re.Pattern, str]] = [
     (re.compile(r'\bOpen[\s-]?Claw\b', re.IGNORECASE), 'Hermes'),
     (re.compile(r'\bClawdBot\b', re.IGNORECASE), 'Hermes'),
@@ -387,10 +391,31 @@ _REBRAND_PATTERNS: List[Tuple[re.Pattern, str]] = [
 ]
 
 
+def _case_preserving_replacement(replacement: str):
+    """Return a re.sub replacement fn that lowercases the result when the
+    matched text was all-lowercase.
+
+    Keeps ``OpenClaw`` → ``Hermes`` but maps ``openclaw`` → ``hermes`` so a
+    filesystem path like ``~/.openclaw/config.yaml`` rewrites to
+    ``~/.hermes/config.yaml`` (the real Hermes home) instead of the broken
+    ``~/.Hermes/config.yaml``.
+    """
+    def _sub(match: "re.Match[str]") -> str:
+        matched = match.group(0)
+        if matched and matched.islower():
+            return replacement.lower()
+        return replacement
+    return _sub
+
+
 def rebrand_text(text: str) -> str:
-    """Replace OpenClaw / ClawdBot / MoltBot brand names with Hermes."""
+    """Replace OpenClaw / ClawdBot / MoltBot brand names with Hermes.
+
+    Preserves case so filesystem-path matches (lowercase) don't become
+    capitalized directory names that don't exist.
+    """
     for pattern, replacement in _REBRAND_PATTERNS:
-        text = pattern.sub(replacement, text)
+        text = pattern.sub(_case_preserving_replacement(replacement), text)
     return text
 
 

plugins/example-dashboard/dashboard/dist/index.js

@@ -91,4 +91,29 @@
 
   // Register this plugin — the dashboard picks it up automatically.
   window.__HERMES_PLUGINS__.register("example", ExamplePage);
+
+  // ─────────────────────────────────────────────────────────────────────
+  // Page-scoped slot demo: inject a small banner at the top of /sessions.
+  //
+  // Built-in pages expose named slots (<page>:top, <page>:bottom) that
+  // plugins can populate without overriding the whole route. The
+  // manifest lists the slots we use in its `slots` array so the shell
+  // knows to render <PluginSlot name="sessions:top" /> there.
+  // ─────────────────────────────────────────────────────────────────────
+  function SessionsTopBanner() {
+    return React.createElement(Card, {
+      className: "border-dashed",
+    },
+      React.createElement(CardContent, { className: "flex items-center gap-3 py-2" },
+        React.createElement(Badge, { variant: "outline" }, "Example"),
+        React.createElement("span", {
+          className: "text-xs text-muted-foreground",
+        }, "This banner was injected into the Sessions page by the example plugin via the ",
+          React.createElement("code", { className: "font-courier" }, "sessions:top"),
+          " slot."),
+      ),
+    );
+  }
+
+  window.__HERMES_PLUGINS__.registerSlot("example", "sessions:top", SessionsTopBanner);
 })();

plugins/example-dashboard/dashboard/manifest.json

@@ -8,6 +8,7 @@
     "path": "/example",
     "position": "after:skills"
   },
+  "slots": ["sessions:top"],
   "entry": "dist/index.js",
   "api": "plugin_api.py"
 }

plugins/memory/hindsight/__init__.py

@@ -3,7 +3,9 @@
 Long-term memory with knowledge graph, entity resolution, and multi-strategy
 retrieval. Supports cloud (API key) and local modes.
 
-Configurable timeout via HINDSIGHT_TIMEOUT env var or config.json.
+Configurable request timeout via HINDSIGHT_TIMEOUT env var or config.json.
+Configurable embedded daemon idle timeout via HINDSIGHT_IDLE_TIMEOUT env var
+or config.json idle_timeout.
 
 Original PR #1811 by benfrank241, adapted to MemoryProvider ABC.
 
@@ -14,6 +16,7 @@ Config via environment variables:
   HINDSIGHT_API_URL                — API endpoint
   HINDSIGHT_MODE                   — cloud or local (default: cloud)
   HINDSIGHT_TIMEOUT                — API request timeout in seconds (default: 120)
+  HINDSIGHT_IDLE_TIMEOUT           — embedded daemon idle timeout seconds; 0 disables shutdown (default: 300)
   HINDSIGHT_RETAIN_TAGS            — comma-separated tags attached to retained memories
   HINDSIGHT_RETAIN_SOURCE          — metadata source value attached to retained memories
   HINDSIGHT_RETAIN_USER_PREFIX     — label used before user turns in retained transcripts
@@ -45,6 +48,7 @@ _DEFAULT_API_URL = "https://api.hindsight.vectorize.io"
 _DEFAULT_LOCAL_URL = "http://localhost:8888"
 _MIN_CLIENT_VERSION = "0.4.22"
 _DEFAULT_TIMEOUT = 120  # seconds — cloud API can take 30-40s per request
+_DEFAULT_IDLE_TIMEOUT = 300  # seconds — Hindsight embedded daemon default
 _VALID_BUDGETS = {"low", "mid", "high"}
 _PROVIDER_DEFAULT_MODELS = {
     "openai": "gpt-4o-mini",
@@ -59,6 +63,17 @@ _PROVIDER_DEFAULT_MODELS = {
 }
 
 
+def _parse_int_setting(value: Any, default: int) -> int:
+    """Parse an integer config/env value, falling back on invalid input."""
+    if value is None or value == "":
+        return default
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        logger.warning("Invalid integer Hindsight setting %r; using default %s", value, default)
+        return default
+
+
 def _check_local_runtime() -> tuple[bool, str | None]:
     """Return whether local embedded Hindsight imports cleanly.
 
@@ -203,6 +218,8 @@ def _load_config() -> dict:
     return {
         "mode": os.environ.get("HINDSIGHT_MODE", "cloud"),
         "apiKey": os.environ.get("HINDSIGHT_API_KEY", ""),
+        "timeout": _parse_int_setting(os.environ.get("HINDSIGHT_TIMEOUT"), _DEFAULT_TIMEOUT),
+        "idle_timeout": _parse_int_setting(os.environ.get("HINDSIGHT_IDLE_TIMEOUT"), _DEFAULT_IDLE_TIMEOUT),
         "retain_tags": os.environ.get("HINDSIGHT_RETAIN_TAGS", ""),
         "retain_source": os.environ.get("HINDSIGHT_RETAIN_SOURCE", ""),
         "retain_user_prefix": os.environ.get("HINDSIGHT_RETAIN_USER_PREFIX", "User"),
@@ -304,6 +321,16 @@ def _build_embedded_profile_env(config: dict[str, Any], *, llm_api_key: str | No
     }
     if current_base_url:
         env_values["HINDSIGHT_API_LLM_BASE_URL"] = str(current_base_url)
+
+    idle_timeout = (
+        config.get("idle_timeout")
+        if config.get("idle_timeout") is not None
+        else os.environ.get("HINDSIGHT_IDLE_TIMEOUT")
+    )
+    if idle_timeout is not None and idle_timeout != "":
+        env_values["HINDSIGHT_EMBED_DAEMON_IDLE_TIMEOUT"] = str(
+            _parse_int_setting(idle_timeout, _DEFAULT_IDLE_TIMEOUT)
+        )
     return env_values
 
 
@@ -412,6 +439,7 @@ class HindsightMemoryProvider(MemoryProvider):
         self._turn_index = 0
         self._client = None
         self._timeout = _DEFAULT_TIMEOUT
+        self._idle_timeout = _DEFAULT_IDLE_TIMEOUT
         self._prefetch_result = ""
         self._prefetch_lock = threading.Lock()
         self._prefetch_thread = None
@@ -592,10 +620,17 @@ class HindsightMemoryProvider(MemoryProvider):
             sys.stdout.write("  LLM API key: ")
             sys.stdout.flush()
             llm_key = getpass.getpass(prompt="") if sys.stdin.isatty() else sys.stdin.readline().strip()
-            # Always write explicitly (including empty) so the provider sees ""
-            # rather than a missing variable.  The daemon reads from .env at
-            # startup and fails when HINDSIGHT_LLM_API_KEY is unset.
-            env_writes["HINDSIGHT_LLM_API_KEY"] = llm_key
+            if llm_key:
+                env_writes["HINDSIGHT_LLM_API_KEY"] = llm_key
+            else:
+                env_path = Path(hermes_home) / ".env"
+                existing_llm_key = ""
+                if env_path.exists():
+                    for line in env_path.read_text().splitlines():
+                        if line.startswith("HINDSIGHT_LLM_API_KEY="):
+                            existing_llm_key = line.split("=", 1)[1]
+                            break
+                env_writes["HINDSIGHT_LLM_API_KEY"] = existing_llm_key
 
         # Step 4: Save everything
         provider_config["bank_id"] = "hermes"
@@ -605,6 +640,11 @@ class HindsightMemoryProvider(MemoryProvider):
         timeout_val = existing_timeout if existing_timeout else _DEFAULT_TIMEOUT
         provider_config["timeout"] = timeout_val
         env_writes["HINDSIGHT_TIMEOUT"] = str(timeout_val)
+        if mode == "local_embedded":
+            existing_idle_timeout = self._config.get("idle_timeout") if self._config else None
+            idle_timeout_val = existing_idle_timeout if existing_idle_timeout is not None else _DEFAULT_IDLE_TIMEOUT
+            provider_config["idle_timeout"] = idle_timeout_val
+            env_writes["HINDSIGHT_IDLE_TIMEOUT"] = str(idle_timeout_val)
         config["memory"]["provider"] = "hindsight"
         save_config(config)
 
@@ -693,6 +733,7 @@ class HindsightMemoryProvider(MemoryProvider):
             {"key": "recall_max_input_chars", "description": "Maximum input query length for auto-recall", "default": 800},
             {"key": "recall_prompt_preamble", "description": "Custom preamble for recalled memories in context"},
             {"key": "timeout", "description": "API request timeout in seconds", "default": _DEFAULT_TIMEOUT},
+            {"key": "idle_timeout", "description": "Embedded daemon idle timeout in seconds (0 disables auto-shutdown)", "default": _DEFAULT_IDLE_TIMEOUT, "when": {"mode": "local_embedded"}},
         ]
 
     def _get_client(self):
@@ -720,6 +761,14 @@ class HindsightMemoryProvider(MemoryProvider):
                 )
                 if self._llm_base_url:
                     kwargs["llm_base_url"] = self._llm_base_url
+                idle_timeout = _parse_int_setting(
+                    self._config.get("idle_timeout")
+                    if self._config.get("idle_timeout") is not None
+                    else os.environ.get("HINDSIGHT_IDLE_TIMEOUT", self._idle_timeout),
+                    _DEFAULT_IDLE_TIMEOUT,
+                )
+                self._idle_timeout = idle_timeout
+                kwargs["idle_timeout"] = idle_timeout
                 self._client = HindsightEmbedded(**kwargs)
             else:
                 from hindsight_client import Hindsight
@@ -736,6 +785,38 @@ class HindsightMemoryProvider(MemoryProvider):
         """Schedule *coro* on the shared loop using the configured timeout."""
         return _run_sync(coro, timeout=self._timeout)
 
+    def _is_retriable_embedded_connection_error(self, exc: Exception) -> bool:
+        """Return True for stale embedded-daemon connection failures."""
+        if self._mode != "local_embedded":
+            return False
+        text = f"{type(exc).__name__}: {exc}".lower()
+        return any(
+            marker in text
+            for marker in (
+                "cannot connect to host",
+                "connection refused",
+                "connect call failed",
+                "clientconnectorerror",
+            )
+        )
+
+    def _run_hindsight_operation(self, operation):
+        """Run an async Hindsight client operation, retrying once after idle shutdown."""
+        client = self._get_client()
+        try:
+            return self._run_sync(operation(client))
+        except Exception as exc:
+            if not self._is_retriable_embedded_connection_error(exc):
+                raise
+            logger.info(
+                "Hindsight embedded daemon appears unreachable; recreating client and retrying once: %s",
+                exc,
+            )
+            self._client = None
+            client = self._get_client()
+            self._client = client
+            return self._run_sync(operation(client))
+
     def initialize(self, session_id: str, **kwargs) -> None:
         self._session_id = str(session_id or "").strip()
         self._parent_session_id = str(kwargs.get("parent_session_id", "") or "").strip()
@@ -790,7 +871,14 @@ class HindsightMemoryProvider(MemoryProvider):
         self._session_turns = []
         self._mode = self._config.get("mode", "cloud")
         # Read timeout from config or env var, fall back to default
-        self._timeout = self._config.get("timeout") or int(os.environ.get("HINDSIGHT_TIMEOUT", str(_DEFAULT_TIMEOUT)))
+        self._timeout = _parse_int_setting(
+            self._config.get("timeout") if self._config.get("timeout") is not None else os.environ.get("HINDSIGHT_TIMEOUT"),
+            _DEFAULT_TIMEOUT,
+        )
+        self._idle_timeout = _parse_int_setting(
+            self._config.get("idle_timeout") if self._config.get("idle_timeout") is not None else os.environ.get("HINDSIGHT_IDLE_TIMEOUT"),
+            _DEFAULT_IDLE_TIMEOUT,
+        )
         # "local" is a legacy alias for "local_embedded"
         if self._mode == "local":
             self._mode = "local_embedded"
@@ -981,10 +1069,9 @@ class HindsightMemoryProvider(MemoryProvider):
 
         def _run():
             try:
-                client = self._get_client()
                 if self._prefetch_method == "reflect":
                     logger.debug("Prefetch: calling reflect (bank=%s, query_len=%d)", self._bank_id, len(query))
-                    resp = self._run_sync(client.areflect(bank_id=self._bank_id, query=query, budget=self._budget))
+                    resp = self._run_hindsight_operation(lambda client: client.areflect(bank_id=self._bank_id, query=query, budget=self._budget))
                     text = resp.text or ""
                 else:
                     recall_kwargs: dict = {
@@ -998,7 +1085,7 @@ class HindsightMemoryProvider(MemoryProvider):
                         recall_kwargs["types"] = self._recall_types
                     logger.debug("Prefetch: calling recall (bank=%s, query_len=%d, budget=%s)",
                                  self._bank_id, len(query), self._budget)
-                    resp = self._run_sync(client.arecall(**recall_kwargs))
+                    resp = self._run_hindsight_operation(lambda client: client.arecall(**recall_kwargs))
                     num_results = len(resp.results) if resp.results else 0
                     logger.debug("Prefetch: recall returned %d results", num_results)
                     text = "\n".join(f"- {r.text}" for r in resp.results if r.text) if resp.results else ""
@@ -1131,12 +1218,14 @@ class HindsightMemoryProvider(MemoryProvider):
                 item.pop("retain_async", None)
                 logger.debug("Hindsight retain: bank=%s, doc=%s, async=%s, content_len=%d, num_turns=%d",
                              self._bank_id, self._document_id, self._retain_async, len(content), len(self._session_turns))
-                self._run_sync(client.aretain_batch(
-                    bank_id=self._bank_id,
-                    items=[item],
-                    document_id=self._document_id,
-                    retain_async=self._retain_async,
-                ))
+                self._run_hindsight_operation(
+                    lambda client: client.aretain_batch(
+                        bank_id=self._bank_id,
+                        items=[item],
+                        document_id=self._document_id,
+                        retain_async=self._retain_async,
+                    )
+                )
                 logger.debug("Hindsight retain succeeded")
             except Exception as e:
                 logger.warning("Hindsight sync failed: %s", e, exc_info=True)
@@ -1152,12 +1241,6 @@ class HindsightMemoryProvider(MemoryProvider):
         return [RETAIN_SCHEMA, RECALL_SCHEMA, REFLECT_SCHEMA]
 
     def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
-        try:
-            client = self._get_client()
-        except Exception as e:
-            logger.warning("Hindsight client init failed: %s", e)
-            return tool_error(f"Hindsight client unavailable: {e}")
-
         if tool_name == "hindsight_retain":
             content = args.get("content", "")
             if not content:
@@ -1171,7 +1254,7 @@ class HindsightMemoryProvider(MemoryProvider):
                 )
                 logger.debug("Tool hindsight_retain: bank=%s, content_len=%d, context=%s",
                              self._bank_id, len(content), context)
-                self._run_sync(client.aretain(**retain_kwargs))
+                self._run_hindsight_operation(lambda client: client.aretain(**retain_kwargs))
                 logger.debug("Tool hindsight_retain: success")
                 return json.dumps({"result": "Memory stored successfully."})
             except Exception as e:
@@ -1194,7 +1277,7 @@ class HindsightMemoryProvider(MemoryProvider):
                     recall_kwargs["types"] = self._recall_types
                 logger.debug("Tool hindsight_recall: bank=%s, query_len=%d, budget=%s",
                              self._bank_id, len(query), self._budget)
-                resp = self._run_sync(client.arecall(**recall_kwargs))
+                resp = self._run_hindsight_operation(lambda client: client.arecall(**recall_kwargs))
                 num_results = len(resp.results) if resp.results else 0
                 logger.debug("Tool hindsight_recall: %d results", num_results)
                 if not resp.results:
@@ -1212,9 +1295,11 @@ class HindsightMemoryProvider(MemoryProvider):
             try:
                 logger.debug("Tool hindsight_reflect: bank=%s, query_len=%d, budget=%s",
                              self._bank_id, len(query), self._budget)
-                resp = self._run_sync(client.areflect(
-                    bank_id=self._bank_id, query=query, budget=self._budget
-                ))
+                resp = self._run_hindsight_operation(
+                    lambda client: client.areflect(
+                        bank_id=self._bank_id, query=query, budget=self._budget
+                    )
+                )
                 logger.debug("Tool hindsight_reflect: response_len=%d", len(resp.text or ""))
                 return json.dumps({"result": resp.text or "No relevant memories found."})
             except Exception as e:
@@ -1231,9 +1316,19 @@ class HindsightMemoryProvider(MemoryProvider):
         if self._client is not None:
             try:
                 if self._mode == "local_embedded":
-                    # Use the public close() API. The RuntimeError from
-                    # aiohttp's "attached to a different loop" is expected
-                    # and harmless — the daemon keeps running independently.
+                    # HindsightEmbedded.close() delegates to its sync client.close().
+                    # When Hermes created/used that client on the shared async loop,
+                    # closing it from this thread can raise "attached to a different
+                    # loop" before aiohttp releases the session. Close the embedded
+                    # inner async client on the shared loop first, then let the
+                    # wrapper clean up daemon/UI bookkeeping.
+                    inner_client = getattr(self._client, "_client", None)
+                    if inner_client is not None and hasattr(inner_client, "aclose"):
+                        _run_sync(inner_client.aclose())
+                        try:
+                            self._client._client = None
+                        except Exception:
+                            pass
                     try:
                         self._client.close()
                     except RuntimeError:

plugins/memory/openviking/__init__.py

@@ -43,7 +43,7 @@ _TIMEOUT = 30.0
 # ---------------------------------------------------------------------------
 # Process-level atexit safety net — ensures pending sessions are committed
 # even if shutdown_memory_provider is never called (e.g. gateway crash,
-# SIGKILL, or exception in _async_flush_memories preventing shutdown).
+# SIGKILL, or exception in the session expiry watcher preventing shutdown).
 # ---------------------------------------------------------------------------
 _last_active_provider: Optional["OpenVikingMemoryProvider"] = None
 

run_agent.py

@@ -40,6 +40,7 @@ from types import SimpleNamespace
 import urllib.request
 import uuid
 from typing import List, Dict, Any, Optional
+from urllib.parse import urlparse, parse_qs, urlunparse
 from openai import OpenAI
 import fire
 from datetime import datetime
@@ -891,7 +892,6 @@ class AIAgent:
         checkpoints_enabled: bool = False,
         checkpoint_max_snapshots: int = 50,
         pass_session_id: bool = False,
-        persist_session: bool = True,
     ):
         """
         Initialize the AI Agent.
@@ -963,7 +963,6 @@ class AIAgent:
         self.background_review_callback = None  # Optional sync callback for gateway delivery
         self.skip_context_files = skip_context_files
         self.pass_session_id = pass_session_id
-        self.persist_session = persist_session
         self._credential_pool = credential_pool
         self.log_prefix_chars = log_prefix_chars
         self.log_prefix = f"{log_prefix} " if log_prefix else ""
@@ -1033,12 +1032,16 @@ class AIAgent:
         # surface.
         # When api_mode was explicitly provided, respect it — the user
         # knows what their endpoint supports (#10473).
+        # Exception: Azure OpenAI serves gpt-5.x on /chat/completions and
+        # does NOT support the Responses API — skip the upgrade for Azure
+        # (openai.azure.com), even though it looks OpenAI-compatible.
         if (
             api_mode is None
             and self.api_mode == "chat_completions"
             and self.provider != "copilot-acp"
             and not str(self.base_url or "").lower().startswith("acp://copilot")
             and not str(self.base_url or "").lower().startswith("acp+tcp://")
+            and not self._is_azure_openai_url()
             and (
                 self._is_direct_openai_url()
                 or self._provider_model_requires_responses_api(
@@ -1314,7 +1317,22 @@ class AIAgent:
             if api_key and base_url:
                 # Explicit credentials from CLI/gateway — construct directly.
                 # The runtime provider resolver already handled auth for us.
-                client_kwargs = {"api_key": api_key, "base_url": base_url}
+                # Extract query params (e.g. Azure api-version) from base_url
+                # and pass via default_query to prevent loss during SDK URL
+                # joining (httpx drops query string when joining paths).
+                _parsed_url = urlparse(base_url)
+                if _parsed_url.query:
+                    _clean_url = urlunparse(_parsed_url._replace(query=""))
+                    _query_params = {
+                        k: v[0] for k, v in parse_qs(_parsed_url.query).items()
+                    }
+                    client_kwargs = {
+                        "api_key": api_key,
+                        "base_url": _clean_url,
+                        "default_query": _query_params,
+                    }
+                else:
+                    client_kwargs = {"api_key": api_key, "base_url": base_url}
                 if _provider_timeout is not None:
                     client_kwargs["timeout"] = _provider_timeout
                 if self.provider == "copilot-acp":
@@ -1578,7 +1596,6 @@ class AIAgent:
         self._memory_enabled = False
         self._user_profile_enabled = False
         self._memory_nudge_interval = 10
-        self._memory_flush_min_turns = 6
         self._turns_since_memory = 0
         self._iters_since_skill = 0
         if not skip_memory:
@@ -1587,7 +1604,6 @@ class AIAgent:
                 self._memory_enabled = mem_config.get("memory_enabled", False)
                 self._user_profile_enabled = mem_config.get("user_profile_enabled", False)
                 self._memory_nudge_interval = int(mem_config.get("nudge_interval", 10))
-                self._memory_flush_min_turns = int(mem_config.get("flush_min_turns", 6))
                 if self._memory_enabled or self._user_profile_enabled:
                     from tools.memory_tool import MemoryStore
                     self._memory_store = MemoryStore(
@@ -1767,43 +1783,64 @@ class AIAgent:
         # Store for reuse in switch_model (so config override persists across model switches)
         self._config_context_length = _config_context_length
 
+        # Resolve custom_providers list once for reuse below (startup
+        # context-length override and plugin context-engine init).
+        try:
+            from hermes_cli.config import get_compatible_custom_providers
+            _custom_providers = get_compatible_custom_providers(_agent_cfg)
+        except Exception:
+            _custom_providers = _agent_cfg.get("custom_providers")
+            if not isinstance(_custom_providers, list):
+                _custom_providers = []
+
         # Check custom_providers per-model context_length
-        if _config_context_length is None:
+        if _config_context_length is None and _custom_providers:
             try:
-                from hermes_cli.config import get_compatible_custom_providers
-                _custom_providers = get_compatible_custom_providers(_agent_cfg)
+                from hermes_cli.config import get_custom_provider_context_length
+                _cp_ctx_resolved = get_custom_provider_context_length(
+                    model=self.model,
+                    base_url=self.base_url,
+                    custom_providers=_custom_providers,
+                )
+                if _cp_ctx_resolved:
+                    _config_context_length = int(_cp_ctx_resolved)
             except Exception:
-                _custom_providers = _agent_cfg.get("custom_providers")
-                if not isinstance(_custom_providers, list):
-                    _custom_providers = []
-            for _cp_entry in _custom_providers:
-                if not isinstance(_cp_entry, dict):
-                    continue
-                _cp_url = (_cp_entry.get("base_url") or "").rstrip("/")
-                if _cp_url and _cp_url == self.base_url.rstrip("/"):
-                    _cp_models = _cp_entry.get("models", {})
-                    if isinstance(_cp_models, dict):
-                        _cp_model_cfg = _cp_models.get(self.model, {})
-                        if isinstance(_cp_model_cfg, dict):
-                            _cp_ctx = _cp_model_cfg.get("context_length")
-                            if _cp_ctx is not None:
-                                try:
-                                    _config_context_length = int(_cp_ctx)
-                                except (TypeError, ValueError):
-                                    logger.warning(
-                                        "Invalid context_length for model %r in "
-                                        "custom_providers: %r — must be a plain "
-                                        "integer (e.g. 256000, not '256K'). "
-                                        "Falling back to auto-detection.",
-                                        self.model, _cp_ctx,
-                                    )
-                                    print(
-                                        f"\n⚠ Invalid context_length for model {self.model!r} in custom_providers: {_cp_ctx!r}\n"
-                                        f"  Must be a plain integer (e.g. 256000, not '256K').\n"
-                                        f"  Falling back to auto-detected context window.\n",
-                                        file=sys.stderr,
-                                    )
-                    break
+                _cp_ctx_resolved = None
+
+            # Surface a clear warning if the user set a context_length but it
+            # wasn't a valid positive int — the helper silently skips those.
+            if _config_context_length is None:
+                _target = self.base_url.rstrip("/") if self.base_url else ""
+                for _cp_entry in _custom_providers:
+                    if not isinstance(_cp_entry, dict):
+                        continue
+                    _cp_url = (_cp_entry.get("base_url") or "").rstrip("/")
+                    if _target and _cp_url == _target:
+                        _cp_models = _cp_entry.get("models", {})
+                        if isinstance(_cp_models, dict):
+                            _cp_model_cfg = _cp_models.get(self.model, {})
+                            if isinstance(_cp_model_cfg, dict):
+                                _cp_ctx = _cp_model_cfg.get("context_length")
+                                if _cp_ctx is not None:
+                                    try:
+                                        _parsed = int(_cp_ctx)
+                                        if _parsed <= 0:
+                                            raise ValueError
+                                    except (TypeError, ValueError):
+                                        logger.warning(
+                                            "Invalid context_length for model %r in "
+                                            "custom_providers: %r — must be a positive "
+                                            "integer (e.g. 256000, not '256K'). "
+                                            "Falling back to auto-detection.",
+                                            self.model, _cp_ctx,
+                                        )
+                                        print(
+                                            f"\n⚠ Invalid context_length for model {self.model!r} in custom_providers: {_cp_ctx!r}\n"
+                                            f"  Must be a positive integer (e.g. 256000, not '256K').\n"
+                                            f"  Falling back to auto-detected context window.\n",
+                                            file=sys.stderr,
+                                        )
+                        break
         
         # Select context engine: config-driven (like memory providers).
         # 1. Check config.yaml context.engine setting
@@ -1853,6 +1890,7 @@ class AIAgent:
                 api_key=getattr(self, "api_key", ""),
                 config_context_length=_config_context_length,
                 provider=self.provider,
+                custom_providers=_custom_providers,
             )
             self.context_compressor.update_model(
                 model=self.model,
@@ -2143,12 +2181,23 @@ class AIAgent:
         # ── Update context compressor ──
         if hasattr(self, "context_compressor") and self.context_compressor:
             from agent.model_metadata import get_model_context_length
+            # Re-read custom_providers from live config so per-model
+            # context_length overrides are honored when switching to a
+            # custom provider mid-session (closes #15779).
+            _sm_custom_providers = None
+            try:
+                from hermes_cli.config import load_config, get_compatible_custom_providers
+                _sm_cfg = load_config()
+                _sm_custom_providers = get_compatible_custom_providers(_sm_cfg)
+            except Exception:
+                _sm_custom_providers = None
             new_context_length = get_model_context_length(
                 self.model,
                 base_url=self.base_url,
                 api_key=self.api_key,
                 provider=self.provider,
                 config_context_length=getattr(self, "_config_context_length", None),
+                custom_providers=_sm_custom_providers,
             )
             self.context_compressor.update_model(
                 model=self.model,
@@ -2399,6 +2448,7 @@ class AIAgent:
                 base_url=aux_base_url,
                 api_key=aux_api_key,
                 config_context_length=getattr(self, "_aux_compression_context_length_config", None),
+                provider=getattr(self, "provider", ""),
             )
 
             # Hard floor: the auxiliary compression model must have at least
@@ -2425,6 +2475,11 @@ class AIAgent:
                 # compression actually works this session.  The hard floor
                 # above guarantees aux_context >= MINIMUM_CONTEXT_LENGTH,
                 # so the new threshold is always >= 64K.
+                #
+                # The compression summariser sends a single user-role
+                # prompt (no system prompt, no tools) to the aux model, so
+                # new_threshold == aux_context is safe: the request is
+                # the raw messages plus a small summarisation instruction.
                 old_threshold = threshold
                 new_threshold = aux_context
                 self.context_compressor.threshold_tokens = new_threshold
@@ -2500,6 +2555,22 @@ class AIAgent:
             )
         return hostname == "api.openai.com"
 
+    def _is_azure_openai_url(self, base_url: str = None) -> bool:
+        """Return True when a base URL targets Azure OpenAI.
+
+        Azure OpenAI exposes an OpenAI-compatible endpoint at
+        ``{resource}.openai.azure.com/openai/v1`` that accepts the
+        standard ``openai`` Python client.  Unlike api.openai.com it
+        does NOT support the Responses API — gpt-5.x models are served
+        on the regular ``/chat/completions`` path — so routing decisions
+        must treat Azure separately from direct OpenAI.
+        """
+        if base_url is not None:
+            url = str(base_url).lower()
+        else:
+            url = getattr(self, "_base_url_lower", "") or ""
+        return "openai.azure.com" in url
+
     def _resolved_api_call_timeout(self) -> float:
         """Resolve the effective per-call request timeout in seconds.
 
@@ -2671,12 +2742,14 @@ class AIAgent:
 
     def _max_tokens_param(self, value: int) -> dict:
         """Return the correct max tokens kwarg for the current provider.
-        
+
         OpenAI's newer models (gpt-4o, o-series, gpt-5+) require
-        'max_completion_tokens'. OpenRouter, local models, and older
+        'max_completion_tokens'. Azure OpenAI also requires
+        'max_completion_tokens' for gpt-5.x models served via the
+        OpenAI-compatible endpoint. OpenRouter, local models, and older
         OpenAI models use 'max_tokens'.
         """
-        if self._is_direct_openai_url():
+        if self._is_direct_openai_url() or self._is_azure_openai_url():
             return {"max_completion_tokens": value}
         return {"max_tokens": value}
 
@@ -3034,13 +3107,28 @@ class AIAgent:
     )
 
     _SKILL_REVIEW_PROMPT = (
-        "Review the conversation above and consider saving or updating a skill if appropriate.\n\n"
-        "Focus on: was a non-trivial approach used to complete a task that required trial "
-        "and error, or changing course due to experiential findings along the way, or did "
-        "the user expect or desire a different method or outcome?\n\n"
-        "If a relevant skill already exists, update it with what you learned. "
-        "Otherwise, create a new skill if the approach is reusable.\n"
-        "If nothing is worth saving, just say 'Nothing to save.' and stop."
+        "Review the conversation above and consider whether a skill should be saved or updated.\n\n"
+        "Work in this order — do not skip steps:\n\n"
+        "1. SURVEY the existing skill landscape first. Call skills_list to see what you "
+        "have. If anything looks potentially relevant, skill_view it before deciding. "
+        "You are looking for the CLASS of task that just happened, not the exact task. "
+        "Example: a successful Tauri build is in the class \"desktop app build "
+        "troubleshooting\", not \"fix my specific Tauri error today\".\n\n"
+        "2. THINK CLASS-FIRST. What general pattern of task did the user just complete? "
+        "What conditions will trigger this pattern again? Describe the class in one "
+        "sentence before looking at what to save.\n\n"
+        "3. PREFER GENERALIZING AN EXISTING SKILL over creating a new one. If a skill "
+        "already covers the class — even partially — update it (skill_manage patch) "
+        "with the new insight. Broaden its \"when to use\" trigger if needed.\n\n"
+        "4. ONLY CREATE A NEW SKILL when no existing skill reasonably covers the class. "
+        "When you create one, name and scope it at the class level "
+        "(\"react-i18n-setup\", not \"add-i18n-to-my-dashboard-app\"). The trigger "
+        "section must describe the class of situations, not this one session.\n\n"
+        "5. If you notice two existing skills that overlap, note it in your response "
+        "so a future review can consolidate them. Do not consolidate now unless the "
+        "overlap is obvious and low-risk.\n\n"
+        "Only act when something is genuinely worth saving. "
+        "If nothing stands out, just say 'Nothing to save.' and stop."
     )
 
     _COMBINED_REVIEW_PROMPT = (
@@ -3050,9 +3138,16 @@ class AIAgent:
         "about how you should behave, their work style, or ways they want you to operate? "
         "If so, save using the memory tool.\n\n"
         "**Skills**: Was a non-trivial approach used to complete a task that required trial "
-        "and error, or changing course due to experiential findings along the way, or did "
-        "the user expect or desire a different method or outcome? If a relevant skill "
-        "already exists, update it. Otherwise, create a new one if the approach is reusable.\n\n"
+        "and error, changing course due to experiential findings, or a different method "
+        "or outcome than the user expected? If so, work in this order:\n"
+        "  a. SURVEY existing skills first (skills_list, then skill_view on candidates).\n"
+        "  b. Identify the CLASS of task, not the specific task "
+        "(\"desktop app build troubleshooting\", not \"fix my Tauri error\").\n"
+        "  c. PREFER UPDATING/GENERALIZING an existing skill that covers the class.\n"
+        "  d. ONLY CREATE A NEW SKILL if no existing one covers the class. Scope at "
+        "the class level, not this one session.\n"
+        "  e. If you notice overlapping skills during the survey, note it so a future "
+        "review can consolidate them.\n\n"
         "Only act if there's something genuinely worth saving. "
         "If nothing stands out, just say 'Nothing to save.' and stop."
     )
@@ -3150,12 +3245,25 @@ class AIAgent:
                 with open(os.devnull, "w") as _devnull, \
                      contextlib.redirect_stdout(_devnull), \
                      contextlib.redirect_stderr(_devnull):
+                    # Inherit the parent agent's live runtime (provider, model,
+                    # base_url, api_key, api_mode) so the fork uses the exact
+                    # same credentials the main turn is using.  Without this,
+                    # AIAgent.__init__ re-runs auto-resolution from env vars,
+                    # which fails for OAuth-only providers, session-scoped
+                    # creds, or credential-pool setups where the resolver can't
+                    # reconstruct auth from scratch -- producing the spurious
+                    # "No LLM provider configured" warning at end of turn.
+                    _parent_runtime = self._current_main_runtime()
                     review_agent = AIAgent(
                         model=self.model,
                         max_iterations=8,
                         quiet_mode=True,
                         platform=self.platform,
                         provider=self.provider,
+                        api_mode=_parent_runtime.get("api_mode") or None,
+                        base_url=_parent_runtime.get("base_url") or None,
+                        api_key=_parent_runtime.get("api_key") or None,
+                        credential_pool=getattr(self, "_credential_pool", None),
                         parent_session_id=self.session_id,
                     )
                     review_agent._memory_write_origin = "background_review"
@@ -3196,10 +3304,19 @@ class AIAgent:
                 logger.warning("Background memory/skill review failed: %s", e)
                 self._emit_auxiliary_failure("background review", e)
             finally:
-                # Close all resources (httpx client, subprocesses, etc.) so
-                # GC doesn't try to clean them up on a dead asyncio event
-                # loop (which produces "Event loop is closed" errors).
+                # Background review agents can initialize memory providers
+                # (for example Hindsight) that own their own network clients.
+                # Explicitly stop those providers before closing the agent so
+                # their aiohttp sessions do not leak until GC/process exit.
+                # Then close all remaining resources (httpx client,
+                # subprocesses, etc.) so GC doesn't try to clean them up on a
+                # dead asyncio event loop (which produces "Event loop is
+                # closed" errors).
                 if review_agent is not None:
+                    try:
+                        review_agent.shutdown_memory_provider()
+                    except Exception:
+                        pass
                     try:
                         review_agent.close()
                     except Exception:
@@ -3256,10 +3373,7 @@ class AIAgent:
         """Save session state to both JSON log and SQLite on any exit path.
 
         Ensures conversations are never lost, even on errors or early returns.
-        Skipped when ``persist_session=False`` (ephemeral helper flows).
         """
-        if not self.persist_session:
-            return
         self._apply_persist_user_message_override(messages)
         self._session_messages = messages
         self._save_session_log(messages)
@@ -3309,6 +3423,7 @@ class AIAgent:
                     reasoning_content=msg.get("reasoning_content") if role == "assistant" else None,
                     reasoning_details=msg.get("reasoning_details") if role == "assistant" else None,
                     codex_reasoning_items=msg.get("codex_reasoning_items") if role == "assistant" else None,
+                    codex_message_items=msg.get("codex_message_items") if role == "assistant" else None,
                 )
             self._last_flushed_db_idx = len(messages)
         except Exception as e:
@@ -5137,6 +5252,8 @@ class AIAgent:
         # response.incomplete instead of response.completed).
         self._codex_streamed_text_parts: list = []
         for attempt in range(max_stream_retries + 1):
+            if self._interrupt_requested:
+                raise InterruptedError("Agent interrupted before Codex stream retry")
             collected_output_items: list = []
             try:
                 with active_client.responses.stream(**api_kwargs) as stream:
@@ -5431,6 +5548,11 @@ class AIAgent:
         # Other anthropic_messages providers (MiniMax, Alibaba, etc.) use their own keys.
         if self.provider != "anthropic":
             return False
+        # Azure endpoints use static API keys — OAuth token rotation doesn't apply.
+        # Refreshing would pick up ~/.claude/.credentials.json OAuth token and break auth.
+        _base = getattr(self, "_anthropic_base_url", "") or ""
+        if "azure.com" in _base:
+            return False
 
         try:
             from agent.anthropic_adapter import resolve_anthropic_token, build_anthropic_client
@@ -6306,6 +6428,14 @@ class AIAgent:
 
             try:
                 for _stream_attempt in range(_max_stream_retries + 1):
+                    # Check for interrupt before each retry attempt.  Without
+                    # this, /stop closes the HTTP connection (outer poll loop),
+                    # but the retry loop opens a FRESH connection — negating the
+                    # interrupt entirely.  On slow providers (ollama-cloud) each
+                    # retry can block for the full stream-read timeout (120s+),
+                    # causing multi-minute delays between /stop and response.
+                    if self._interrupt_requested:
+                        raise InterruptedError("Agent interrupted before stream retry")
                     try:
                         if self.api_mode == "anthropic_messages":
                             self._try_refresh_anthropic_client_credentials()
@@ -6779,10 +6909,15 @@ class AIAgent:
             # Determine api_mode from provider / base URL / model
             fb_api_mode = "chat_completions"
             fb_base_url = str(fb_client.base_url)
+            _fb_is_azure = self._is_azure_openai_url(fb_base_url)
             if fb_provider == "openai-codex":
                 fb_api_mode = "codex_responses"
             elif fb_provider == "anthropic" or fb_base_url.rstrip("/").lower().endswith("/anthropic"):
                 fb_api_mode = "anthropic_messages"
+            elif _fb_is_azure:
+                # Azure OpenAI serves gpt-5.x on /chat/completions — does NOT
+                # support the Responses API. Stay on chat_completions.
+                fb_api_mode = "chat_completions"
             elif self._is_direct_openai_url(fb_base_url):
                 fb_api_mode = "codex_responses"
             elif self._provider_model_requires_responses_api(
@@ -7655,6 +7790,13 @@ class AIAgent:
         if codex_items:
             msg["codex_reasoning_items"] = codex_items
 
+        # Codex Responses API: preserve exact assistant message items (with
+        # id/phase) so follow-up turns can replay structured items instead of
+        # flattening to plain text. This is required for prefix cache hits.
+        codex_message_items = getattr(assistant_message, "codex_message_items", None)
+        if codex_message_items:
+            msg["codex_message_items"] = codex_message_items
+
         if assistant_message.tool_calls:
             tool_calls = []
             for tool_call in assistant_message.tool_calls:
@@ -7740,25 +7882,53 @@ class AIAgent:
         if source_msg.get("role") != "assistant":
             return
 
-        explicit_reasoning = source_msg.get("reasoning_content")
-        if isinstance(explicit_reasoning, str):
-            api_msg["reasoning_content"] = explicit_reasoning
+        # 1. Explicit reasoning_content already set — preserve it verbatim
+        # (includes DeepSeek/Kimi's own empty-string placeholder written at
+        # creation time, and any valid reasoning content from the same provider).
+        existing = source_msg.get("reasoning_content")
+        if isinstance(existing, str):
+            api_msg["reasoning_content"] = existing
             return
 
+        # 2. Healthy session: promote 'reasoning' field to 'reasoning_content'
+        # for providers that use the internal 'reasoning' key.
+        # This must happen BEFORE the DeepSeek/Kimi tool-call check so that
+        # genuine reasoning content is not overwritten by the empty-string
+        # fallback (#15812 regression in PR #15478).
         normalized_reasoning = source_msg.get("reasoning")
         if isinstance(normalized_reasoning, str) and normalized_reasoning:
             api_msg["reasoning_content"] = normalized_reasoning
             return
 
-        # Providers that require an echoed reasoning_content on every
-        # assistant tool-call turn. Detection logic lives in the per-provider
-        # helpers so both the creation path (_build_assistant_message) and
-        # this replay path stay in sync.
-        if source_msg.get("tool_calls") and (
+        # 3. DeepSeek / Kimi thinking mode: tool-call turns that lack
+        # reasoning_content are "poisoned history" — a prior provider (MiniMax,
+        # etc.) left them empty. DeepSeek returns HTTP 400 if reasoning_content
+        # is absent on replay; inject "" to satisfy the provider's requirement
+        # without forwarding any cross-provider reasoning content.
+        needs_empty_reasoning = (
+            source_msg.get("tool_calls")
+            and (
+                self._needs_kimi_tool_reasoning()
+                or self._needs_deepseek_tool_reasoning()
+            )
+        )
+        if needs_empty_reasoning:
+            api_msg["reasoning_content"] = ""
+            return
+
+        # 4. DeepSeek / Kimi thinking mode: all assistant messages need
+        # reasoning_content. Inject "" to satisfy the provider's requirement
+        # when no explicit reasoning content is present.
+        if (
             self._needs_kimi_tool_reasoning()
             or self._needs_deepseek_tool_reasoning()
         ):
             api_msg["reasoning_content"] = ""
+            return
+
+        # 5. reasoning_content was present but not a string (e.g. None after
+        # context compaction).  Don't pass null to the API.
+        api_msg.pop("reasoning_content", None)
 
     @staticmethod
     def _sanitize_tool_calls_for_strict_api(api_msg: dict) -> dict:
@@ -7910,251 +8080,6 @@ class AIAgent:
         """
         return self.api_mode != "codex_responses"
 
-    def flush_memories(self, messages: list = None, min_turns: int = None):
-        """Give the model one turn to persist memories before context is lost.
-
-        Called before compression, session reset, or CLI exit. Injects a flush
-        message, makes one API call, executes any memory tool calls, then
-        strips all flush artifacts from the message list.
-
-        Args:
-            messages: The current conversation messages. If None, uses
-                      self._session_messages (last run_conversation state).
-            min_turns: Minimum user turns required to trigger the flush.
-                       None = use config value (flush_min_turns).
-                       0 = always flush (used for compression).
-        """
-        if self._memory_flush_min_turns == 0 and min_turns is None:
-            return
-        if "memory" not in self.valid_tool_names or not self._memory_store:
-            return
-        effective_min = min_turns if min_turns is not None else self._memory_flush_min_turns
-        if self._user_turn_count < effective_min:
-            return
-
-        if messages is None:
-            messages = getattr(self, '_session_messages', None)
-        if not messages or len(messages) < 3:
-            return
-
-        flush_content = (
-            "[System: The session is being compressed. "
-            "Save anything worth remembering — prioritize user preferences, "
-            "corrections, and recurring patterns over task-specific details.]"
-        )
-        _sentinel = f"__flush_{id(self)}_{time.monotonic()}"
-        flush_msg = {"role": "user", "content": flush_content, "_flush_sentinel": _sentinel}
-        messages.append(flush_msg)
-
-        try:
-            # Build API messages for the flush call
-            _needs_sanitize = self._should_sanitize_tool_calls()
-            api_messages = []
-            for msg in messages:
-                api_msg = msg.copy()
-                self._copy_reasoning_content_for_api(msg, api_msg)
-                api_msg.pop("reasoning", None)
-                api_msg.pop("finish_reason", None)
-                api_msg.pop("_flush_sentinel", None)
-                api_msg.pop("_thinking_prefill", None)
-                if _needs_sanitize:
-                    self._sanitize_tool_calls_for_strict_api(api_msg)
-                api_messages.append(api_msg)
-
-            if self._cached_system_prompt:
-                api_messages = [{"role": "system", "content": self._cached_system_prompt}] + api_messages
-
-            # Make one API call with only the memory tool available
-            memory_tool_def = None
-            for t in (self.tools or []):
-                if t.get("function", {}).get("name") == "memory":
-                    memory_tool_def = t
-                    break
-
-            if not memory_tool_def:
-                messages.pop()  # remove flush msg
-                return
-
-            # Use auxiliary client for the flush call when available --
-            # it's cheaper and avoids Codex Responses API incompatibility.
-            from agent.auxiliary_client import (
-                call_llm as _call_llm,
-                _fixed_temperature_for_model,
-                OMIT_TEMPERATURE,
-            )
-            _aux_available = True
-            # Kimi models manage temperature server-side — omit it entirely.
-            # Other models with a fixed contract get that value; everyone else
-            # gets the historical 0.3 default.
-            _fixed_temp = _fixed_temperature_for_model(self.model, self.base_url)
-            _omit_temperature = _fixed_temp is OMIT_TEMPERATURE
-            if _omit_temperature:
-                _flush_temperature = None
-            elif _fixed_temp is not None:
-                _flush_temperature = _fixed_temp
-            else:
-                _flush_temperature = 0.3
-            aux_error = None
-            try:
-                response = _call_llm(
-                    task="flush_memories",
-                    messages=api_messages,
-                    tools=[memory_tool_def],
-                    temperature=_flush_temperature,
-                    max_tokens=5120,
-                    # timeout resolved from auxiliary.flush_memories.timeout config
-                )
-            except Exception as e:
-                aux_error = e
-                _aux_available = False
-                response = None
-
-            if not _aux_available and self.api_mode == "codex_responses":
-                # No auxiliary client -- use the Codex Responses path directly
-                codex_kwargs = self._build_api_kwargs(api_messages)
-                _ct_flush = self._get_transport()
-                if _ct_flush is not None:
-                    codex_kwargs["tools"] = _ct_flush.convert_tools([memory_tool_def])
-                elif not codex_kwargs.get("tools"):
-                    codex_kwargs["tools"] = [memory_tool_def]
-                if _flush_temperature is not None:
-                    codex_kwargs["temperature"] = _flush_temperature
-                else:
-                    codex_kwargs.pop("temperature", None)
-                if "max_output_tokens" in codex_kwargs:
-                    codex_kwargs["max_output_tokens"] = 5120
-                response = self._run_codex_stream(codex_kwargs)
-            elif not _aux_available and self.api_mode == "anthropic_messages":
-                # Native Anthropic — use the transport for kwargs
-                _tflush = self._get_transport()
-                ant_kwargs = _tflush.build_kwargs(
-                    model=self.model, messages=api_messages,
-                    tools=[memory_tool_def], max_tokens=5120,
-                    reasoning_config=None,
-                    preserve_dots=self._anthropic_preserve_dots(),
-                )
-                response = self._anthropic_messages_create(ant_kwargs)
-            elif not _aux_available:
-                api_kwargs = {
-                    "model": self.model,
-                    "messages": api_messages,
-                    "tools": [memory_tool_def],
-                    **self._max_tokens_param(5120),
-                }
-                if _flush_temperature is not None:
-                    api_kwargs["temperature"] = _flush_temperature
-                from agent.auxiliary_client import _get_task_timeout
-                response = self._ensure_primary_openai_client(reason="flush_memories").chat.completions.create(
-                    **api_kwargs, timeout=_get_task_timeout("flush_memories")
-                )
-
-            if aux_error is not None:
-                logger.warning("Auxiliary memory flush failed; used fallback path: %s", aux_error)
-                self._emit_auxiliary_failure("memory flush", aux_error)
-
-            def _openai_tool_calls(resp):
-                if resp is not None and hasattr(resp, "choices") and resp.choices:
-                    msg = getattr(resp.choices[0], "message", None)
-                    calls = getattr(msg, "tool_calls", None)
-                    if calls:
-                        return calls
-                return []
-
-            def _codex_output_tool_calls(resp):
-                calls = []
-                for item in getattr(resp, "output", []) or []:
-                    if getattr(item, "type", None) == "function_call":
-                        calls.append(SimpleNamespace(
-                            id=getattr(item, "call_id", None),
-                            type="function",
-                            function=SimpleNamespace(
-                                name=getattr(item, "name", ""),
-                                arguments=getattr(item, "arguments", "{}"),
-                            ),
-                        ))
-                return calls
-
-            # Extract tool calls from the response, handling all API formats
-            tool_calls = []
-            if self.api_mode == "codex_responses" and not _aux_available:
-                _ct_flush = self._get_transport()
-                _cnr_flush = _ct_flush.normalize_response(response) if _ct_flush is not None else None
-                if _cnr_flush and _cnr_flush.tool_calls:
-                    tool_calls = [
-                        SimpleNamespace(
-                            id=tc.id, type="function",
-                            function=SimpleNamespace(name=tc.name, arguments=tc.arguments),
-                        ) for tc in _cnr_flush.tool_calls
-                    ]
-                else:
-                    tool_calls = _codex_output_tool_calls(response)
-            elif self.api_mode == "anthropic_messages" and not _aux_available:
-                _tfn = self._get_transport()
-                _flush_result = _tfn.normalize_response(response, strip_tool_prefix=self._is_anthropic_oauth)
-                if _flush_result and _flush_result.tool_calls:
-                    tool_calls = [
-                        SimpleNamespace(
-                            id=tc.id, type="function",
-                            function=SimpleNamespace(name=tc.name, arguments=tc.arguments),
-                        ) for tc in _flush_result.tool_calls
-                    ]
-            elif self.api_mode in ("chat_completions", "bedrock_converse"):
-                # chat_completions / bedrock — normalize through transport
-                _tfn = self._get_transport()
-                _flush_result = _tfn.normalize_response(response) if _tfn is not None else None
-                if _flush_result and _flush_result.tool_calls:
-                    tool_calls = _flush_result.tool_calls
-                else:
-                    tool_calls = _openai_tool_calls(response)
-            elif _aux_available and hasattr(response, "choices") and response.choices:
-                # Auxiliary client returned OpenAI-shaped response while main
-                # api_mode is codex/anthropic — extract tool_calls from .choices
-                tool_calls = _openai_tool_calls(response)
-
-            for tc in tool_calls:
-                if tc.function.name == "memory":
-                    try:
-                        args = json.loads(tc.function.arguments)
-                        flush_target = args.get("target", "memory")
-                        from tools.memory_tool import memory_tool as _memory_tool
-                        _memory_tool(
-                            action=args.get("action"),
-                            target=flush_target,
-                            content=args.get("content"),
-                            old_text=args.get("old_text"),
-                            store=self._memory_store,
-                        )
-                        if self._memory_manager and args.get("action") in ("add", "replace"):
-                            try:
-                                self._memory_manager.on_memory_write(
-                                    args.get("action", ""),
-                                    flush_target,
-                                    args.get("content", ""),
-                                    metadata=self._build_memory_write_metadata(
-                                        write_origin="memory_flush",
-                                        execution_context="flush_memories",
-                                    ),
-                                )
-                            except Exception:
-                                pass
-                        if not self.quiet_mode:
-                            print(f"  🧠 Memory flush: saved to {args.get('target', 'memory')}")
-                    except Exception as e:
-                        logger.warning("Memory flush tool call failed: %s", e)
-                        self._emit_auxiliary_failure("memory flush tool", e)
-        except Exception as e:
-            logger.warning("Memory flush API call failed: %s", e)
-            self._emit_auxiliary_failure("memory flush", e)
-        finally:
-            # Strip flush artifacts: remove everything from the flush message onward.
-            # Use sentinel marker instead of identity check for robustness.
-            while messages and messages[-1].get("_flush_sentinel") != _sentinel:
-                messages.pop()
-                if not messages:
-                    break
-            if messages and messages[-1].get("_flush_sentinel") == _sentinel:
-                messages.pop()
-
     def _compress_context(self, messages: list, system_message: str, *, approx_tokens: int = None, task_id: str = "default", focus_topic: str = None) -> tuple:
         """Compress conversation context and split the session in SQLite.
 
@@ -8173,8 +8098,6 @@ class AIAgent:
             f"{approx_tokens:,}" if approx_tokens else "unknown", self.model,
             focus_topic,
         )
-        # Pre-compression memory flush: let the model save memories before they're lost
-        self.flush_memories(messages, min_turns=0)
 
         # Notify external memory provider before compression discards context
         if self._memory_manager:
@@ -8237,6 +8160,22 @@ class AIAgent:
             except Exception as e:
                 logger.warning("Session DB compression split failed — new session will NOT be indexed: %s", e)
 
+        # Notify the context engine that the session_id rotated because of
+        # compression (not a fresh /new). Plugin engines (e.g. hermes-lcm) use
+        # boundary_reason="compression" to preserve DAG lineage across the
+        # rollover instead of re-initializing fresh per-session state.
+        # See hermes-lcm#68. Built-in ContextCompressor ignores kwargs.
+        try:
+            _old_sid = locals().get("old_session_id")
+            if _old_sid and hasattr(self.context_compressor, "on_session_start"):
+                self.context_compressor.on_session_start(
+                    self.session_id or "",
+                    boundary_reason="compression",
+                    old_session_id=_old_sid,
+                )
+        except Exception as _ce_err:
+            logger.debug("context engine on_session_start (compression): %s", _ce_err)
+
         # Warn on repeated compressions (quality degrades with each pass)
         _cc = self.context_compressor.compression_count
         if _cc >= 2:
@@ -11126,36 +11065,69 @@ class AIAgent:
                                 continue
 
                     # ── Nous Portal: record rate limit & skip retries ─────
-                    # When Nous returns a 429, record the reset time to a
-                    # shared file so ALL sessions (cron, gateway, auxiliary)
-                    # know not to pile on.  Then skip further retries —
-                    # each one burns another RPH request and deepens the
-                    # rate limit hole.  The retry loop's top-of-iteration
-                    # guard will catch this on the next pass and try
-                    # fallback or bail with a clear message.
+                    # When Nous returns a 429 that is a genuine account-
+                    # level rate limit, record the reset time to a shared
+                    # file so ALL sessions (cron, gateway, auxiliary) know
+                    # not to pile on, then skip further retries -- each
+                    # one burns another RPH request and deepens the hole.
+                    # The retry loop's top-of-iteration guard will catch
+                    # this on the next pass and try fallback or bail.
+                    #
+                    # IMPORTANT: Nous Portal multiplexes multiple upstream
+                    # providers (DeepSeek, Kimi, MiMo, Hermes).  A 429 can
+                    # also mean an UPSTREAM provider is out of capacity
+                    # for one specific model -- transient, clears in
+                    # seconds, nothing to do with the caller's quota.
+                    # Tripping the cross-session breaker on that would
+                    # block every Nous model for minutes.  We use
+                    # ``is_genuine_nous_rate_limit`` to tell the two
+                    # apart via the 429's own x-ratelimit-* headers and
+                    # the last-known-good state captured on the previous
+                    # successful response.
                     if (
                         is_rate_limited
                         and self.provider == "nous"
                         and classified.reason == FailoverReason.rate_limit
                         and not recovered_with_pool
                     ):
+                        _genuine_nous_rate_limit = False
                         try:
-                            from agent.nous_rate_guard import record_nous_rate_limit
+                            from agent.nous_rate_guard import (
+                                is_genuine_nous_rate_limit,
+                                record_nous_rate_limit,
+                            )
                             _err_resp = getattr(api_error, "response", None)
                             _err_hdrs = (
                                 getattr(_err_resp, "headers", None)
                                 if _err_resp else None
                             )
-                            record_nous_rate_limit(
+                            _genuine_nous_rate_limit = is_genuine_nous_rate_limit(
                                 headers=_err_hdrs,
-                                error_context=error_context,
+                                last_known_state=self._rate_limit_state,
                             )
+                            if _genuine_nous_rate_limit:
+                                record_nous_rate_limit(
+                                    headers=_err_hdrs,
+                                    error_context=error_context,
+                                )
+                            else:
+                                logging.info(
+                                    "Nous 429 looks like upstream capacity "
+                                    "(no exhausted bucket in headers or "
+                                    "last-known state) -- not tripping "
+                                    "cross-session breaker."
+                                )
                         except Exception:
                             pass
-                        # Skip straight to max_retries — the top-of-loop
-                        # guard will handle fallback or bail cleanly.
-                        retry_count = max_retries
-                        continue
+                        if _genuine_nous_rate_limit:
+                            # Skip straight to max_retries -- the
+                            # top-of-loop guard will handle fallback or
+                            # bail cleanly.
+                            retry_count = max_retries
+                            continue
+                        # Upstream capacity 429: fall through to normal
+                        # retry logic.  A different model (or the same
+                        # model a moment later) will typically succeed.
 
                     is_payload_too_large = (
                         classified.reason == FailoverReason.payload_too_large
@@ -11757,16 +11729,26 @@ class AIAgent:
                     interim_has_content = bool((interim_msg.get("content") or "").strip())
                     interim_has_reasoning = bool(interim_msg.get("reasoning", "").strip()) if isinstance(interim_msg.get("reasoning"), str) else False
                     interim_has_codex_reasoning = bool(interim_msg.get("codex_reasoning_items"))
+                    interim_has_codex_message_items = bool(interim_msg.get("codex_message_items"))
 
-                    if interim_has_content or interim_has_reasoning or interim_has_codex_reasoning:
+                    if (
+                        interim_has_content
+                        or interim_has_reasoning
+                        or interim_has_codex_reasoning
+                        or interim_has_codex_message_items
+                    ):
                         last_msg = messages[-1] if messages else None
                         # Duplicate detection: two consecutive incomplete assistant
                         # messages with identical content AND reasoning are collapsed.
-                        # For reasoning-only messages (codex_reasoning_items differ but
-                        # visible content/reasoning are both empty), we also compare
-                        # the encrypted items to avoid silently dropping new state.
+                        # For provider-state-only changes (encrypted reasoning
+                        # items or replayable message ids/phases/statuses differ
+                        # while visible content/reasoning are unchanged), compare
+                        # those opaque payloads too so we don't silently drop the
+                        # newer continuation state.
                         last_codex_items = last_msg.get("codex_reasoning_items") if isinstance(last_msg, dict) else None
                         interim_codex_items = interim_msg.get("codex_reasoning_items")
+                        last_codex_message_items = last_msg.get("codex_message_items") if isinstance(last_msg, dict) else None
+                        interim_codex_message_items = interim_msg.get("codex_message_items")
                         duplicate_interim = (
                             isinstance(last_msg, dict)
                             and last_msg.get("role") == "assistant"
@@ -11774,6 +11756,7 @@ class AIAgent:
                             and (last_msg.get("content") or "") == (interim_msg.get("content") or "")
                             and (last_msg.get("reasoning") or "") == (interim_msg.get("reasoning") or "")
                             and last_codex_items == interim_codex_items
+                            and last_codex_message_items == interim_codex_message_items
                         )
                         if not duplicate_interim:
                             messages.append(interim_msg)

scripts/build_model_catalog.py

@@ -0,0 +1,95 @@
+#!/usr/bin/env python3
+"""Build the Hermes Model Catalog — a centralized JSON manifest of curated models.
+
+This script reads the in-repo hardcoded curated lists (``OPENROUTER_MODELS``,
+``_PROVIDER_MODELS["nous"]``) and writes them to a JSON manifest that the
+Hermes CLI fetches at runtime. Publishing the catalog through the docs site
+lets maintainers update model lists without shipping a Hermes release.
+
+The runtime fetcher falls back to the same in-repo hardcoded lists if the
+manifest is unreachable, so this script is a convenience for keeping the
+manifest in sync — not a source of truth.
+
+Usage::
+
+    python scripts/build_model_catalog.py
+
+Output: ``website/static/api/model-catalog.json``
+
+Live URL (after ``deploy-site.yml`` runs on merge to main):
+``https://hermes-agent.nousresearch.com/docs/api/model-catalog.json``
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+from datetime import datetime, timezone
+
+REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, REPO_ROOT)
+
+# Ensure HERMES_HOME is set for imports that touch it at module level.
+os.environ.setdefault("HERMES_HOME", os.path.join(os.path.expanduser("~"), ".hermes"))
+
+from hermes_cli.models import OPENROUTER_MODELS, _PROVIDER_MODELS  # noqa: E402
+
+OUTPUT_PATH = os.path.join(REPO_ROOT, "website", "static", "api", "model-catalog.json")
+CATALOG_VERSION = 1
+
+
+def build_catalog() -> dict:
+    return {
+        "version": CATALOG_VERSION,
+        "updated_at": datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
+        "metadata": {
+            "source": "hermes-agent repo",
+            "docs": "https://hermes-agent.nousresearch.com/docs/reference/model-catalog",
+        },
+        "providers": {
+            "openrouter": {
+                "metadata": {
+                    "display_name": "OpenRouter",
+                    "note": (
+                        "Descriptions drive picker badges. Live /api/v1/models "
+                        "filters curated ids by tool-calling support and free pricing."
+                    ),
+                },
+                "models": [
+                    {"id": mid, "description": desc}
+                    for mid, desc in OPENROUTER_MODELS
+                ],
+            },
+            "nous": {
+                "metadata": {
+                    "display_name": "Nous Portal",
+                    "note": (
+                        "Free-tier gating is determined live via Portal pricing "
+                        "(partition_nous_models_by_tier), not this manifest."
+                    ),
+                },
+                "models": [
+                    {"id": mid}
+                    for mid in _PROVIDER_MODELS.get("nous", [])
+                ],
+            },
+        },
+    }
+
+
+def main() -> int:
+    catalog = build_catalog()
+    os.makedirs(os.path.dirname(OUTPUT_PATH), exist_ok=True)
+    with open(OUTPUT_PATH, "w") as fh:
+        json.dump(catalog, fh, indent=2)
+        fh.write("\n")
+
+    print(f"Wrote {OUTPUT_PATH}")
+    for provider, block in catalog["providers"].items():
+        print(f"  {provider}: {len(block['models'])} models")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

scripts/install.sh

@@ -29,10 +29,25 @@ BOLD='\033[1m'
 REPO_URL_SSH="git@github.com:NousResearch/hermes-agent.git"
 REPO_URL_HTTPS="https://github.com/NousResearch/hermes-agent.git"
 HERMES_HOME="${HERMES_HOME:-$HOME/.hermes}"
-INSTALL_DIR="${HERMES_INSTALL_DIR:-$HERMES_HOME/hermes-agent}"
+# INSTALL_DIR is resolved AFTER arg parsing and OS detection so we can pick an
+# FHS-style layout for root installs.  Track whether the user gave us an
+# explicit directory — if so we never override it.
+if [ -n "${HERMES_INSTALL_DIR:-}" ]; then
+    INSTALL_DIR="$HERMES_INSTALL_DIR"
+    INSTALL_DIR_EXPLICIT=true
+else
+    INSTALL_DIR=""
+    INSTALL_DIR_EXPLICIT=false
+fi
 PYTHON_VERSION="3.11"
 NODE_VERSION="22"
 
+# FHS-style root install layout (set by resolve_install_layout when applicable):
+#   code at /usr/local/lib/hermes-agent, command at /usr/local/bin/hermes,
+#   data still at /root/.hermes (HERMES_HOME).  Matches Claude Code / Codex CLI
+#   and keeps Docker bind-mounted /root/ volumes lean.
+ROOT_FHS_LAYOUT=false
+
 # Options
 USE_VENV=true
 RUN_SETUP=true
@@ -64,6 +79,7 @@ while [[ $# -gt 0 ]]; do
             ;;
         --dir)
             INSTALL_DIR="$2"
+            INSTALL_DIR_EXPLICIT=true
             shift 2
             ;;
         --hermes-home)
@@ -79,9 +95,20 @@ while [[ $# -gt 0 ]]; do
             echo "  --no-venv      Don't create virtual environment"
             echo "  --skip-setup   Skip interactive setup wizard"
             echo "  --branch NAME  Git branch to install (default: main)"
-            echo "  --dir PATH     Installation directory (default: ~/.hermes/hermes-agent)"
+            echo "  --dir PATH     Installation directory"
+            echo "                   default (non-root):  ~/.hermes/hermes-agent"
+            echo "                   default (root, Linux): /usr/local/lib/hermes-agent"
             echo "  --hermes-home PATH  Data directory (default: ~/.hermes, or \$HERMES_HOME)"
             echo "  -h, --help     Show this help"
+            echo ""
+            echo "Notes:"
+            echo "  When running as root on Linux, Hermes installs the code under"
+            echo "  /usr/local/lib/hermes-agent and links the command into"
+            echo "  /usr/local/bin/hermes (FHS layout — matches Claude Code / Codex CLI)."
+            echo "  Data, config, sessions, and logs still live in \$HERMES_HOME"
+            echo "  (default /root/.hermes).  This keeps Docker bind-mounted volumes"
+            echo "  small and ensures the command is on PATH for all shells."
+            echo "  Existing installs at \$HERMES_HOME/hermes-agent are preserved in-place."
             exit 0
             ;;
         *)
@@ -163,9 +190,60 @@ is_termux() {
     [ -n "${TERMUX_VERSION:-}" ] || [[ "${PREFIX:-}" == *"com.termux/files/usr"* ]]
 }
 
+# Decide where the repo checkout + venv live, and where the `hermes` command
+# symlink goes.  Called after detect_os so $OS/$DISTRO are known.
+#
+# Defaults:
+#   - Non-root, any OS:       INSTALL_DIR = $HERMES_HOME/hermes-agent
+#                             command link in $HOME/.local/bin
+#   - Termux (any uid):       INSTALL_DIR = $HERMES_HOME/hermes-agent
+#                             command link in $PREFIX/bin (already on PATH)
+#   - Root on Linux (new):    INSTALL_DIR = /usr/local/lib/hermes-agent
+#                             command link in /usr/local/bin
+#                             (unless a legacy install already exists at
+#                              $HERMES_HOME/hermes-agent — then preserve it)
+#
+# Always no-op when the user set --dir or $HERMES_INSTALL_DIR.
+resolve_install_layout() {
+    if [ "$INSTALL_DIR_EXPLICIT" = true ]; then
+        log_info "Install directory: $INSTALL_DIR (explicit)"
+        return 0
+    fi
+
+    # Termux: package manager manages /data/data/..., keep code in HERMES_HOME.
+    if is_termux; then
+        INSTALL_DIR="$HERMES_HOME/hermes-agent"
+        return 0
+    fi
+
+    # Root on Linux: prefer FHS layout unless a legacy install already exists.
+    # macOS root installs keep the legacy layout because /usr/local/ on macOS
+    # is Homebrew territory and we don't want to fight that.
+    if [ "$OS" = "linux" ] && [ "$(id -u)" -eq 0 ]; then
+        if [ -d "$HERMES_HOME/hermes-agent/.git" ]; then
+            INSTALL_DIR="$HERMES_HOME/hermes-agent"
+            log_info "Existing install detected at $INSTALL_DIR — keeping legacy layout"
+            log_info "  (new root installs use /usr/local/lib/hermes-agent)"
+            return 0
+        fi
+        INSTALL_DIR="/usr/local/lib/hermes-agent"
+        ROOT_FHS_LAYOUT=true
+        log_info "Root install on Linux — using FHS layout"
+        log_info "  Code:    $INSTALL_DIR"
+        log_info "  Command: /usr/local/bin/hermes"
+        log_info "  Data:    $HERMES_HOME (unchanged)"
+        return 0
+    fi
+
+    # Default: non-root, non-Termux → legacy user-scoped layout.
+    INSTALL_DIR="$HERMES_HOME/hermes-agent"
+}
+
 get_command_link_dir() {
     if is_termux && [ -n "${PREFIX:-}" ]; then
         echo "$PREFIX/bin"
+    elif [ "$ROOT_FHS_LAYOUT" = true ]; then
+        echo "/usr/local/bin"
     else
         echo "$HOME/.local/bin"
     fi
@@ -174,6 +252,8 @@ get_command_link_dir() {
 get_command_link_display_dir() {
     if is_termux && [ -n "${PREFIX:-}" ]; then
         echo '$PREFIX/bin'
+    elif [ "$ROOT_FHS_LAYOUT" = true ]; then
+        echo '/usr/local/bin'
     else
         echo '~/.local/bin'
     fi
@@ -975,6 +1055,41 @@ setup_path() {
         return 0
     fi
 
+    # FHS layout: /usr/local/bin is normally on PATH for login shells (via
+    # /etc/profile pathmunge), but on RHEL/CentOS/Rocky/Alma 8+ non-login
+    # interactive root shells (su, sudo -s, tmux panes, some web terminals)
+    # only source /etc/bashrc, which does NOT add /usr/local/bin — and
+    # /root/.bash_profile doesn't either.  So verify with `command -v` and
+    # fall back to writing a PATH guard into /root/.bashrc when needed.
+    if [ "$ROOT_FHS_LAYOUT" = true ]; then
+        export PATH="$command_link_dir:$PATH"
+        # Probe a fresh non-login interactive bash the way the user will use it.
+        # `bash -i -c` sources ~/.bashrc but NOT ~/.bash_profile or /etc/profile,
+        # which is the exact scenario where RHEL root loses /usr/local/bin.
+        if env -i HOME="$HOME" TERM="${TERM:-dumb}" bash -i -c 'command -v hermes' \
+                >/dev/null 2>&1; then
+            log_info "/usr/local/bin is already on PATH for all shells"
+            log_success "hermes command ready"
+            return 0
+        fi
+
+        log_info "hermes not on PATH in non-login shells (common on RHEL-family)"
+        PATH_LINE='export PATH="/usr/local/bin:$PATH"'
+        PATH_COMMENT='# Hermes Agent — ensure /usr/local/bin is on PATH (RHEL non-login shells)'
+        for SHELL_CONFIG in "$HOME/.bashrc" "$HOME/.bash_profile"; do
+            [ -f "$SHELL_CONFIG" ] || continue
+            if ! grep -v '^[[:space:]]*#' "$SHELL_CONFIG" 2>/dev/null \
+                    | grep -qE 'PATH=.*(/usr/local/bin|\$command_link_dir)'; then
+                echo "" >> "$SHELL_CONFIG"
+                echo "$PATH_COMMENT" >> "$SHELL_CONFIG"
+                echo "$PATH_LINE" >> "$SHELL_CONFIG"
+                log_success "Added /usr/local/bin to PATH in $SHELL_CONFIG"
+            fi
+        done
+        log_success "hermes command ready"
+        return 0
+    fi
+
     # Check if ~/.local/bin is on PATH; if not, add it to shell config.
     # Detect the user's actual login shell (not the shell running this script,
     # which is always bash when piped from curl).
@@ -1339,12 +1454,12 @@ print_success() {
     echo ""
 
     # Show file locations
-    echo -e "${CYAN}${BOLD}📁 Your files (all in ~/.hermes/):${NC}"
+    echo -e "${CYAN}${BOLD}📁 Your files:${NC}"
     echo ""
-    echo -e "   ${YELLOW}Config:${NC}    ~/.hermes/config.yaml"
-    echo -e "   ${YELLOW}API Keys:${NC}  ~/.hermes/.env"
-    echo -e "   ${YELLOW}Data:${NC}      ~/.hermes/cron/, sessions/, logs/"
-    echo -e "   ${YELLOW}Code:${NC}      ~/.hermes/hermes-agent/"
+    echo -e "   ${YELLOW}Config:${NC}    $HERMES_HOME/config.yaml"
+    echo -e "   ${YELLOW}API Keys:${NC}  $HERMES_HOME/.env"
+    echo -e "   ${YELLOW}Data:${NC}      $HERMES_HOME/cron/, sessions/, logs/"
+    echo -e "   ${YELLOW}Code:${NC}      $INSTALL_DIR"
     echo ""
 
     echo -e "${CYAN}─────────────────────────────────────────────────────────${NC}"
@@ -1364,6 +1479,9 @@ print_success() {
     if [ "$DISTRO" = "termux" ]; then
         echo -e "${YELLOW}⚡ 'hermes' was linked into $(get_command_link_display_dir), which is already on PATH in Termux.${NC}"
         echo ""
+    elif [ "$ROOT_FHS_LAYOUT" = true ]; then
+        echo -e "${YELLOW}⚡ 'hermes' was linked into /usr/local/bin and is ready to use — no shell reload needed.${NC}"
+        echo ""
     else
         echo -e "${YELLOW}⚡ Reload your shell to use 'hermes' command:${NC}"
         echo ""
@@ -1415,6 +1533,7 @@ main() {
     print_banner
 
     detect_os
+    resolve_install_layout
     install_uv
     check_python
     check_git

scripts/profile-tui.py

@@ -0,0 +1,614 @@
+#!/usr/bin/env python3
+"""Drive the Hermes TUI under HERMES_DEV_PERF and summarize the pipeline.
+
+Usage:
+  scripts/profile-tui.py [--session SID] [--hold KEY] [--seconds N] [--rate HZ]
+
+Defaults: picks the session with the most messages, holds PageUp for 8s at
+~30 Hz (matching xterm key-repeat), summarizes ~/.hermes/perf.log on exit.
+
+The --tui build must exist (run `npm run build` in ui-tui first). This script
+launches `node dist/entry.js` directly with HERMES_TUI_RESUME set so it
+bypasses the hermes_cli wrapper — we want repeatable timing, not the CLI's
+session-picker flow.
+
+Environment overrides:
+  HERMES_PERF_LOG     (default ~/.hermes/perf.log)
+  HERMES_PERF_NODE    (default node from $PATH)
+  HERMES_TUI_DIR      (default /home/bb/hermes-agent/ui-tui)
+
+Exit code is 0 if the harness ran and parsed results, 2 if the TUI crashed
+or produced no perf data (suggests HERMES_DEV_PERF wiring is broken).
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import pty
+import select
+import signal
+import sqlite3
+import sys
+import time
+from pathlib import Path
+from typing import Any
+
+
+DEFAULT_TUI_DIR = Path(os.environ.get("HERMES_TUI_DIR", "/home/bb/hermes-agent/ui-tui"))
+DEFAULT_LOG = Path(os.environ.get("HERMES_PERF_LOG", str(Path.home() / ".hermes" / "perf.log")))
+DEFAULT_STATE_DB = Path.home() / ".hermes" / "state.db"
+
+# Keystroke escape sequences.  Matches what xterm/VT220 send when the
+# terminal has bracketed-paste disabled and the key-repeat handler fires.
+KEYS = {
+    "page_up": b"\x1b[5~",
+    "page_down": b"\x1b[6~",
+    "wheel_up": b"\x1b[M`!!",      # mouse wheel up (SGR-less) — best-effort
+    "shift_up": b"\x1b[1;2A",
+    "shift_down": b"\x1b[1;2B",
+}
+
+
+def pick_longest_session(db: Path) -> str:
+    conn = sqlite3.connect(db)
+    row = conn.execute(
+        "SELECT id FROM sessions s ORDER BY "
+        "(SELECT COUNT(*) FROM messages m WHERE m.session_id = s.id) DESC LIMIT 1"
+    ).fetchone()
+    if not row:
+        sys.exit(f"no sessions in {db}")
+    return row[0]
+
+
+def drain(fd: int, timeout: float) -> bytes:
+    """Read whatever's available from fd within `timeout`, then return."""
+    chunks = []
+    end = time.monotonic() + timeout
+    while time.monotonic() < end:
+        r, _, _ = select.select([fd], [], [], max(0.0, end - time.monotonic()))
+        if not r:
+            break
+        try:
+            data = os.read(fd, 4096)
+        except OSError:
+            break
+        if not data:
+            break
+        chunks.append(data)
+    return b"".join(chunks)
+
+
+def hold_key(fd: int, seq: bytes, seconds: float, rate_hz: int) -> int:
+    """Write `seq` to fd at ~rate_hz for `seconds`. Returns keystrokes sent."""
+    interval = 1.0 / max(1, rate_hz)
+    end = time.monotonic() + seconds
+    sent = 0
+    while time.monotonic() < end:
+        try:
+            os.write(fd, seq)
+            sent += 1
+        except OSError:
+            break
+        # Drain stdout to keep the PTY buffer flowing; ignore content.
+        drain(fd, 0)
+        time.sleep(interval)
+    return sent
+
+
+def summarize(log: Path, since_ts_ms: int) -> dict[str, Any]:
+    """Parse perf.log, keep only events newer than since_ts_ms, return stats."""
+    react_events: list[dict[str, Any]] = []
+    frame_events: list[dict[str, Any]] = []
+    if not log.exists():
+        return {"error": f"no log at {log}", "react": [], "frame": []}
+    for line in log.read_text().splitlines():
+        line = line.strip()
+        if not line:
+            continue
+        try:
+            row = json.loads(line)
+        except json.JSONDecodeError:
+            continue
+        if int(row.get("ts", 0)) < since_ts_ms:
+            continue
+        src = row.get("src")
+        if src == "react":
+            react_events.append(row)
+        elif src == "frame":
+            frame_events.append(row)
+
+    return {
+        "react": react_events,
+        "frame": frame_events,
+    }
+
+
+def pct(values: list[float], p: float) -> float:
+    if not values:
+        return 0.0
+    s = sorted(values)
+    idx = min(len(s) - 1, int(len(s) * p))
+    return s[idx]
+
+
+def format_report(data: dict[str, Any]) -> str:
+    react = data.get("react") or []
+    frames = data.get("frame") or []
+    out = []
+
+    out.append("═══ React Profiler ═══")
+    if not react:
+        out.append("  (no react events — HERMES_DEV_PERF wired? threshold too high?)")
+    else:
+        by_id: dict[str, list[float]] = {}
+        for r in react:
+            by_id.setdefault(r["id"], []).append(r["actualMs"])
+        out.append(f"  {'pane':<14} {'count':>6} {'p50':>8} {'p95':>8} {'p99':>8} {'max':>8}")
+        for pid, ms in sorted(by_id.items(), key=lambda kv: -pct(kv[1], 0.99)):
+            out.append(
+                f"  {pid:<14} {len(ms):>6} {pct(ms,0.50):>8.2f} {pct(ms,0.95):>8.2f} "
+                f"{pct(ms,0.99):>8.2f} {max(ms):>8.2f}"
+            )
+
+    out.append("")
+    out.append("═══ Ink pipeline ═══")
+    if not frames:
+        out.append("  (no frame events — onFrame wiring broken?)")
+    else:
+        dur = [f["durationMs"] for f in frames]
+        phases_present = any(f.get("phases") for f in frames)
+        out.append(f"  frames captured: {len(frames)}")
+        out.append(
+            f"  durationMs  p50={pct(dur,0.50):.2f}  p95={pct(dur,0.95):.2f}  "
+            f"p99={pct(dur,0.99):.2f}  max={max(dur):.2f}"
+        )
+        # Effective FPS during the run: frames / elapsed seconds.
+        ts = sorted(f["ts"] for f in frames)
+        if len(ts) >= 2:
+            elapsed_s = (ts[-1] - ts[0]) / 1000.0
+            fps = len(frames) / elapsed_s if elapsed_s > 0 else float("inf")
+            out.append(f"  throughput: {len(frames)} frames / {elapsed_s:.2f}s = {fps:.1f} fps")
+
+        if phases_present:
+            fields = ["yoga", "renderer", "diff", "optimize", "write", "commit"]
+            out.append("")
+            out.append(f"  {'phase':<10} {'p50':>8} {'p95':>8} {'p99':>8} {'max':>8}   (ms)")
+            for field in fields:
+                vals = [f["phases"][field] for f in frames if f.get("phases")]
+                if vals:
+                    out.append(
+                        f"  {field:<10} {pct(vals,0.50):>8.2f} {pct(vals,0.95):>8.2f} "
+                        f"{pct(vals,0.99):>8.2f} {max(vals):>8.2f}"
+                    )
+            # Derived: sum of phases vs durationMs (reveals hidden time).
+            sum_ps = [
+                sum(f["phases"][k] for k in fields)
+                for f in frames if f.get("phases")
+            ]
+            if sum_ps:
+                dur_match = [f["durationMs"] for f in frames if f.get("phases")]
+                deltas = [d - s for d, s in zip(dur_match, sum_ps)]
+                out.append(
+                    f"  {'dur-Σphases':<10} {pct(deltas,0.50):>8.2f} {pct(deltas,0.95):>8.2f} "
+                    f"{pct(deltas,0.99):>8.2f} {max(deltas):>8.2f}   (unaccounted-for time)"
+                )
+
+            # Yoga counters
+            visited = [f["phases"]["yogaVisited"] for f in frames if f.get("phases")]
+            measured = [f["phases"]["yogaMeasured"] for f in frames if f.get("phases")]
+            cache_hits = [f["phases"]["yogaCacheHits"] for f in frames if f.get("phases")]
+            live = [f["phases"]["yogaLive"] for f in frames if f.get("phases")]
+            out.append("")
+            out.append("  Yoga counters (per frame):")
+            for name, vals in (
+                ("visited", visited),
+                ("measured", measured),
+                ("cacheHits", cache_hits),
+                ("live", live),
+            ):
+                if vals:
+                    out.append(f"    {name:<11} p50={pct(vals,0.5):.0f}  p99={pct(vals,0.99):.0f}  max={max(vals)}")
+
+            # Patch counts — proxy for "how much changed each frame"
+            patches = [f["phases"]["patches"] for f in frames if f.get("phases")]
+            if patches:
+                out.append(
+                    f"  patches     p50={pct(patches,0.5):.0f}  p99={pct(patches,0.99):.0f}  "
+                    f"max={max(patches)}  total={sum(patches)}"
+                )
+            optimized = [
+                f["phases"].get("optimizedPatches", 0)
+                for f in frames if f.get("phases")
+            ]
+            if any(optimized):
+                out.append(
+                    f"  optimized   p50={pct(optimized,0.5):.0f}  p99={pct(optimized,0.99):.0f}  "
+                    f"max={max(optimized)}  total={sum(optimized)}"
+                    f"  (ratio: {sum(optimized)/max(1,sum(patches)):.2f})"
+                )
+
+            # Write bytes + drain telemetry — the outer-terminal bottleneck gauge.
+            bytes_written = [
+                f["phases"].get("writeBytes", 0)
+                for f in frames if f.get("phases")
+            ]
+            if any(bytes_written):
+                total_b = sum(bytes_written)
+                kb = total_b / 1024
+                out.append(
+                    f"  writeBytes  p50={pct(bytes_written,0.5):.0f}B  p99={pct(bytes_written,0.99):.0f}B  "
+                    f"max={max(bytes_written)}B  total={kb:.1f}KB"
+                )
+            drains = [
+                f["phases"].get("prevFrameDrainMs", 0)
+                for f in frames if f.get("phases")
+            ]
+            if any(d > 0 for d in drains):
+                nonzero = [d for d in drains if d > 0]
+                out.append(
+                    f"  drainMs     p50={pct(nonzero,0.5):.2f}  p95={pct(nonzero,0.95):.2f}  "
+                    f"p99={pct(nonzero,0.99):.2f}  max={max(nonzero):.2f}   (terminal flush latency)"
+                )
+            backpressure = sum(1 for f in frames if f.get("phases", {}).get("backpressure"))
+            if backpressure:
+                out.append(
+                    f"  backpressure: {backpressure}/{len(frames)} frames "
+                    f"({100*backpressure/len(frames):.0f}%)   (Node stdout buffer full — terminal slow)"
+                )
+
+        # Flickers
+        flicker_frames = [f for f in frames if f.get("flickers")]
+        if flicker_frames:
+            out.append("")
+            out.append(f"  ⚠ flickers detected in {len(flicker_frames)} frames")
+            reasons: dict[str, int] = {}
+            for f in flicker_frames:
+                for fl in f["flickers"]:
+                    reasons[fl["reason"]] = reasons.get(fl["reason"], 0) + 1
+            for reason, n in sorted(reasons.items(), key=lambda kv: -kv[1]):
+                out.append(f"    {reason}: {n}")
+
+    return "\n".join(out)
+
+
+def key_metrics(data: dict[str, Any]) -> dict[str, float]:
+    """Flatten the report into a dict of scalar metrics for A/B diffing."""
+    metrics: dict[str, float] = {}
+    frames = data.get("frame") or []
+    react = data.get("react") or []
+
+    if frames:
+        durs = [f["durationMs"] for f in frames]
+        metrics["frames"] = len(frames)
+        metrics["dur_p50"] = pct(durs, 0.50)
+        metrics["dur_p95"] = pct(durs, 0.95)
+        metrics["dur_p99"] = pct(durs, 0.99)
+        metrics["dur_max"] = max(durs)
+
+        ts = sorted(f["ts"] for f in frames)
+        if len(ts) >= 2:
+            elapsed = (ts[-1] - ts[0]) / 1000.0
+            metrics["fps_throughput"] = len(frames) / elapsed if elapsed > 0 else 0.0
+            # Interframe gaps distribution — complementary view to throughput:
+            gaps = [ts[i] - ts[i - 1] for i in range(1, len(ts))]
+            if gaps:
+                metrics["gap_p50_ms"] = pct(gaps, 0.50)
+                metrics["gap_p99_ms"] = pct(gaps, 0.99)
+                metrics["gaps_under_16ms"] = sum(1 for g in gaps if g < 16)
+                metrics["gaps_over_200ms"] = sum(1 for g in gaps if g >= 200)
+
+        for phase in ("renderer", "yoga", "diff", "write"):
+            vals = [f["phases"][phase] for f in frames if f.get("phases")]
+            if vals:
+                metrics[f"{phase}_p99"] = pct(vals, 0.99)
+                metrics[f"{phase}_max"] = max(vals)
+
+        patches = [f["phases"]["patches"] for f in frames if f.get("phases")]
+        if patches:
+            metrics["patches_total"] = sum(patches)
+            metrics["patches_p99"] = pct(patches, 0.99)
+
+        optimized = [
+            f["phases"].get("optimizedPatches", 0) for f in frames if f.get("phases")
+        ]
+        if any(optimized):
+            metrics["optimized_total"] = sum(optimized)
+
+        bytes_list = [
+            f["phases"].get("writeBytes", 0) for f in frames if f.get("phases")
+        ]
+        if any(bytes_list):
+            metrics["writeBytes_total"] = sum(bytes_list)
+
+        drains = [
+            f["phases"].get("prevFrameDrainMs", 0)
+            for f in frames if f.get("phases")
+        ]
+        drain_nonzero = [d for d in drains if d > 0]
+        if drain_nonzero:
+            metrics["drain_p99"] = pct(drain_nonzero, 0.99)
+            metrics["drain_max"] = max(drain_nonzero)
+
+        bp = sum(1 for f in frames if f.get("phases", {}).get("backpressure"))
+        metrics["backpressure_frames"] = bp
+
+    if react:
+        for pid in set(e["id"] for e in react):
+            ms = [e["actualMs"] for e in react if e["id"] == pid]
+            metrics[f"react_{pid}_p99"] = pct(ms, 0.99)
+            metrics[f"react_{pid}_max"] = max(ms)
+
+    return metrics
+
+
+def format_diff(before: dict[str, float], after: dict[str, float]) -> str:
+    """Render a side-by-side A/B comparison table."""
+    keys = sorted(set(before) | set(after))
+    lines = [f"{'metric':<28} {'before':>12} {'after':>12} {'delta':>12}  {'%':>6}"]
+    lines.append("─" * 76)
+    for k in keys:
+        b = before.get(k, 0.0)
+        a = after.get(k, 0.0)
+        d = a - b
+        pct_change = ((a / b) - 1) * 100 if b not in (0, 0.0) else float("inf") if a else 0
+
+        # Flag improvements vs regressions. For _p99 / _max / _total / gaps_over /
+        # patches / writeBytes / backpressure, LOWER is better.  For fps / gaps_under,
+        # HIGHER is better.
+        lower_is_better = any(
+            token in k
+            for token in (
+                "p50",
+                "p95",
+                "p99",
+                "_max",
+                "_total",
+                "gaps_over",
+                "backpressure",
+                "drain",
+            )
+        )
+        higher_is_better = "fps_" in k or "gaps_under" in k
+        mark = ""
+        if d and not (lower_is_better or higher_is_better):
+            mark = ""
+        elif d < 0 and lower_is_better:
+            mark = "↓"
+        elif d > 0 and higher_is_better:
+            mark = "↑"
+        elif d > 0 and lower_is_better:
+            mark = "↑"  # regression
+        elif d < 0 and higher_is_better:
+            mark = "↓"  # regression
+
+        pct_str = "—" if pct_change == float("inf") else f"{pct_change:+6.1f}%"
+        lines.append(
+            f"{k:<28} {b:>12.2f} {a:>12.2f} {d:>+12.2f}  {pct_str} {mark}"
+        )
+
+    return "\n".join(lines)
+
+
+def run_once(args: argparse.Namespace) -> dict[str, Any]:
+    tui_dir = Path(args.tui_dir).resolve()
+    entry = tui_dir / "dist" / "entry.js"
+    if not entry.exists():
+        sys.exit(f"{entry} missing — run `npm run build` in {tui_dir} first")
+
+    sid = args.session or pick_longest_session(DEFAULT_STATE_DB)
+    print(f"• session: {sid}")
+    print(f"• hold: {args.hold} x {args.rate}Hz for {args.seconds}s after {args.warmup}s warmup")
+    print(f"• terminal: {args.cols}x{args.rows}")
+
+    log = Path(args.log)
+    if not args.keep_log and log.exists():
+        log.unlink()
+
+    since_ms = int(time.time() * 1000)
+
+    env = os.environ.copy()
+    env["HERMES_DEV_PERF"] = "1"
+    env["HERMES_DEV_PERF_MS"] = str(args.threshold_ms)
+    env["HERMES_DEV_PERF_LOG"] = str(log)
+    env["HERMES_TUI_RESUME"] = sid
+    env["COLUMNS"] = str(args.cols)
+    env["LINES"] = str(args.rows)
+    env["TERM"] = env.get("TERM", "xterm-256color")
+
+    # Pass through extra flags the TUI wrapper recognizes (e.g. --no-fullscreen).
+    # Stored on args as `extra_flags` list.
+    node = os.environ.get("HERMES_PERF_NODE", "node")
+    node_args = [node, str(entry), *getattr(args, "extra_flags", [])]
+
+    pid, fd = pty.fork()
+    if pid == 0:
+        os.execvpe(node, node_args, env)
+
+    try:
+        import fcntl, struct, termios
+        winsize = struct.pack("HHHH", args.rows, args.cols, 0, 0)
+        fcntl.ioctl(fd, termios.TIOCSWINSZ, winsize)
+
+        print(f"• pid: {pid}  fd: {fd}")
+        print(f"• warmup {args.warmup}s (drain startup output)…")
+        drain(fd, args.warmup)
+
+        print(f"• holding {args.hold}…")
+        sent = hold_key(fd, KEYS[args.hold], args.seconds, args.rate)
+        print(f"  sent {sent} keystrokes")
+
+        drain(fd, 0.5)
+    finally:
+        try:
+            os.kill(pid, signal.SIGTERM)
+            for _ in range(10):
+                pid_done, _ = os.waitpid(pid, os.WNOHANG)
+                if pid_done == pid:
+                    break
+                time.sleep(0.1)
+            else:
+                os.kill(pid, signal.SIGKILL)
+                os.waitpid(pid, 0)
+        except (ProcessLookupError, ChildProcessError):
+            pass
+        try:
+            os.close(fd)
+        except OSError:
+            pass
+
+    time.sleep(0.2)
+    return summarize(log, since_ms)
+
+
+def main() -> int:
+    p = argparse.ArgumentParser()
+    p.add_argument("--session", help="session id to resume (default: longest in db)")
+    p.add_argument("--hold", default="page_up", choices=sorted(KEYS.keys()), help="key to hold")
+    p.add_argument("--seconds", type=float, default=8.0, help="how long to hold the key")
+    p.add_argument("--rate", type=int, default=30, help="keystrokes per second")
+    p.add_argument("--warmup", type=float, default=3.0, help="seconds to wait after launch before input")
+    p.add_argument("--threshold-ms", type=float, default=0.0, help="HERMES_DEV_PERF_MS (0 = capture all)")
+    p.add_argument("--cols", type=int, default=120)
+    p.add_argument("--rows", type=int, default=40)
+    p.add_argument("--keep-log", action="store_true", help="don't wipe perf.log before run")
+    p.add_argument("--tui-dir", default=str(DEFAULT_TUI_DIR))
+    p.add_argument("--log", default=str(DEFAULT_LOG))
+    p.add_argument("--save", metavar="LABEL",
+                   help="save the final metrics as /tmp/perf-<LABEL>.json for later --compare")
+    p.add_argument("--compare", metavar="LABEL",
+                   help="diff against /tmp/perf-<LABEL>.json after running")
+    p.add_argument("--loop", action="store_true",
+                   help="watch for source changes, rebuild, rerun, and diff vs previous run")
+    p.add_argument("--extra-flag", dest="extra_flags", action="append", default=[],
+                   help="pass through to node dist/entry.js (repeatable)")
+    args = p.parse_args()
+
+    if args.loop:
+        return loop_mode(args)
+
+    # Single-shot path.
+    data = run_once(args)
+    print()
+    print(format_report(data))
+
+    metrics = key_metrics(data)
+
+    if args.save:
+        path = Path(f"/tmp/perf-{args.save}.json")
+        path.write_text(json.dumps(metrics, indent=2))
+        print(f"\n• saved: {path}")
+
+    if args.compare:
+        path = Path(f"/tmp/perf-{args.compare}.json")
+        if not path.exists():
+            print(f"\n⚠ no baseline at {path} — run with --save {args.compare} first")
+        else:
+            before = json.loads(path.read_text())
+            print(f"\n═══ A/B diff vs /tmp/perf-{args.compare}.json ═══")
+            print(format_diff(before, metrics))
+
+    if not data["react"] and not data["frame"]:
+        return 2
+    return 0
+
+
+def loop_mode(args: argparse.Namespace) -> int:
+    """Watch source files, rebuild, rerun, print A/B diff against previous run.
+
+    Keeps a rolling 'previous run' baseline in memory so each iteration
+    reports delta vs the last one — visibility into whether the last
+    edit moved the needle.  Press Ctrl+C to stop.
+    """
+    import subprocess
+
+    tui_dir = Path(args.tui_dir).resolve()
+    src_root = tui_dir / "src"
+    pkg_root = tui_dir / "packages" / "hermes-ink" / "src"
+
+    def collect_mtimes() -> dict[str, float]:
+        mtimes: dict[str, float] = {}
+        for root in (src_root, pkg_root):
+            if not root.exists():
+                continue
+            for path in root.rglob("*"):
+                if path.suffix in {".ts", ".tsx"} and "__tests__" not in str(path):
+                    try:
+                        mtimes[str(path)] = path.stat().st_mtime
+                    except OSError:
+                        pass
+        return mtimes
+
+    previous_metrics: dict[str, float] | None = None
+    previous_mtimes = collect_mtimes()
+    iteration = 0
+
+    print(f"• loop mode — watching {src_root} + {pkg_root} for *.ts(x) changes")
+    print("• edit any TS file, the harness rebuilds + reruns automatically")
+    print("• Ctrl+C to stop\n")
+
+    try:
+        while True:
+            iteration += 1
+            print(f"\n{'═' * 76}")
+            print(f"Iteration {iteration}  @ {time.strftime('%H:%M:%S')}")
+            print("═" * 76)
+
+            if iteration > 1:
+                print("• rebuilding…")
+                result = subprocess.run(
+                    ["npm", "run", "build"],
+                    cwd=tui_dir,
+                    capture_output=True,
+                    text=True,
+                )
+                if result.returncode != 0:
+                    print("✗ build failed:")
+                    print(result.stdout[-2000:])
+                    print(result.stderr[-2000:])
+                    print("\n• waiting for source changes to retry…")
+                    previous_mtimes = wait_for_change(previous_mtimes, collect_mtimes)
+                    continue
+                print("✓ build ok")
+
+            data = run_once(args)
+            metrics = key_metrics(data)
+
+            print()
+            print(format_report(data))
+
+            if previous_metrics is not None:
+                print(f"\n═══ A/B diff vs iteration {iteration - 1} ═══")
+                print(format_diff(previous_metrics, metrics))
+
+            previous_metrics = metrics
+
+            print("\n• waiting for source changes…")
+            previous_mtimes = wait_for_change(previous_mtimes, collect_mtimes)
+    except KeyboardInterrupt:
+        print("\n• loop stopped")
+        return 0
+
+
+def wait_for_change(prev: dict[str, float], collect) -> dict[str, float]:
+    """Poll every 1s until a watched file's mtime changes. Debounced 500ms."""
+    while True:
+        time.sleep(1)
+        current = collect()
+
+        changed = [
+            path for path, mtime in current.items() if prev.get(path) != mtime
+        ]
+
+        if changed:
+            print(f"  ↻ {len(changed)} file(s) changed:")
+            for path in changed[:5]:
+                print(f"    {path}")
+            # Debounce — editor save bursts can take ~500ms to settle
+            time.sleep(0.5)
+            return collect()
+
+
+if __name__ == "__main__":
+    sys.exit(main())

scripts/release.py

@@ -43,14 +43,22 @@ AUTHOR_MAP = {
     "teknium1@gmail.com": "teknium1",
     "teknium@nousresearch.com": "teknium1",
     "127238744+teknium1@users.noreply.github.com": "teknium1",
+    "johnnncenaaa77@gmail.com": "johnncenae",
+    "focusflow.app.help@gmail.com": "yes999zc",
     "343873859@qq.com": "DrStrangerUJN",
     "uzmpsk.dilekakbas@gmail.com": "dlkakbs",
     "jefferson@heimdallstrategy.com": "Mind-Dragon",
     "130918800+devorun@users.noreply.github.com": "devorun",
+    "sonoyuncudmr@gmail.com": "Sonoyunchu",
     "maks.mir@yahoo.com": "say8hi",
     "web3blind@users.noreply.github.com": "web3blind",
     "julia@alexland.us": "alexg0bot",
     "1060770+benjaminsehl@users.noreply.github.com": "benjaminsehl",
+    "nerijusn76@gmail.com": "Nerijusas",
+    "itonov@proton.me": "Ito-69",
+    "glesstech@gmail.com": "georgeglessner",
+    "maxim.smetanin@gmail.com": "maxims-oss",
+    "yoimexex@gmail.com": "Yoimex",
     # contributors (from noreply pattern)
     "david.vv@icloud.com": "davidvv",
     "wangqiang@wangqiangdeMac-mini.local": "xiaoqiang243",
@@ -65,9 +73,14 @@ AUTHOR_MAP = {
     "thomasgeorgevii09@gmail.com": "tochukwuada",
     "harryykyle1@gmail.com": "hharry11",
     "kshitijk4poor@gmail.com": "kshitijk4poor",
+    "1294707+Tosko4@users.noreply.github.com": "Tosko4",
     "keira.voss94@gmail.com": "keiravoss94",
     "16443023+stablegenius49@users.noreply.github.com": "stablegenius49",
+    "fqsy1416@gmail.com": "EKKOLearnAI",
+    "octo-patch@github.com": "octo-patch",
+    "math0r-be@github.com": "math0r-be",
     "simbamax99@gmail.com": "simbam99",
+    "iris@growthpillars.co": "irispillars",
     "185121704+stablegenius49@users.noreply.github.com": "stablegenius49",
     "101283333+batuhankocyigit@users.noreply.github.com": "batuhankocyigit",
     "255305877+ismell0992-afk@users.noreply.github.com": "ismell0992-afk",
@@ -92,6 +105,8 @@ AUTHOR_MAP = {
     "104278804+Sertug17@users.noreply.github.com": "Sertug17",
     "112503481+caentzminger@users.noreply.github.com": "caentzminger",
     "258577966+voidborne-d@users.noreply.github.com": "voidborne-d",
+    "liusway405@gmail.com": "voidborne-d",
+    "xydarcher@uestc.edu.cn": "Readon",
     "sir_even@icloud.com": "sirEven",
     "36056348+sirEven@users.noreply.github.com": "sirEven",
     "70424851+insecurejezza@users.noreply.github.com": "insecurejezza",
@@ -110,9 +125,22 @@ AUTHOR_MAP = {
     "Mibayy@users.noreply.github.com": "Mibayy",
     "mibayy@users.noreply.github.com": "Mibayy",
     "135070653+sgaofen@users.noreply.github.com": "sgaofen",
+    "lzy.dev@gmail.com": "zhiyanliu",
+    "me@janstepanovsky.cz": "hhhonzik",
+    "139848623+hhuang91@users.noreply.github.com": "hhuang91",
+    "s.ozaki@ebinou.net": "Satoshi-agi",
+    "10774721+kunlabs@users.noreply.github.com": "kunlabs",
+    "110560187+Wang-tianhao@users.noreply.github.com": "Wang-tianhao",
+    "170458616+ghostmfr@users.noreply.github.com": "ghostmfr",
+    "1848670+mewwts@users.noreply.github.com": "mewwts",
+    "1930707+haru398801@users.noreply.github.com": "haru398801",
+    "rapabelias@gmail.com": "badgerbees",
+    "xnb888@proton.me": "xnbi",
+    "xiahu889889@proton.me": "xiahu88988",
     "nocoo@users.noreply.github.com": "nocoo",
     "30841158+n-WN@users.noreply.github.com": "n-WN",
     "tsuijinglei@gmail.com": "hiddenpuppy",
+    "buraysandro9@gmail.com": "ygd58",
     "jerome@clawwork.ai": "HiddenPuppy",
     "jerome.benoit@sap.com": "jerome-benoit",
     "wysie@users.noreply.github.com": "Wysie",
@@ -175,12 +203,17 @@ AUTHOR_MAP = {
     "jaisehgal11299@gmail.com": "jaisup",
     "percydikec@gmail.com": "PercyDikec",
     "noonou7@gmail.com": "HenkDz",
+    # Azure Foundry salvage (PRs #9029, #4599, #10086, #8766)
+    "tech@smartlogics.net": "TechPrototyper",
+    "637186+HangGlidersRule@users.noreply.github.com": "HangGlidersRule",
+    "pein892@gmail.com": "pein892",
     "dean.kerr@gmail.com": "deankerr",
     "socrates1024@gmail.com": "socrates1024",
     "seanalt555@gmail.com": "Salt-555",
     "satelerd@gmail.com": "satelerd",
     "dan@danlynn.com": "danklynn",
     "mattmaximo@hotmail.com": "MattMaximo",
+    "MatthewRHardwick@gmail.com": "mrhwick",
     "149063006+j3ffffff@users.noreply.github.com": "j3ffffff",
     "A-FdL-Prog@users.noreply.github.com": "A-FdL-Prog",
     "l0hde@users.noreply.github.com": "l0hde",
@@ -367,6 +400,17 @@ AUTHOR_MAP = {
     "zzn+pa@zzn.im": "xinbenlv",
     "zaynjarvis@gmail.com": "ZaynJarvis",
     "zhiheng.liu@bytedance.com": "ZaynJarvis",
+    "izhaolongfei@gmail.com": "loongfay",
+    "296659110@qq.com": "lrt4836",
+    "fe.daniel91@gmail.com": "beforeload",
+    "libo1106@foxmail.com": "libo1106",
+    "295367131@qq.com": "295367131",
+    "295367132@qq.com": "IxAres",
+    "danieldliu@tencent.com": "danieldliu",
+    "loongzhao@tencent.com": "loongzhao",
+    "Bartok9@users.noreply.github.com": "Bartok9",
+    "LeonSGP43@users.noreply.github.com": "LeonSGP43",
+    "kshitijk4poor@users.noreply.github.com": "kshitijk4poor",
     "mbelleau@Michels-MacBook-Pro.local": "malaiwah",
     "michel.belleau@malaiwah.com": "malaiwah",
     "gnanasekaran.sekareee@gmail.com": "gnanam1990",
@@ -409,6 +453,7 @@ AUTHOR_MAP = {
     "105142614+VTRiot@users.noreply.github.com": "VTRiot",
     "vivien000812@gmail.com": "iamagenius00",
     "89228157+Feranmi10@users.noreply.github.com": "Feranmi10",
+    "oluwadareferanmi11@gmail.com": "Feranmi10",
     "simon@gtcl.us": "simon-gtcl",
     "suzukaze.haduki@gmail.com": "houko",
     "cliff@cigii.com": "cgarwood82",
@@ -504,6 +549,7 @@ AUTHOR_MAP = {
     "screenmachine@gmail.com": "teknium1",
     "chenzeshi@live.com": "chen1749144759",
     "mor.aleksandr@yahoo.com": "MorAlekss",
+    "ash@users.noreply.github.com": "ash",
 }
 
 

skills/autonomous-ai-agents/hermes-agent/SKILL.md

@@ -281,7 +281,6 @@ Type these during an interactive chat session.
 ### Utility
 ```
 /branch (/fork)      Branch the current session
-/btw                 Ephemeral side question (doesn't interrupt main task)
 /fast                Toggle priority/fast processing
 /browser             Open CDP browser connection
 /history             Show conversation history (CLI)
@@ -403,6 +402,63 @@ Tool changes take effect on `/reset` (new session). They do NOT apply mid-conver
 
 ---
 
+## Security & Privacy Toggles
+
+Common "why is Hermes doing X to my output / tool calls / commands?" toggles — and the exact commands to change them. Most of these need a fresh session (`/reset` in chat, or start a new `hermes` invocation) because they're read once at startup.
+
+### Secret redaction in tool output
+
+Hermes auto-redacts strings that look like API keys, tokens, and secrets in all tool output (terminal stdout, `read_file`, web content, subagent summaries, etc.) so the model never sees raw credentials. If the user is intentionally working with mock tokens, share-management tokens, or their own secrets and the redaction is getting in the way:
+
+```bash
+hermes config set security.redact_secrets false      # disable globally
+```
+
+**Restart required.** `security.redact_secrets` is snapshotted at import time — setting it mid-session (e.g. via `export HERMES_REDACT_SECRETS=false` from a tool call) will NOT take effect for the running process. Tell the user to run `hermes config set security.redact_secrets false` in a terminal, then start a new session. This is deliberate — it prevents an LLM from turning off redaction on itself mid-task.
+
+Re-enable with:
+```bash
+hermes config set security.redact_secrets true
+```
+
+### PII redaction in gateway messages
+
+Separate from secret redaction. When enabled, the gateway hashes user IDs and strips phone numbers from the session context before it reaches the model:
+
+```bash
+hermes config set privacy.redact_pii true    # enable
+hermes config set privacy.redact_pii false   # disable (default)
+```
+
+### Command approval prompts
+
+By default (`approvals.mode: manual`), Hermes prompts the user before running shell commands flagged as destructive (`rm -rf`, `git reset --hard`, etc.). The modes are:
+
+- `manual` — always prompt (default)
+- `smart` — use an auxiliary LLM to auto-approve low-risk commands, prompt on high-risk
+- `off` — skip all approval prompts (equivalent to `--yolo`)
+
+```bash
+hermes config set approvals.mode smart       # recommended middle ground
+hermes config set approvals.mode off         # bypass everything (not recommended)
+```
+
+Per-invocation bypass without changing config:
+- `hermes --yolo …`
+- `export HERMES_YOLO_MODE=1`
+
+Note: YOLO / `approvals.mode: off` does NOT turn off secret redaction. They are independent.
+
+### Shell hooks allowlist
+
+Some shell-hook integrations require explicit allowlisting before they fire. Managed via `~/.hermes/shell-hooks-allowlist.json` — prompted interactively the first time a hook wants to run.
+
+### Disabling the web/browser/image-gen tools
+
+To keep the model away from network or media tools entirely, open `hermes tools` and toggle per-platform. Takes effect on next session (`/reset`). See the Tools & Skills section above.
+
+---
+
 ## Voice & Transcription
 
 ### STT (Voice → Text)

skills/feeds/DESCRIPTION.md

@@ -1,3 +0,0 @@
----
-description: Skills for monitoring, aggregating, and processing RSS feeds, blogs, and web content sources.
----

skills/mlops/inference/obliteratus/SKILL.md

@@ -17,6 +17,13 @@ Remove refusal behaviors (guardrails) from open-weight LLMs without retraining o
 
 **License warning:** OBLITERATUS is AGPL-3.0. NEVER import it as a Python library. Always invoke via CLI (`obliteratus` command) or subprocess. This keeps Hermes Agent's MIT license clean.
 
+## Video Guide
+
+Walkthrough of OBLITERATUS used by a Hermes agent to abliterate Gemma:
+https://www.youtube.com/watch?v=8fG9BrNTeHs ("OBLITERATUS: An AI Agent Removed Gemma 4's Safety Guardrails")
+
+Useful when the user wants a visual overview of the end-to-end workflow before running it themselves.
+
 ## When to Use This Skill
 
 Trigger when the user:

skills/productivity/airtable/SKILL.md

@@ -0,0 +1,228 @@
+---
+name: airtable
+description: Airtable REST API via curl. Records CRUD, filters, upserts.
+version: 1.1.0
+author: community
+license: MIT
+prerequisites:
+  env_vars: [AIRTABLE_API_KEY]
+  commands: [curl]
+metadata:
+  hermes:
+    tags: [Airtable, Productivity, Database, API]
+    homepage: https://airtable.com/developers/web/api/introduction
+---
+
+# Airtable — Bases, Tables & Records
+
+Work with Airtable's REST API directly via `curl` using the `terminal` tool. No MCP server, no OAuth flow, no Python SDK — just `curl` and a personal access token.
+
+## Prerequisites
+
+1. Create a **Personal Access Token (PAT)** at https://airtable.com/create/tokens (tokens start with `pat...`).
+2. Grant these scopes (minimum):
+   - `data.records:read` — read rows
+   - `data.records:write` — create / update / delete rows
+   - `schema.bases:read` — list bases and tables
+3. **Important:** in the same token UI, add each base you want to access to the token's **Access** list. PATs are scoped per-base — a valid token on the wrong base returns `403`.
+4. Store the token in `~/.hermes/.env` (or via `hermes setup`):
+   ```
+   AIRTABLE_API_KEY=pat_your_token_here
+   ```
+
+> Note: legacy `key...` API keys were deprecated Feb 2024. Only PATs and OAuth tokens work now.
+
+## API Basics
+
+- **Endpoint:** `https://api.airtable.com/v0`
+- **Auth header:** `Authorization: Bearer $AIRTABLE_API_KEY`
+- **All requests** use JSON (`Content-Type: application/json` for any POST/PATCH/PUT body).
+- **Object IDs:** bases `app...`, tables `tbl...`, records `rec...`, fields `fld...`. IDs never change; names can. Prefer IDs in automations.
+- **Rate limit:** 5 requests/sec/base. `429` → back off. Burst on a single base will be throttled.
+
+Base curl pattern:
+```bash
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE?maxRecords=5" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+`-s` suppresses curl's progress bar — keep it set for every call so the tool output stays clean for Hermes. Pipe through `python3 -m json.tool` (always present) or `jq` (if installed) for readable JSON.
+
+## Field Types (request body shapes)
+
+| Field type | Write shape |
+|---|---|
+| Single line text | `"Name": "hello"` |
+| Long text | `"Notes": "multi\nline"` |
+| Number | `"Score": 42` |
+| Checkbox | `"Done": true` |
+| Single select | `"Status": "Todo"` (name must already exist unless `typecast: true`) |
+| Multi-select | `"Tags": ["urgent", "bug"]` |
+| Date | `"Due": "2026-04-01"` |
+| DateTime (UTC) | `"At": "2026-04-01T14:30:00.000Z"` |
+| URL / Email / Phone | `"Link": "https://…"` |
+| Attachment | `"Files": [{"url": "https://…"}]` (Airtable fetches + rehosts) |
+| Linked record | `"Owner": ["recXXXXXXXXXXXXXX"]` (array of record IDs) |
+| User | `"AssignedTo": {"id": "usrXXXXXXXXXXXXXX"}` |
+
+Pass `"typecast": true` at the top level of a create/update body to let Airtable auto-coerce values (e.g. create a new select option on the fly, convert `"42"` → `42`).
+
+## Common Queries
+
+### List bases the token can see
+```bash
+curl -s "https://api.airtable.com/v0/meta/bases" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+### List tables + schema for a base
+```bash
+curl -s "https://api.airtable.com/v0/meta/bases/$BASE_ID/tables" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+Use this BEFORE mutating — confirms exact field names and IDs, surfaces `options.choices` for select fields, and shows primary-field names.
+
+### List records (first 10)
+```bash
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE?maxRecords=10" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+### Get a single record
+```bash
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE/$RECORD_ID" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+### Filter records (filterByFormula)
+Airtable formulas must be URL-encoded. Let Python stdlib do it — never hand-encode:
+```bash
+FORMULA="{Status}='Todo'"
+ENC=$(python3 -c 'import sys, urllib.parse; print(urllib.parse.quote(sys.argv[1], safe=""))' "$FORMULA")
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE?filterByFormula=$ENC&maxRecords=20" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+Useful formula patterns:
+- Exact match: `{Email}='user@example.com'`
+- Contains: `FIND('bug', LOWER({Title}))`
+- Multiple conditions: `AND({Status}='Todo', {Priority}='High')`
+- Or: `OR({Owner}='alice', {Owner}='bob')`
+- Not empty: `NOT({Assignee}='')`
+- Date comparison: `IS_AFTER({Due}, TODAY())`
+
+### Sort + select specific fields
+```bash
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE?sort%5B0%5D%5Bfield%5D=Priority&sort%5B0%5D%5Bdirection%5D=asc&fields%5B%5D=Name&fields%5B%5D=Status" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+Square brackets in query params MUST be URL-encoded (`%5B` / `%5D`).
+
+### Use a named view
+```bash
+curl -s "https://api.airtable.com/v0/$BASE_ID/$TABLE?view=Grid%20view&maxRecords=50" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+Views apply their saved filter + sort server-side.
+
+## Common Mutations
+
+### Create a record
+```bash
+curl -s -X POST "https://api.airtable.com/v0/$BASE_ID/$TABLE" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"fields":{"Name":"New task","Status":"Todo","Priority":"High"}}' | python3 -m json.tool
+```
+
+### Create up to 10 records in one call
+```bash
+curl -s -X POST "https://api.airtable.com/v0/$BASE_ID/$TABLE" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "typecast": true,
+    "records": [
+      {"fields": {"Name": "Task A", "Status": "Todo"}},
+      {"fields": {"Name": "Task B", "Status": "In progress"}}
+    ]
+  }' | python3 -m json.tool
+```
+Batch endpoints are capped at **10 records per request**. For larger inserts, loop in batches of 10 with a short sleep to respect 5 req/sec/base.
+
+### Update a record (PATCH — merges, preserves unchanged fields)
+```bash
+curl -s -X PATCH "https://api.airtable.com/v0/$BASE_ID/$TABLE/$RECORD_ID" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"fields":{"Status":"Done"}}' | python3 -m json.tool
+```
+
+### Upsert by a merge field (no ID needed)
+```bash
+curl -s -X PATCH "https://api.airtable.com/v0/$BASE_ID/$TABLE" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "performUpsert": {"fieldsToMergeOn": ["Email"]},
+    "records": [
+      {"fields": {"Email": "user@example.com", "Status": "Active"}}
+    ]
+  }' | python3 -m json.tool
+```
+`performUpsert` creates records whose merge-field values are new, patches records whose merge-field values already exist. Great for idempotent syncs.
+
+### Delete a record
+```bash
+curl -s -X DELETE "https://api.airtable.com/v0/$BASE_ID/$TABLE/$RECORD_ID" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+### Delete up to 10 records in one call
+```bash
+curl -s -X DELETE "https://api.airtable.com/v0/$BASE_ID/$TABLE?records%5B%5D=rec1&records%5B%5D=rec2" \
+  -H "Authorization: Bearer $AIRTABLE_API_KEY" | python3 -m json.tool
+```
+
+## Pagination
+
+List endpoints return at most **100 records per page**. If the response includes `"offset": "..."`, pass it back on the next call. Loop until the field is absent:
+
+```bash
+OFFSET=""
+while :; do
+  URL="https://api.airtable.com/v0/$BASE_ID/$TABLE?pageSize=100"
+  [ -n "$OFFSET" ] && URL="$URL&offset=$OFFSET"
+  RESP=$(curl -s "$URL" -H "Authorization: Bearer $AIRTABLE_API_KEY")
+  echo "$RESP" | python3 -c 'import json,sys; d=json.load(sys.stdin); [print(r["id"], r["fields"].get("Name","")) for r in d["records"]]'
+  OFFSET=$(echo "$RESP" | python3 -c 'import json,sys; d=json.load(sys.stdin); print(d.get("offset",""))')
+  [ -z "$OFFSET" ] && break
+done
+```
+
+## Typical Hermes Workflow
+
+1. **Confirm auth.** `curl -s -o /dev/null -w "%{http_code}\n" https://api.airtable.com/v0/meta/bases -H "Authorization: Bearer $AIRTABLE_API_KEY"` — expect `200`.
+2. **Find the base.** List bases (step above) OR ask the user for the `app...` ID directly if the token lacks `schema.bases:read`.
+3. **Inspect the schema.** `GET /v0/meta/bases/$BASE_ID/tables` — cache the exact field names and primary-field name locally in the session before mutating anything.
+4. **Read before you write.** For "update X where Y", `filterByFormula` first to resolve the `rec...` ID, then `PATCH /v0/$BASE_ID/$TABLE/$RECORD_ID`. Never guess record IDs.
+5. **Batch writes.** Combine related creates into one 10-record POST to stay under the 5 req/sec budget.
+6. **Destructive ops.** Deletions can't be undone via API. If the user says "delete all Xs", echo back the filter + record count and confirm before firing.
+
+## Pitfalls
+
+- **`filterByFormula` MUST be URL-encoded.** Field names with spaces or non-ASCII also need encoding (`{My Field}` → `%7BMy%20Field%7D`). Use Python stdlib (pattern above) — never hand-escape.
+- **Empty fields are omitted from responses.** A missing `"Assignee"` key doesn't mean the field doesn't exist — it means this record's value is empty. Check the schema (step 3) before concluding a field is missing.
+- **PATCH vs PUT.** `PATCH` merges supplied fields into the record. `PUT` replaces the record entirely and clears any field you didn't include. Default to `PATCH`.
+- **Single-select options must exist.** Writing `"Status": "Shipping"` when `Shipping` isn't in the field's option list errors with `INVALID_MULTIPLE_CHOICE_OPTIONS` unless you pass `"typecast": true` (which auto-creates the option).
+- **Per-base token scoping.** A `403` on one base while another works means the token's Access list doesn't include that base — not a scope or auth issue. Send the user to https://airtable.com/create/tokens to grant it.
+- **Rate limits are per base, not per token.** 5 req/sec on `baseA` and 5 req/sec on `baseB` is fine; 6 req/sec on `baseA` alone will throttle. Monitor the `Retry-After` header on `429`.
+
+## Important Notes for Hermes
+
+- **Always use the `terminal` tool with `curl`.** Do NOT use `web_extract` (it can't send auth headers) or `browser_navigate` (needs UI auth and is slow).
+- **`AIRTABLE_API_KEY` flows from `~/.hermes/.env` into the subprocess automatically** when this skill is loaded — no need to re-export it before each `curl` call.
+- **Escape curly braces in formulas carefully.** In a heredoc body, `{Status}` is literal. In a shell argument, `{Status}` is safe outside `{...}` brace-expansion context — but pass dynamic strings through `python3 urllib.parse.quote` before splicing into a URL.
+- **Pretty-print with `python3 -m json.tool`** (always present) rather than `jq` (optional). Only reach for `jq` when you need filtering/projection.
+- **Pagination is per-page, not global.** Airtable's 100-record cap is a hard limit; there is no way to bump it. Loop with `offset` until the field is absent.
+- **Read the `errors` array** on non-2xx responses — Airtable returns structured error codes like `AUTHENTICATION_REQUIRED`, `INVALID_PERMISSIONS`, `MODEL_ID_NOT_FOUND`, `INVALID_MULTIPLE_CHOICE_OPTIONS` that tell you exactly what's wrong.

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

@@ -289,6 +289,7 @@ def exchange_auth_code(code: str):
         sys.exit(1)
 
     pending_auth = _load_pending_auth()
+    raw_callback = code
     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.")
@@ -298,19 +299,13 @@ def exchange_auth_code(code: str):
     from google_auth_oauthlib.flow import Flow
     from urllib.parse import parse_qs, urlparse
 
-    # Extract granted scopes from the callback URL if present
-    if returned_state and "scope" in parse_qs(urlparse(code).query if isinstance(code, str) and code.startswith("http") else {}):
-        granted_scopes = parse_qs(urlparse(code).query)["scope"][0].split()
-    else:
-        # Try to extract from code_or_url parameter
-        if isinstance(code, str) and code.startswith("http"):
-            params = parse_qs(urlparse(code).query)
-            if "scope" in params:
-                granted_scopes = params["scope"][0].split()
-            else:
-                granted_scopes = SCOPES
-        else:
-            granted_scopes = SCOPES
+    # Extract granted scopes from the callback URL if the user pasted the full redirect URL.
+    granted_scopes = list(SCOPES)
+    if isinstance(raw_callback, str) and raw_callback.startswith("http"):
+        params = parse_qs(urlparse(raw_callback).query)
+        scope_val = (params.get("scope") or [""])[0].strip()
+        if scope_val:
+            granted_scopes = scope_val.split()
 
     flow = Flow.from_client_secrets_file(
         str(CLIENT_SECRET_PATH),

skills/productivity/maps/scripts/maps_client.py

@@ -926,13 +926,18 @@ def cmd_timezone(args):
                 os_ = offset_info.get("seconds", 0)
                 sign = "+" if oh >= 0 else "-"
                 utc_offset = f"{sign}{abs(oh):02d}:{om:02d}"
+                if os_:
+                    utc_offset = f"{utc_offset}:{os_:02d}"
             elif tz_data.get("standardUtcOffset"):
                 offset_info2 = tz_data["standardUtcOffset"]
                 if isinstance(offset_info2, dict):
                     oh = offset_info2.get("hours", 0)
                     om = abs(offset_info2.get("minutes", 0))
+                    os_ = offset_info2.get("seconds", 0)
                     sign = "+" if oh >= 0 else "-"
                     utc_offset = f"{sign}{abs(oh):02d}:{om:02d}"
+                    if os_:
+                        utc_offset = f"{utc_offset}:{os_:02d}"
             timezone_src = "timeapi.io"
     except (RuntimeError, KeyError, TypeError):
         pass  # API may be down; continue to fallback

skills/software-development/debugging-hermes-tui-commands/SKILL.md

@@ -0,0 +1,151 @@
+---
+name: debugging-hermes-tui-commands
+description: Use when debugging or adding Hermes TUI slash commands across the Python backend (hermes_cli/commands.py), the tui_gateway bridge, and the TypeScript/Ink frontend. Covers autocomplete gaps, gateway dispatch issues, and live UI-state wiring.
+version: 1.0.0
+author: Hermes Agent
+license: MIT
+metadata:
+  hermes:
+    tags: [debugging, hermes-agent, tui, slash-commands, typescript, python]
+    related_skills: [python-debugpy, node-inspect-debugger, systematic-debugging]
+---
+
+# Debugging Hermes TUI Slash Commands
+
+## Overview
+
+Hermes slash commands span three layers — Python command registry, tui_gateway JSON-RPC bridge, and the Ink/TypeScript frontend. When a command misbehaves (missing from autocomplete, works in CLI but not TUI, config persists but UI doesn't update), the bug is almost always one layer being out of sync with another.
+
+Use this skill when you encounter issues with slash commands in the Hermes TUI, particularly when commands aren't showing in autocomplete, aren't working properly in the TUI, or need to be added/updated.
+
+## When to Use
+
+- A slash command exists in one part of the codebase but doesn't work fully
+- A command needs to be added to both backend and frontend
+- Command autocomplete isn't working for specific commands
+- Command behavior is inconsistent between CLI and TUI
+- A command persists config but doesn't apply live in the TUI
+
+## Architecture Overview
+
+```
+Python backend (hermes_cli/commands.py)     <- canonical COMMAND_REGISTRY
+       │
+       ▼
+TUI gateway (tui_gateway/server.py)         <- slash.exec / command.dispatch
+       │
+       ▼
+TUI frontend (ui-tui/src/app/slash/)        <- local handlers + fallthrough
+```
+
+Command definitions must be registered consistently across Python and TypeScript to work properly. The Python `COMMAND_REGISTRY` is the source of truth for: CLI dispatch, gateway help, Telegram BotCommand menu, Slack subcommand map, and autocomplete data shipped to Ink.
+
+## Investigation Steps
+
+1. **Check if the command exists in the TUI frontend:**
+   ```bash
+   search_files --pattern "/commandname" --file_glob "*.ts" --path ui-tui/
+   search_files --pattern "/commandname" --file_glob "*.tsx" --path ui-tui/
+   ```
+
+2. **Examine the TUI command definition:**
+   ```bash
+   read_file ui-tui/src/app/slash/commands/core.ts
+   # If not there:
+   search_files --pattern "commandname" --path ui-tui/src/app/slash/commands --target files
+   ```
+
+3. **Check if the command exists in the Python backend:**
+   ```bash
+   search_files --pattern "CommandDef" --file_glob "*.py" --path hermes_cli/
+   search_files --pattern "commandname" --path hermes_cli/commands.py --context 3
+   ```
+
+4. **Examine the gateway implementation:**
+   ```bash
+   search_files --pattern "complete.slash|slash.exec" --path tui_gateway/
+   ```
+
+## Fix: Missing Command Autocomplete
+
+If a command exists in the TUI but doesn't show in autocomplete:
+
+1. Add a `CommandDef` entry to `COMMAND_REGISTRY` in `hermes_cli/commands.py`:
+   ```python
+   CommandDef("commandname", "Description of the command", "Session",
+              cli_only=True, aliases=("alias",),
+              args_hint="[arg1|arg2|arg3]",
+              subcommands=("arg1", "arg2", "arg3")),
+   ```
+
+2. Pick `cli_only` vs gateway availability carefully:
+   - `cli_only=True` — only in the interactive CLI/TUI
+   - `gateway_only=True` — only in messaging platforms
+   - neither — available everywhere
+   - `gateway_config_gate="display.foo"` — config-gated availability in the gateway
+
+3. Ensure `subcommands` matches the expected tab-completion options shown by the TUI.
+
+4. If the command runs server-side, add a handler in `HermesCLI.process_command()` in `cli.py`:
+   ```python
+   elif canonical == "commandname":
+       self._handle_commandname(cmd_original)
+   ```
+
+5. For gateway-available commands, add a handler in `gateway/run.py`:
+   ```python
+   if canonical == "commandname":
+       return await self._handle_commandname(event)
+   ```
+
+## Common Issues
+
+1. **Command shows in TUI but not in autocomplete.** The command is defined in the TUI codebase but missing from `COMMAND_REGISTRY` in `hermes_cli/commands.py`. Autocomplete data ships from Python.
+
+2. **Command shows in autocomplete but doesn't work.** Check the command handler in `tui_gateway/server.py` and the frontend handler in `ui-tui/src/app/createSlashHandler.ts`. If the command is local-only in Ink, it must be handled in `app.tsx` built-in branch; otherwise it falls through to `slash.exec` and must have a Python handler.
+
+3. **Command behavior differs between CLI and TUI.** The command might have different implementations. Check both `cli.py::process_command` and the TUI's local handler. Local TUI handlers take precedence over gateway dispatch.
+
+4. **Command persists config but doesn't apply live.** For TUI-local commands, updating `config.set` is not enough. Also patch the relevant nanostore state immediately (usually `patchUiState(...)`) and pass any new state through rendering components. Example: `/details collapsed` must update live detail visibility, not just save `details_mode`; in-session global `/details <mode>` may need a separate command-override flag so live commands can override built-in section defaults while startup/config sync preserves default-expanded thinking/tools behavior.
+
+5. **Gateway dispatch silently ignores the command.** The gateway only dispatches commands it knows about. Check `GATEWAY_KNOWN_COMMANDS` (derived from `COMMAND_REGISTRY` automatically) includes the canonical name. If the command is `cli_only` with a `gateway_config_gate`, verify the gated config value is truthy.
+
+## Debugging Tactics
+
+When surface-level inspection doesn't reveal the bug:
+
+- **Python side hangs or misbehaves:** use the `python-debugpy` skill to break inside `_SlashWorker.exec` or the command handler. `remote-pdb` set at the handler entry is the fastest path.
+- **Ink side not reacting:** use the `node-inspect-debugger` skill to break in `app.tsx`'s slash dispatch or the local command branch. `sb('dist/app.js', <line>)` after `npm run build`.
+- **Registry mismatch / unclear which side is wrong:** compare the canonical `COMMAND_REGISTRY` entry against the TUI's local command list side-by-side.
+
+## Pitfalls
+
+- Don't forget to set the appropriate category for the command in `CommandDef` (e.g., "Session", "Configuration", "Tools & Skills", "Info", "Exit")
+- Make sure any aliases are properly registered in the `aliases` tuple — no other file changes are needed, everything downstream (Telegram menu, Slack mapping, autocomplete, help) derives from it
+- For commands with subcommands, ensure the `subcommands` tuple in `CommandDef` matches what's in the TUI code
+- `cli_only=True` commands won't work in gateway/messaging platforms — unless you add a `gateway_config_gate` and the gate is truthy
+- After adding live UI state, search every consumer of the old prop/helper and thread the new state through all render paths, not just the active streaming path. TUI detail rendering has at least two important paths: live `StreamingAssistant`/`ToolTrail` and transcript/pending `MessageLine` rows. A `/clean` pass should explicitly check both.
+- Rebuild the TUI (`npm --prefix ui-tui run build`) before testing — tsx watch mode may lag on first launch
+
+## Verification
+
+After fixing:
+
+1. Rebuild the TUI:
+   ```bash
+   cd /home/bb/hermes-agent && npm --prefix ui-tui run build
+   ```
+
+2. Run the TUI and test the command:
+   ```bash
+   hermes --tui
+   ```
+
+3. Type `/` and verify the command appears in autocomplete suggestions with the expected description and args hint.
+
+4. Execute the command and confirm:
+   - Expected behavior fires
+   - Any persisted config updates correctly (`read_file ~/.hermes/config.yaml`)
+   - Live UI state reflects the change immediately (not just after restart)
+
+5. If the command is also gateway-available, test it from at least one messaging platform (or run the gateway tests: `scripts/run_tests.sh tests/gateway/`).

skills/software-development/hermes-agent-skill-authoring/SKILL.md

@@ -0,0 +1,164 @@
+---
+name: hermes-agent-skill-authoring
+description: Use when authoring or updating a SKILL.md inside the hermes-agent repo itself (skills/ tree, committed to a branch). Covers required frontmatter, validator limits, peer-matching structure, and the write_file-vs-skill_manage distinction for in-repo skills.
+version: 1.0.0
+author: Hermes Agent
+license: MIT
+metadata:
+  hermes:
+    tags: [skills, authoring, hermes-agent, conventions, skill-md]
+    related_skills: [writing-plans, requesting-code-review]
+---
+
+# Authoring Hermes-Agent Skills (in-repo)
+
+## Overview
+
+There are two places a SKILL.md can live:
+
+1. **User-local:** `~/.hermes/skills/<maybe-category>/<name>/SKILL.md` — personal, not shared. Created via `skill_manage(action='create')`.
+2. **In-repo (this skill is about this case):** `/home/bb/hermes-agent/skills/<category>/<name>/SKILL.md` — committed, shipped with the package. Use `write_file` + `git add`. `skill_manage(action='create')` does NOT target this tree.
+
+## When to Use
+
+- User asks you to add a skill "in this branch / repo / commit"
+- You're committing a reusable workflow that should ship with hermes-agent
+- You're editing an existing skill under `/home/bb/hermes-agent/skills/` (use `patch` for small edits, `write_file` for rewrites; `skill_manage` still works for patch on in-repo skills, but not for `create`)
+
+## Required Frontmatter
+
+Source of truth: `tools/skill_manager_tool.py::_validate_frontmatter`. Hard requirements:
+
+- Starts with `---` as the first bytes (no leading blank line).
+- Closes with `\n---\n` before the body.
+- Parses as a YAML mapping.
+- `name` field present.
+- `description` field present, ≤ **1024 chars** (`MAX_DESCRIPTION_LENGTH`).
+- Non-empty body after the closing `---`.
+
+Peer-matched shape used by every skill under `skills/software-development/`:
+
+```yaml
+---
+name: my-skill-name               # lowercase, hyphens, ≤64 chars (MAX_NAME_LENGTH)
+description: Use when <trigger>. <one-line behavior>.
+version: 1.0.0
+author: Hermes Agent
+license: MIT
+metadata:
+  hermes:
+    tags: [short, descriptive, tags]
+    related_skills: [other-skill, another-skill]
+---
+```
+
+`version` / `author` / `license` / `metadata` are NOT enforced by the validator, but every peer has them — omit and your skill sticks out.
+
+## Size Limits
+
+- Description: ≤ 1024 chars (enforced).
+- Full SKILL.md: ≤ 100,000 chars (enforced as `MAX_SKILL_CONTENT_CHARS`, ~36k tokens).
+- Peer skills in `software-development/` sit at **8-14k chars**. Aim for that range. If you're pushing past 20k, split into `references/*.md` and reference them from SKILL.md.
+
+## Peer-Matched Structure
+
+Every in-repo skill follows roughly:
+
+```
+# <Title>
+
+## Overview
+One or two paragraphs: what and why.
+
+## When to Use
+- Bulleted triggers
+- "Don't use for:" counter-triggers
+
+## <Topic sections specific to the skill>
+- Quick-reference tables are common
+- Code blocks with exact commands
+- Hermes-specific recipes (tests via scripts/run_tests.sh, ui-tui paths, etc.)
+
+## Common Pitfalls
+Numbered list of mistakes and their fixes.
+
+## Verification Checklist
+- [ ] Checkbox list of post-action verifications
+
+## One-Shot Recipes (optional)
+Named scenarios → concrete command sequences.
+```
+
+Not every section is mandatory, but `Overview` + `When to Use` + actionable body + pitfalls are the minimum for the skill to feel like a peer.
+
+## Directory Placement
+
+```
+skills/<category>/<skill-name>/SKILL.md
+```
+
+Categories currently in repo (confirm with `ls skills/`): `autonomous-ai-agents`, `creative`, `data-science`, `devops`, `dogfood`, `email`, `gaming`, `github`, `leisure`, `mcp`, `media`, `mlops/*`, `note-taking`, `productivity`, `red-teaming`, `research`, `smart-home`, `social-media`, `software-development`.
+
+Pick the closest existing category. Don't invent new top-level categories casually.
+
+## Workflow
+
+1. **Survey peers** in the target category:
+   ```
+   ls skills/<category>/
+   ```
+   Read 2-3 peer SKILL.md files to match tone and structure.
+2. **Check validator constraints** in `tools/skill_manager_tool.py` if unsure.
+3. **Draft** with `write_file` to `skills/<category>/<name>/SKILL.md`.
+4. **Validate locally**:
+   ```python
+   import yaml, re, pathlib
+   content = pathlib.Path("skills/<category>/<name>/SKILL.md").read_text()
+   assert content.startswith("---")
+   m = re.search(r'\n---\s*\n', content[3:])
+   fm = yaml.safe_load(content[3:m.start()+3])
+   assert "name" in fm and "description" in fm
+   assert len(fm["description"]) <= 1024
+   assert len(content) <= 100_000
+   ```
+5. **Git add + commit** on the active branch.
+6. **Note:** the CURRENT session's skill loader is cached — `skill_view` / `skills_list` will not see the new skill until a new session. This is expected, not a bug.
+
+## Cross-Referencing Other Skills
+
+`metadata.hermes.related_skills` unions both trees (`skills/` in-repo and `~/.hermes/skills/`) at load time. You CAN reference a user-local skill from an in-repo skill, but it won't resolve for other users who clone the repo fresh. Prefer referencing only in-repo skills from in-repo skills. If a frequently-referenced skill lives only in `~/.hermes/skills/`, consider promoting it to the repo.
+
+## Editing Existing In-Repo Skills
+
+- **Small fix (typo, added pitfall, tightened trigger):** `skill_manage(action='patch', name=..., old_string=..., new_string=...)` works fine on in-repo skills.
+- **Major rewrite:** `write_file` the whole SKILL.md. `skill_manage(action='edit')` also works but requires supplying the full new content.
+- **Adding supporting files:** `write_file` to `skills/<category>/<name>/references/<file>.md`, `templates/<file>`, or `scripts/<file>`. `skill_manage(action='write_file')` also works and enforces the references/templates/scripts/assets subdir allowlist.
+- **Always commit** the edit — in-repo skills are source, not runtime state.
+
+## Common Pitfalls
+
+1. **Using `skill_manage(action='create')` for an in-repo skill.** It writes to `~/.hermes/skills/`, not the repo tree. Use `write_file` for in-repo creation.
+
+2. **Leading whitespace before `---`.** The validator checks `content.startswith("---")`; any leading blank line or BOM fails validation.
+
+3. **Description too generic.** Peer descriptions start with "Use when ..." and describe the *trigger class*, not the one task. "Use when debugging X" > "Debug X".
+
+4. **Forgetting the author/license/metadata block.** Not validator-enforced, but every peer has it; omitting makes the skill look half-finished.
+
+5. **Writing a skill that duplicates a peer.** Before creating, `ls skills/<category>/` and open 2-3 peers. Prefer extending an existing skill to creating a narrow sibling.
+
+6. **Expecting the current session to see the new skill.** It won't. The skill loader is initialized at session start. Verify in a fresh session or via `skill_view` using the exact path.
+
+7. **Linking to skills that don't exist in-repo.** `related_skills: [some-user-local-skill]` works for you but breaks for other clones. Prefer only in-repo links.
+
+## Verification Checklist
+
+- [ ] File is at `skills/<category>/<name>/SKILL.md` (not in `~/.hermes/skills/`)
+- [ ] Frontmatter starts at byte 0 with `---`, closes with `\n---\n`
+- [ ] `name`, `description`, `version`, `author`, `license`, `metadata.hermes.{tags, related_skills}` all present
+- [ ] Name ≤ 64 chars, lowercase + hyphens
+- [ ] Description ≤ 1024 chars and starts with "Use when ..."
+- [ ] Total file ≤ 100,000 chars (aim for 8-15k)
+- [ ] Structure: `# Title` → `## Overview` → `## When to Use` → body → `## Common Pitfalls` → `## Verification Checklist`
+- [ ] `related_skills` references resolve in-repo (or are explicitly OK to be user-local)
+- [ ] `git add skills/<category>/<name>/ && git commit` completed on the intended branch

skills/software-development/node-inspect-debugger/SKILL.md

@@ -0,0 +1,318 @@
+---
+name: node-inspect-debugger
+description: Use when debugging Node.js code (ui-tui, tui_gateway child processes, any Node script/test) with real breakpoints, stepping, scope inspection, and expression evaluation. Drives `node --inspect` via the Chrome DevTools Protocol from the terminal — no browser required.
+version: 1.0.0
+author: Hermes Agent
+license: MIT
+metadata:
+  hermes:
+    tags: [debugging, nodejs, node-inspect, cdp, breakpoints, ui-tui]
+    related_skills: [systematic-debugging, python-debugpy, debugging-hermes-tui-commands]
+---
+
+# Node.js Inspect Debugger
+
+## Overview
+
+When `console.log` isn't enough, drive Node's built-in V8 inspector programmatically from the terminal. You get real breakpoints, step in/over/out, call-stack walking, local/closure scope dumps, and arbitrary expression evaluation in the paused frame.
+
+Two tools, pick one:
+
+- **`node inspect`** — built-in, zero install, CLI REPL. Best for quick poking.
+- **`ndb` / CDP via `chrome-remote-interface`** — scriptable from Node/Python; best when you want to automate many breakpoints, collect state across runs, or debug non-interactively from an agent loop.
+
+**Prefer `node inspect` first.** It's always available and the REPL is fast.
+
+## When to Use
+
+- A Node test fails and you need to see intermediate state
+- ui-tui crashes or behaves wrong and you want to inspect React/Ink state pre-render
+- tui_gateway child processes (`_SlashWorker`, PTY bridge workers) misbehave
+- You need to inspect a value in a closure that `console.log` can't reach without patching
+- Perf: attach to a running process to capture a CPU profile or heap snapshot
+
+**Don't use for:** things `console.log` solves in under a minute. Breakpoint-driven debugging is heavier; use it when the payoff is real.
+
+## Quick Reference: `node inspect` REPL
+
+Launch paused on first line:
+
+```bash
+node inspect path/to/script.js
+# or with tsx
+node --inspect-brk $(which tsx) path/to/script.ts
+```
+
+The `debug>` prompt accepts:
+
+| Command | Action |
+|---|---|
+| `c` or `cont` | continue |
+| `n` or `next` | step over |
+| `s` or `step` | step into |
+| `o` or `out` | step out |
+| `pause` | pause running code |
+| `sb('file.js', 42)` | set breakpoint at file.js line 42 |
+| `sb(42)` | set breakpoint at line 42 of current file |
+| `sb('functionName')` | break when function is called |
+| `cb('file.js', 42)` | clear breakpoint |
+| `breakpoints` | list all breakpoints |
+| `bt` | backtrace (call stack) |
+| `list(5)` | show 5 lines of source around current position |
+| `watch('expr')` | evaluate expr on every pause |
+| `watchers` | show watched expressions |
+| `repl` | drop into REPL in current scope (Ctrl+C to exit REPL) |
+| `exec expr` | evaluate expression once |
+| `restart` | restart script |
+| `kill` | kill the script |
+| `.exit` | quit debugger |
+
+**In the `repl` sub-mode:** type any JS expression, including access to locals/closure variables. `Ctrl+C` exits back to `debug>`.
+
+## Attaching to a Running Process
+
+When the process is already running (e.g. a long-lived dev server or the TUI gateway):
+
+```bash
+# 1. Send SIGUSR1 to enable the inspector on an existing process
+kill -SIGUSR1 <pid>
+# Node prints: Debugger listening on ws://127.0.0.1:9229/<uuid>
+
+# 2. Attach the debugger CLI
+node inspect -p <pid>
+# or by URL
+node inspect ws://127.0.0.1:9229/<uuid>
+```
+
+To start a process with the inspector from the beginning:
+
+```bash
+node --inspect script.js           # listen on 127.0.0.1:9229, keep running
+node --inspect-brk script.js       # listen AND pause on first line
+node --inspect=0.0.0.0:9230 script.js   # custom host:port
+```
+
+For TypeScript via tsx:
+
+```bash
+node --inspect-brk --import tsx script.ts
+# or older tsx
+node --inspect-brk -r tsx/cjs script.ts
+```
+
+## Programmatic CDP (scripting from terminal)
+
+When you want to automate — set many breakpoints, capture scope state, script a repro — use `chrome-remote-interface`:
+
+```bash
+npm i -g chrome-remote-interface        # or project-local
+# Start your target:
+node --inspect-brk=9229 target.js &
+```
+
+Driver script (save as `/tmp/cdp-debug.js`):
+
+```javascript
+const CDP = require('chrome-remote-interface');
+
+(async () => {
+  const client = await CDP({ port: 9229 });
+  const { Debugger, Runtime } = client;
+
+  Debugger.paused(async ({ callFrames, reason }) => {
+    const top = callFrames[0];
+    console.log(`PAUSED: ${reason} @ ${top.url}:${top.location.lineNumber + 1}`);
+
+    // Walk scopes for locals
+    for (const scope of top.scopeChain) {
+      if (scope.type === 'local' || scope.type === 'closure') {
+        const { result } = await Runtime.getProperties({
+          objectId: scope.object.objectId,
+          ownProperties: true,
+        });
+        for (const p of result) {
+          console.log(`  ${scope.type}.${p.name} =`, p.value?.value ?? p.value?.description);
+        }
+      }
+    }
+
+    // Evaluate an expression in the paused frame
+    const { result } = await Debugger.evaluateOnCallFrame({
+      callFrameId: top.callFrameId,
+      expression: 'typeof state !== "undefined" ? JSON.stringify(state) : "n/a"',
+    });
+    console.log('state =', result.value ?? result.description);
+
+    await Debugger.resume();
+  });
+
+  await Runtime.enable();
+  await Debugger.enable();
+
+  // Set a breakpoint by URL regex + line
+  await Debugger.setBreakpointByUrl({
+    urlRegex: '.*app\\.tsx$',
+    lineNumber: 119,       // 0-indexed
+    columnNumber: 0,
+  });
+
+  await Runtime.runIfWaitingForDebugger();
+})();
+```
+
+Run it:
+
+```bash
+node /tmp/cdp-debug.js
+```
+
+Hermes-specific note: `chrome-remote-interface` is NOT in `ui-tui/package.json`. Install it to a throwaway location if you don't want to dirty the project:
+
+```bash
+mkdir -p /tmp/cdp-tools && cd /tmp/cdp-tools && npm i chrome-remote-interface
+NODE_PATH=/tmp/cdp-tools/node_modules node /tmp/cdp-debug.js
+```
+
+## Debugging Hermes ui-tui
+
+The TUI is built Ink + tsx. Two common scenarios:
+
+### Debugging a single Ink component under dev
+
+`ui-tui/package.json` has `npm run dev` (tsx --watch). Add `--inspect-brk` by running tsx directly:
+
+```bash
+cd /home/bb/hermes-agent/ui-tui
+npm run build    # produce dist/ once so transpile isn't needed on first load
+node --inspect-brk dist/entry.js
+# In another terminal:
+node inspect -p <node pid>
+```
+
+Then inside `debug>`:
+
+```
+sb('dist/app.js', 220)     # or wherever the suspect render is
+cont
+```
+
+When it pauses, `repl` → inspect `props`, state refs, `useInput` handler values, etc.
+
+### Debugging a running `hermes --tui`
+
+The TUI spawns Node from the Python CLI. Easiest path:
+
+```bash
+# 1. Launch TUI
+hermes --tui &
+TUI_PID=$(pgrep -f 'ui-tui/dist/entry' | head -1)
+
+# 2. Enable inspector on that Node PID
+kill -SIGUSR1 "$TUI_PID"
+
+# 3. Find the WS URL
+curl -s http://127.0.0.1:9229/json/list | jq -r '.[0].webSocketDebuggerUrl'
+
+# 4. Attach
+node inspect ws://127.0.0.1:9229/<uuid>
+```
+
+Interacting with the TUI (typing in its window) continues to advance execution; your debugger can pause it on a breakpoint at any `sb(...)`.
+
+### Debugging `_SlashWorker` / PTY child processes
+
+Those are Python, not Node — use the `python-debugpy` skill for them. Only Node portions (Ink UI, tui_gateway client, tsx-run tests under `ui-tui/`) use this skill.
+
+## Running Vitest Tests Under the Debugger
+
+```bash
+cd /home/bb/hermes-agent/ui-tui
+# Run a single test file paused on entry
+node --inspect-brk ./node_modules/vitest/vitest.mjs run --no-file-parallelism src/app/foo.test.tsx
+```
+
+In another terminal: `node inspect -p <pid>`, then `sb('src/app/foo.tsx', 42)`, `cont`.
+
+Use `--no-file-parallelism` (vitest) or `--runInBand` (jest) so only one worker exists — debugging a pool is painful.
+
+## Heap Snapshots & CPU Profiles (Non-interactive)
+
+From the CDP driver above, swap Debugger for `HeapProfiler` / `Profiler`:
+
+```javascript
+// CPU profile for 5 seconds
+await client.Profiler.enable();
+await client.Profiler.start();
+await new Promise(r => setTimeout(r, 5000));
+const { profile } = await client.Profiler.stop();
+require('fs').writeFileSync('/tmp/cpu.cpuprofile', JSON.stringify(profile));
+// Open /tmp/cpu.cpuprofile in Chrome DevTools → Performance tab
+```
+
+```javascript
+// Heap snapshot
+await client.HeapProfiler.enable();
+const chunks = [];
+client.HeapProfiler.addHeapSnapshotChunk(({ chunk }) => chunks.push(chunk));
+await client.HeapProfiler.takeHeapSnapshot({ reportProgress: false });
+require('fs').writeFileSync('/tmp/heap.heapsnapshot', chunks.join(''));
+```
+
+## Common Pitfalls
+
+1. **Wrong line numbers in TS source.** Breakpoints hit the emitted JS, not the `.ts`. Either (a) break in the built `dist/*.js`, or (b) enable sourcemaps (`node --enable-source-maps`) and use `sb('src/app.tsx', N)` — but only with CDP clients that follow sourcemaps. `node inspect` CLI does not.
+
+2. **`--inspect` vs `--inspect-brk`.** `--inspect` starts the inspector but doesn't pause; your script races past your first breakpoint if you attach too late. Use `--inspect-brk` when you need to set breakpoints before any code runs.
+
+3. **Port collisions.** Default is `9229`. If multiple Node processes are inspecting, pass `--inspect=0` (random port) and read the actual URL from `/json/list`:
+   ```bash
+   curl -s http://127.0.0.1:9229/json/list   # lists all inspectable targets on the host
+   ```
+
+4. **Child processes.** `--inspect` on a parent does NOT inspect its children. Use `NODE_OPTIONS='--inspect-brk' node parent.js` to propagate to every child; be aware they all need unique ports (Node auto-increments when `NODE_OPTIONS='--inspect'` is inherited).
+
+5. **Background kills.** If you `Ctrl+C` out of `node inspect` while the target is paused, the target stays paused. Either `cont` first, or `kill` the target explicitly.
+
+6. **Running `node inspect` through an agent terminal.** It's a PTY-friendly REPL. In Hermes, launch it with `terminal(pty=true)` or `background=true` + `process(action='submit', data='...')`. Non-PTY foreground mode will work for one-shot commands but not for interactive stepping.
+
+7. **Security.** `--inspect=0.0.0.0:9229` exposes arbitrary code execution. Always bind to `127.0.0.1` (the default) unless you have an isolated network.
+
+## Verification Checklist
+
+After setting up a debug session, verify:
+
+- [ ] `curl -s http://127.0.0.1:9229/json/list` returns exactly the target you expect
+- [ ] First breakpoint actually hits (if it doesn't, you likely missed `--inspect-brk` or attached after execution completed)
+- [ ] Source listing at pause shows the right file (mismatch = sourcemap issue, see pitfall 1)
+- [ ] `exec process.pid` in `repl` returns the PID you meant to attach to
+
+## One-Shot Recipes
+
+**"Why is this variable undefined at line X?"**
+```bash
+node --inspect-brk script.js &
+node inspect -p $!
+# debug>
+sb('script.js', X)
+cont
+# paused. Now:
+repl
+> myVariable
+> Object.keys(this)
+```
+
+**"What's the call path into this function?"**
+```
+debug> sb('suspectFn')
+debug> cont
+# paused on entry
+debug> bt
+```
+
+**"This async chain hangs — where?"**
+```
+# Start with --inspect (no -brk), let it run to the hang, then:
+debug> pause
+debug> bt
+# Now you see the stuck frame
+```

skills/software-development/python-debugpy/SKILL.md

@@ -0,0 +1,374 @@
+---
+name: python-debugpy
+description: Use when debugging Python code (run_agent.py, cli.py, tui_gateway, tests, scripts) with real breakpoints, stepping, scope inspection, and post-mortem analysis. Covers `pdb` for interactive REPL debugging and `debugpy` for remote/headless DAP-driven sessions.
+version: 1.0.0
+author: Hermes Agent
+license: MIT
+metadata:
+  hermes:
+    tags: [debugging, python, pdb, debugpy, breakpoints, dap, post-mortem]
+    related_skills: [systematic-debugging, node-inspect-debugger, debugging-hermes-tui-commands]
+---
+
+# Python Debugger (pdb + debugpy)
+
+## Overview
+
+Three tools, picked by situation:
+
+| Tool | When |
+|---|---|
+| **`breakpoint()` + pdb** | Local, interactive, simplest. Add `breakpoint()` in the source, run normally, get a REPL at that line. |
+| **`python -m pdb`** | Launch an existing script under pdb with no source edits. Useful for quick poking. |
+| **`debugpy`** | Remote / headless / "attach to already-running process." Talks DAP, scriptable from terminal, works for long-lived processes (gateway, daemon, PTY children). |
+
+**Start with `breakpoint()`.** It's the cheapest thing that works.
+
+## When to Use
+
+- A test fails and the traceback doesn't reveal why a value is wrong
+- You need to step through a function and watch a collection mutate
+- A long-running process (hermes gateway, tui_gateway) misbehaves and you can't restart it
+- Post-mortem: an exception fired in prod-ish code and you want to inspect locals at the crash site
+- A subprocess / child (Python `_SlashWorker`, PTY bridge worker) is the actual bug site
+
+**Don't use for:** things `print()` / `logging.debug` solve in under a minute, or things `pytest -vv --tb=long --showlocals` already reveals.
+
+## pdb Quick Reference
+
+Inside any pdb prompt (`(Pdb)`):
+
+| Command | Action |
+|---|---|
+| `h` / `h cmd` | help |
+| `n` | next line (step over) |
+| `s` | step into |
+| `r` | return from current function |
+| `c` | continue |
+| `unt N` | continue until line N |
+| `j N` | jump to line N (same function only) |
+| `l` / `ll` | list source around current line / full function |
+| `w` | where (stack trace) |
+| `u` / `d` | move up / down in the stack |
+| `a` | print args of the current function |
+| `p expr` / `pp expr` | print / pretty-print expression |
+| `display expr` | auto-print expr on every stop |
+| `b file:line` | set breakpoint |
+| `b func` | break on function entry |
+| `b file:line, cond` | conditional breakpoint |
+| `cl N` | clear breakpoint N |
+| `tbreak file:line` | one-shot breakpoint |
+| `!stmt` | execute arbitrary Python (assignments included) |
+| `interact` | drop into full Python REPL in current scope (Ctrl+D to exit) |
+| `q` | quit |
+
+The `interact` command is the most powerful — you can import anything, inspect complex objects, even call methods that mutate state. Locals are read-only by default; use `!x = 42` from the `(Pdb)` prompt to mutate.
+
+## Recipe 1: Local breakpoint
+
+Easiest. Edit the file:
+
+```python
+def compute(x, y):
+    result = some_helper(x)
+    breakpoint()           # <-- drops into pdb here
+    return result + y
+```
+
+Run the code normally. You land at the `breakpoint()` line with full access to locals.
+
+**Don't forget to remove `breakpoint()` before committing.** Use `git diff` or a pre-commit grep:
+```bash
+rg -n 'breakpoint\(\)' --type py
+```
+
+## Recipe 2: Launch a script under pdb (no source edits)
+
+```bash
+python -m pdb path/to/script.py arg1 arg2
+# Lands at first line of script
+(Pdb) b path/to/script.py:42
+(Pdb) c
+```
+
+## Recipe 3: Debug a pytest test
+
+The hermes test runner and pytest both support this:
+
+```bash
+# Drop to pdb on failure (or on any raised exception):
+scripts/run_tests.sh tests/path/to/test_file.py::test_name --pdb
+
+# Drop to pdb at the START of the test:
+scripts/run_tests.sh tests/path/to/test_file.py::test_name --trace
+
+# Show locals in tracebacks without pdb:
+scripts/run_tests.sh tests/path/to/test_file.py --showlocals --tb=long
+```
+
+Note: `scripts/run_tests.sh` uses xdist (`-n 4`) by default, and pdb does NOT work under xdist. Add `-p no:xdist` or run a single test with `-n 0`:
+
+```bash
+scripts/run_tests.sh tests/foo_test.py::test_bar --pdb -p no:xdist
+# or
+source .venv/bin/activate
+python -m pytest tests/foo_test.py::test_bar --pdb
+```
+
+This bypasses the hermetic-env guarantees — fine for debugging, but re-run under the wrapper to confirm before pushing.
+
+## Recipe 4: Post-mortem on any exception
+
+```python
+import pdb, sys
+try:
+    run_the_thing()
+except Exception:
+    pdb.post_mortem(sys.exc_info()[2])
+```
+
+Or wrap a whole script:
+
+```bash
+python -m pdb -c continue script.py
+# When it crashes, pdb catches it and you're in the frame of the exception
+```
+
+Or set a global hook in a repl/jupyter:
+
+```python
+import sys
+def excepthook(etype, value, tb):
+    import pdb; pdb.post_mortem(tb)
+sys.excepthook = excepthook
+```
+
+## Recipe 5: Remote debug with debugpy (attach to running process)
+
+For long-lived processes: Hermes gateway, tui_gateway, a daemon, a process that's already misbehaving and can't be restarted clean.
+
+### Setup
+
+```bash
+source /home/bb/hermes-agent/.venv/bin/activate
+pip install debugpy
+```
+
+### Pattern A: Source-edit — process waits for debugger at launch
+
+Add near the top of the entry point (or inside the function you want to debug):
+
+```python
+import debugpy
+debugpy.listen(("127.0.0.1", 5678))
+print("debugpy listening on 5678, waiting for client...", flush=True)
+debugpy.wait_for_client()
+debugpy.breakpoint()       # optional: pause immediately once attached
+```
+
+Start the process; it blocks on `wait_for_client()`.
+
+### Pattern B: No source edit — launch with `-m debugpy`
+
+```bash
+python -m debugpy --listen 127.0.0.1:5678 --wait-for-client your_script.py arg1
+```
+
+Equivalent for module entry:
+
+```bash
+python -m debugpy --listen 127.0.0.1:5678 --wait-for-client -m your.module
+```
+
+### Pattern C: Attach to an already-running process
+
+Needs the PID and debugpy preinstalled in the target's environment:
+
+```bash
+python -m debugpy --listen 127.0.0.1:5678 --pid <pid>
+# debugpy injects itself into the process. Then attach a client as below.
+```
+
+Some kernels/security configs block the ptrace-based injection (`/proc/sys/kernel/yama/ptrace_scope`). Fix with:
+```bash
+echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
+```
+
+### Connecting a client from the terminal
+
+The easiest terminal-side DAP client is VS Code CLI or a small script. From inside Hermes you have two practical options:
+
+**Option 1: `debugpy`'s own CLI REPL** — not an official feature, but a tiny DAP client script:
+
+```python
+# /tmp/dap_client.py
+import socket, json, itertools, time, sys
+
+HOST, PORT = "127.0.0.1", 5678
+s = socket.create_connection((HOST, PORT))
+seq = itertools.count(1)
+
+def send(msg):
+    msg["seq"] = next(seq)
+    body = json.dumps(msg).encode()
+    s.sendall(f"Content-Length: {len(body)}\r\n\r\n".encode() + body)
+
+def recv():
+    header = b""
+    while b"\r\n\r\n" not in header:
+        header += s.recv(1)
+    length = int(header.decode().split("Content-Length:")[1].split("\r\n")[0].strip())
+    body = b""
+    while len(body) < length:
+        body += s.recv(length - len(body))
+    return json.loads(body)
+
+send({"type": "request", "command": "initialize", "arguments": {"adapterID": "python"}})
+print(recv())
+send({"type": "request", "command": "attach", "arguments": {}})
+print(recv())
+send({"type": "request", "command": "setBreakpoints",
+      "arguments": {"source": {"path": sys.argv[1]},
+                    "breakpoints": [{"line": int(sys.argv[2])}]}})
+print(recv())
+send({"type": "request", "command": "configurationDone"})
+# ... loop reading events and sending continue/stepIn/etc.
+```
+
+This is fine for one-off automation but painful as an interactive UX.
+
+**Option 2: Attach from VS Code / Cursor / Zed** — if the user has one open, they can add a `launch.json`:
+
+```json
+{
+  "name": "Attach to Hermes",
+  "type": "debugpy",
+  "request": "attach",
+  "connect": { "host": "127.0.0.1", "port": 5678 },
+  "justMyCode": false,
+  "pathMappings": [
+    { "localRoot": "${workspaceFolder}", "remoteRoot": "/home/bb/hermes-agent" }
+  ]
+}
+```
+
+**Option 3: Ditch DAP, use `remote-pdb`** — usually what you actually want from a terminal agent:
+
+```bash
+pip install remote-pdb
+```
+
+In your code:
+```python
+from remote_pdb import set_trace
+set_trace(host="127.0.0.1", port=4444)   # blocks until connection
+```
+
+Then from the terminal:
+```bash
+nc 127.0.0.1 4444
+# You get a (Pdb) prompt exactly as if debugging locally.
+```
+
+`remote-pdb` is the cleanest agent-friendly choice when `debugpy`'s DAP protocol is overkill. Use `debugpy` only when you actually need IDE integration.
+
+## Debugging Hermes-specific Processes
+
+### Tests
+See Recipe 3. Always add `-p no:xdist` or run single tests without xdist.
+
+### `run_agent.py` / CLI — one-shot
+Easiest: add `breakpoint()` near the suspect line, then run `hermes` normally. Control returns to your terminal at the pause point.
+
+### `tui_gateway` subprocess (spawned by `hermes --tui`)
+The gateway runs as a child of the Node TUI. Options:
+
+**A. Source-edit the gateway:**
+```python
+# tui_gateway/server.py near the top of serve()
+import debugpy
+debugpy.listen(("127.0.0.1", 5678))
+debugpy.wait_for_client()
+```
+Start `hermes --tui`. The TUI will appear frozen (its backend is waiting). Attach a client; execution resumes when you `continue`.
+
+**B. Use `remote-pdb` at a specific handler:**
+```python
+from remote_pdb import set_trace
+set_trace(host="127.0.0.1", port=4444)   # in the RPC handler you want to trap
+```
+Trigger the matching slash command from the TUI, then `nc 127.0.0.1 4444` in another terminal.
+
+### `_SlashWorker` subprocess
+Same pattern — `remote-pdb` with `set_trace()` inside the worker's `exec` path. The worker is persistent across slash commands, so the first trigger blocks until you connect; subsequent slash commands pass through normally unless you re-arm.
+
+### Gateway (`gateway/run.py`)
+Long-lived. Use `remote-pdb` at a handler, or `debugpy` with `--wait-for-client` if you're restarting the gateway anyway.
+
+## Common Pitfalls
+
+1. **pdb under pytest-xdist silently does nothing.** You won't see the prompt, the test just hangs. Always use `-p no:xdist` or `-n 0`.
+
+2. **`breakpoint()` in CI / non-TTY contexts hangs the process.** Safe locally; never commit it. Add a pre-commit grep as a safety net.
+
+3. **`PYTHONBREAKPOINT=0`** disables all `breakpoint()` calls. Check the env if your breakpoint isn't hitting:
+   ```bash
+   echo $PYTHONBREAKPOINT
+   ```
+
+4. **`debugpy.listen` blocks only if you also call `wait_for_client()`.** Without it, execution continues and your first breakpoint may fire before the client is attached.
+
+5. **Attach to PID fails on hardened kernels.** `ptrace_scope=1` (Ubuntu default) allows only same-user ptrace of child processes. Workaround: `echo 0 > /proc/sys/kernel/yama/ptrace_scope` (needs root) or launch under `debugpy` from the start.
+
+6. **Threads.** `pdb` only debugs the current thread. For multithreaded code, use `debugpy` (thread-aware DAP) or set `threading.settrace()` per thread.
+
+7. **asyncio.** `pdb` works in coroutines but `await` inside pdb requires Python 3.13+ or `await` from `interact` mode on older versions. For 3.11/3.12, use `asyncio.run_coroutine_threadsafe` tricks or `!stmt`-based awaits via `asyncio.ensure_future`.
+
+8. **`scripts/run_tests.sh` strips credentials and sets `HOME=<tmpdir>`.** If your bug depends on user config or real API keys, it won't reproduce under the wrapper. Debug with raw `pytest` first to repro, then re-confirm under the wrapper.
+
+9. **Forking / multiprocessing.** pdb does not follow forks. Each child needs its own `breakpoint()` or `set_trace()`. For Hermes subagents, debug one process at a time.
+
+## Verification Checklist
+
+- [ ] After `pip install debugpy`, confirm: `python -c "import debugpy; print(debugpy.__version__)"`
+- [ ] For remote debug, confirm the port is actually listening: `ss -tlnp | grep 5678`
+- [ ] First breakpoint actually hits (if it doesn't, you likely have `PYTHONBREAKPOINT=0`, you're under xdist, or execution finished before attach)
+- [ ] `where` / `w` shows the expected call stack
+- [ ] Post-debug cleanup: no stray `breakpoint()` / `set_trace()` in committed code
+  ```bash
+  rg -n 'breakpoint\(\)|set_trace\(|debugpy\.listen' --type py
+  ```
+
+## One-Shot Recipes
+
+**"Why is this dict missing a key?"**
+```python
+# add above the KeyError site
+breakpoint()
+# then in pdb:
+(Pdb) pp d
+(Pdb) pp list(d.keys())
+(Pdb) w                # how did we get here
+```
+
+**"This test passes in isolation but fails in the suite."**
+```bash
+scripts/run_tests.sh tests/the_test.py --pdb -p no:xdist
+# But if it only fails WITH other tests:
+source .venv/bin/activate
+python -m pytest tests/ -x --pdb -p no:xdist
+# Now it pdb-traps at the exact failing test after state accumulated.
+```
+
+**"My async handler deadlocks."**
+```python
+# Add at handler entry
+import remote_pdb; remote_pdb.set_trace(host="127.0.0.1", port=4444)
+```
+Trigger the handler. `nc 127.0.0.1 4444`, then `w` to see the suspended frame, `!import asyncio; asyncio.all_tasks()` to see what else is pending.
+
+**"Post-mortem on a crash in an Ink child process / subprocess."**
+```bash
+PYTHONFAULTHANDLER=1 python -m pdb -c continue path/to/entrypoint.py
+# On crash, pdb lands at the frame of the exception with full locals
+```

skills/yuanbao/SKILL.md

@@ -0,0 +1,107 @@
+---
+name: yuanbao
+description: Yuanbao (元宝) group interaction — @mention users, query group info and members
+version: 1.0.0
+metadata:
+  hermes:
+    tags: [yuanbao, mention, at, group, members, 元宝, 派, 艾特]
+    related_skills: []
+---
+
+# Yuanbao Group Interaction
+
+## CRITICAL: How Messaging Works
+
+**Your text reply IS the message sent to the group/user.** The gateway automatically delivers your response text to the chat. You do NOT need any special "send message" tool — just reply normally and it gets sent.
+
+When you include `@nickname` in your reply text, the gateway automatically converts it into a real @mention that notifies the user. This is built-in — you have full @mention capability.
+
+**NEVER say you cannot send messages or @mention users. NEVER suggest the user do it manually. NEVER add disclaimers about permissions. Just reply with the text you want sent.**
+
+## Available Tools
+
+| Tool | When to use |
+|------|------------|
+| `yb_query_group_info` | Query group name, owner, member count |
+| `yb_query_group_members` | Find a user, list bots, list all members, or get nickname for @mention |
+| `yb_send_dm` | Send a private/direct message (DM / 私信) to a user, with optional media files |
+
+## @Mention Workflow
+
+When you need to @mention / 艾特 someone:
+
+1. Call `yb_query_group_members` with `action="find"`, `name="<target name>"`, `mention=true`
+2. Get the exact nickname from the response
+3. Include `@nickname` in your reply text — the gateway handles the rest
+
+Example: user says "帮我艾特元宝"
+
+Step 1 — tool call:
+```json
+{ "group_code": "328306697", "action": "find", "name": "元宝", "mention": true }
+```
+
+Step 2 — your reply (this gets sent to the group with a working @mention):
+```
+@元宝 你好，有人找你！
+```
+
+**That's it.** No extra explanation needed. Keep it short and natural.
+
+**Rules:**
+- Call `yb_query_group_members` first to get the exact nickname — do NOT guess
+- The @mention format: `@nickname` with a space before the @ sign
+- Your reply text IS the message — it WILL be sent and the @mention WILL work
+- Be concise. Do NOT explain how @mention works to the user.
+
+## Send DM (Private Message) Workflow
+
+When someone asks to send a private message / 私信 / DM to a user:
+
+1. Call `yb_send_dm` with `group_code`, `name` (target user's name), and `message`
+2. The tool automatically finds the user and sends the DM
+3. Report the result to the user
+
+Example: user says "给 @用户aea3 私信发一个 hello"
+
+```json
+yb_send_dm({ "group_code": "535168412", "name": "用户aea3", "message": "hello" })
+```
+
+Example with media: user says "给 @用户aea3 私信发一张图片"
+
+```json
+yb_send_dm({
+  "group_code": "535168412",
+  "name": "用户aea3",
+  "message": "Here is the image",
+  "media_files": [{"path": "/tmp/photo.jpg"}]
+})
+```
+
+**Rules:**
+- Extract `group_code` from the current chat_id (e.g. `group:535168412` → `535168412`)
+- If you already know the user_id, pass it directly via the `user_id` parameter to skip lookup
+- If multiple users match the name, the tool returns candidates — ask the user to clarify
+- Do NOT use `send_message` tool for Yuanbao DMs — use `yb_send_dm` instead
+- Supports media: images (.jpg/.png/.gif/.webp/.bmp) sent as image messages, other files as documents
+
+## Query Group Info
+
+```json
+yb_query_group_info({ "group_code": "328306697" })
+```
+
+## Query Members
+
+| Action | Description |
+|--------|-------------|
+| `find` | Search by name (partial match, case-insensitive) |
+| `list_bots` | List bots and Yuanbao AI assistants |
+| `list_all` | List all members |
+
+## Notes
+
+- `group_code` comes from chat_id: `group:328306697` → `328306697`
+- Groups are called "派 (Pai)" in the Yuanbao app
+- Member roles: `user`, `yuanbao_ai`, `bot`

tests/agent/test_auxiliary_named_custom_providers.py

@@ -386,7 +386,7 @@ class TestProvidersDictApiModeAnthropicMessages:
                 },
             },
             "auxiliary": {
-                "flush_memories": {
+                "compression": {
                     "provider": "myrelay",
                     "model": "claude-sonnet-4.6",
                 },
@@ -399,11 +399,11 @@ class TestProvidersDictApiModeAnthropicMessages:
             AnthropicAuxiliaryClient,
             AsyncAnthropicAuxiliaryClient,
         )
-        async_client, async_model = get_async_text_auxiliary_client("flush_memories")
+        async_client, async_model = get_async_text_auxiliary_client("compression")
         assert isinstance(async_client, AsyncAnthropicAuxiliaryClient)
         assert async_model == "claude-sonnet-4.6"
 
-        sync_client, sync_model = get_text_auxiliary_client("flush_memories")
+        sync_client, sync_model = get_text_auxiliary_client("compression")
         assert isinstance(sync_client, AnthropicAuxiliaryClient)
         assert sync_model == "claude-sonnet-4.6"
 

tests/agent/test_model_metadata.py

@@ -192,6 +192,43 @@ class TestDefaultContextLengths:
                     f"{model_id}: expected {expected_ctx}, got {actual}"
                 )
 
+    def test_deepseek_v4_models_1m_context(self):
+        from agent.model_metadata import get_model_context_length
+        from unittest.mock import patch as mock_patch
+
+        expected_keys = {
+            "deepseek-v4-pro": 1_000_000,
+            "deepseek-v4-flash": 1_000_000,
+            "deepseek-chat": 1_000_000,
+            "deepseek-reasoner": 1_000_000,
+        }
+        for key, value in expected_keys.items():
+            assert key in DEFAULT_CONTEXT_LENGTHS, f"{key} missing"
+            assert DEFAULT_CONTEXT_LENGTHS[key] == value, (
+                f"{key} should be {value}, got {DEFAULT_CONTEXT_LENGTHS[key]}"
+            )
+
+        # Longest-first substring matching must resolve both the bare V4
+        # ids (native DeepSeek) and the vendor-prefixed forms (OpenRouter
+        # / Nous Portal) to 1M without probing down to the legacy 128K
+        # ``deepseek`` substring fallback.
+        with mock_patch("agent.model_metadata.fetch_model_metadata", return_value={}), \
+             mock_patch("agent.model_metadata.fetch_endpoint_model_metadata", return_value={}), \
+             mock_patch("agent.model_metadata.get_cached_context_length", return_value=None):
+            cases = [
+                ("deepseek-v4-pro", 1_000_000),
+                ("deepseek-v4-flash", 1_000_000),
+                ("deepseek/deepseek-v4-pro", 1_000_000),
+                ("deepseek/deepseek-v4-flash", 1_000_000),
+                ("deepseek-chat", 1_000_000),
+                ("deepseek-reasoner", 1_000_000),
+            ]
+            for model_id, expected_ctx in cases:
+                actual = get_model_context_length(model_id)
+                assert actual == expected_ctx, (
+                    f"{model_id}: expected {expected_ctx}, got {actual}"
+                )
+
     def test_all_values_positive(self):
         for key, value in DEFAULT_CONTEXT_LENGTHS.items():
             assert value > 0, f"{key} has non-positive context length"
@@ -303,7 +340,9 @@ class TestCodexOAuthContextLength:
         from agent.model_metadata import get_model_context_length
 
         # OpenRouter — should hit its own catalog path first; when mocked
-        # empty, falls through to hardcoded DEFAULT_CONTEXT_LENGTHS (400k).
+        # empty, falls through to hardcoded DEFAULT_CONTEXT_LENGTHS (1.05M,
+        # matching the real direct-API value — Codex OAuth's 272k cap is
+        # provider-specific and must not leak here).
         with patch("agent.model_metadata.fetch_model_metadata", return_value={}), \
              patch("agent.model_metadata.fetch_endpoint_model_metadata", return_value={}), \
              patch("agent.model_metadata.get_cached_context_length", return_value=None), \
@@ -314,7 +353,7 @@ class TestCodexOAuthContextLength:
                 api_key="",
                 provider="openrouter",
             )
-        assert ctx == 400_000, (
+        assert ctx == 1_050_000, (
             f"Non-Codex gpt-5.5 resolved to {ctx}; Codex 272k override "
             "leaked outside openai-codex provider"
         )
@@ -459,9 +498,10 @@ class TestGetModelContextLength:
 
     @patch("agent.model_metadata.fetch_model_metadata")
     def test_api_missing_context_length_key(self, mock_fetch):
-        """Model in API but without context_length → defaults to 128000."""
+        """Model in API but without context_length → defaults to the top
+        probe tier (currently 256K)."""
         mock_fetch.return_value = {"test/model": {"name": "Test"}}
-        assert get_model_context_length("test/model") == 128000
+        assert get_model_context_length("test/model") == CONTEXT_PROBE_TIERS[0]
 
     @patch("agent.model_metadata.fetch_model_metadata")
     def test_cache_takes_priority_over_api(self, mock_fetch, tmp_path):
@@ -814,14 +854,17 @@ class TestContextProbeTiers:
         for i in range(len(CONTEXT_PROBE_TIERS) - 1):
             assert CONTEXT_PROBE_TIERS[i] > CONTEXT_PROBE_TIERS[i + 1]
 
-    def test_first_tier_is_128k(self):
-        assert CONTEXT_PROBE_TIERS[0] == 128_000
+    def test_first_tier_is_256k(self):
+        assert CONTEXT_PROBE_TIERS[0] == 256_000
 
     def test_last_tier_is_8k(self):
         assert CONTEXT_PROBE_TIERS[-1] == 8_000
 
 
 class TestGetNextProbeTier:
+    def test_from_256k(self):
+        assert get_next_probe_tier(256_000) == 128_000
+
     def test_from_128k(self):
         assert get_next_probe_tier(128_000) == 64_000
 
@@ -841,8 +884,8 @@ class TestGetNextProbeTier:
         assert get_next_probe_tier(100_000) == 64_000
 
     def test_above_max_tier(self):
-        """Value above 128K should return 128K."""
-        assert get_next_probe_tier(500_000) == 128_000
+        """Value above 256K should return 256K."""
+        assert get_next_probe_tier(500_000) == 256_000
 
     def test_zero_returns_none(self):
         assert get_next_probe_tier(0) is None

tests/agent/test_nous_rate_guard.py

@@ -251,3 +251,141 @@ class TestAuxiliaryClientIntegration:
         monkeypatch.setattr(aux, "_read_nous_auth", lambda: None)
         result = aux._try_nous()
         assert result == (None, None)
+
+
+class TestIsGenuineNousRateLimit:
+    """Tell a real account-level 429 apart from an upstream-capacity 429.
+
+    Nous Portal multiplexes upstreams (DeepSeek, Kimi, MiMo, Hermes).
+    A 429 from an upstream out of capacity should NOT trip the
+    cross-session breaker; a real user-quota 429 should.
+    """
+
+    def test_exhausted_hourly_bucket_in_429_headers_is_genuine(self):
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        headers = {
+            "x-ratelimit-limit-requests-1h": "800",
+            "x-ratelimit-remaining-requests-1h": "0",
+            "x-ratelimit-reset-requests-1h": "3100",
+            "x-ratelimit-limit-requests": "200",
+            "x-ratelimit-remaining-requests": "198",
+            "x-ratelimit-reset-requests": "40",
+        }
+        assert is_genuine_nous_rate_limit(headers=headers) is True
+
+    def test_exhausted_tokens_bucket_is_genuine(self):
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        headers = {
+            "x-ratelimit-limit-tokens": "800000",
+            "x-ratelimit-remaining-tokens": "0",
+            "x-ratelimit-reset-tokens": "45",  # < 60s threshold -> not genuine
+            "x-ratelimit-limit-tokens-1h": "8000000",
+            "x-ratelimit-remaining-tokens-1h": "0",
+            "x-ratelimit-reset-tokens-1h": "1800",  # >= 60s threshold -> genuine
+        }
+        assert is_genuine_nous_rate_limit(headers=headers) is True
+
+    def test_healthy_headers_on_429_are_upstream_capacity(self):
+        # Classic upstream-capacity symptom: Nous edge reports plenty of
+        # headroom on every bucket, but returns 429 anyway because
+        # upstream (DeepSeek / Kimi / ...) is out of capacity.
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        headers = {
+            "x-ratelimit-limit-requests": "200",
+            "x-ratelimit-remaining-requests": "198",
+            "x-ratelimit-reset-requests": "40",
+            "x-ratelimit-limit-requests-1h": "800",
+            "x-ratelimit-remaining-requests-1h": "750",
+            "x-ratelimit-reset-requests-1h": "3100",
+            "x-ratelimit-limit-tokens": "800000",
+            "x-ratelimit-remaining-tokens": "790000",
+            "x-ratelimit-reset-tokens": "40",
+            "x-ratelimit-limit-tokens-1h": "8000000",
+            "x-ratelimit-remaining-tokens-1h": "7800000",
+            "x-ratelimit-reset-tokens-1h": "3100",
+        }
+        assert is_genuine_nous_rate_limit(headers=headers) is False
+
+    def test_bare_429_with_no_headers_is_upstream(self):
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        assert is_genuine_nous_rate_limit(headers=None) is False
+        assert is_genuine_nous_rate_limit(headers={}) is False
+        assert is_genuine_nous_rate_limit(
+            headers={"content-type": "application/json"}
+        ) is False
+
+    def test_exhausted_bucket_with_short_reset_is_not_genuine(self):
+        # remaining == 0 but reset in < 60s: almost certainly a
+        # secondary per-minute throttle that will clear immediately --
+        # not worth tripping the cross-session breaker.
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        headers = {
+            "x-ratelimit-limit-requests": "200",
+            "x-ratelimit-remaining-requests": "0",
+            "x-ratelimit-reset-requests": "30",
+        }
+        assert is_genuine_nous_rate_limit(headers=headers) is False
+
+    def test_last_known_state_with_exhausted_bucket_triggers_genuine(self):
+        # Headers on the 429 lack rate-limit info, but the previous
+        # successful response already showed the hourly bucket
+        # exhausted -- the 429 is almost certainly that limit
+        # continuing.
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+        from agent.rate_limit_tracker import parse_rate_limit_headers
+
+        prior_headers = {
+            "x-ratelimit-limit-requests-1h": "800",
+            "x-ratelimit-remaining-requests-1h": "0",
+            "x-ratelimit-reset-requests-1h": "2000",
+            "x-ratelimit-limit-requests": "200",
+            "x-ratelimit-remaining-requests": "100",
+            "x-ratelimit-reset-requests": "30",
+            "x-ratelimit-limit-tokens": "800000",
+            "x-ratelimit-remaining-tokens": "700000",
+            "x-ratelimit-reset-tokens": "30",
+            "x-ratelimit-limit-tokens-1h": "8000000",
+            "x-ratelimit-remaining-tokens-1h": "7000000",
+            "x-ratelimit-reset-tokens-1h": "2000",
+        }
+        last_state = parse_rate_limit_headers(prior_headers, provider="nous")
+        assert is_genuine_nous_rate_limit(
+            headers=None, last_known_state=last_state
+        ) is True
+
+    def test_last_known_state_all_healthy_stays_upstream(self):
+        # Prior state was healthy; bare 429 arrives; should be treated
+        # as upstream capacity.
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+        from agent.rate_limit_tracker import parse_rate_limit_headers
+
+        prior_headers = {
+            "x-ratelimit-limit-requests-1h": "800",
+            "x-ratelimit-remaining-requests-1h": "750",
+            "x-ratelimit-reset-requests-1h": "2000",
+            "x-ratelimit-limit-requests": "200",
+            "x-ratelimit-remaining-requests": "180",
+            "x-ratelimit-reset-requests": "30",
+            "x-ratelimit-limit-tokens": "800000",
+            "x-ratelimit-remaining-tokens": "790000",
+            "x-ratelimit-reset-tokens": "30",
+            "x-ratelimit-limit-tokens-1h": "8000000",
+            "x-ratelimit-remaining-tokens-1h": "7900000",
+            "x-ratelimit-reset-tokens-1h": "2000",
+        }
+        last_state = parse_rate_limit_headers(prior_headers, provider="nous")
+        assert is_genuine_nous_rate_limit(
+            headers=None, last_known_state=last_state
+        ) is False
+
+    def test_none_last_state_and_no_headers_is_upstream(self):
+        from agent.nous_rate_guard import is_genuine_nous_rate_limit
+
+        assert is_genuine_nous_rate_limit(
+            headers=None, last_known_state=None
+        ) is False

tests/agent/test_onboarding.py

@@ -0,0 +1,228 @@
+"""Tests for agent/onboarding.py — contextual first-touch hint helpers."""
+
+from __future__ import annotations
+
+import yaml
+import pytest
+
+from agent.onboarding import (
+    BUSY_INPUT_FLAG,
+    OPENCLAW_RESIDUE_FLAG,
+    TOOL_PROGRESS_FLAG,
+    busy_input_hint_cli,
+    busy_input_hint_gateway,
+    detect_openclaw_residue,
+    is_seen,
+    mark_seen,
+    openclaw_residue_hint_cli,
+    tool_progress_hint_cli,
+    tool_progress_hint_gateway,
+)
+
+
+class TestIsSeen:
+    def test_empty_config_unseen(self):
+        assert is_seen({}, BUSY_INPUT_FLAG) is False
+
+    def test_missing_onboarding_unseen(self):
+        assert is_seen({"display": {}}, BUSY_INPUT_FLAG) is False
+
+    def test_onboarding_not_dict_unseen(self):
+        assert is_seen({"onboarding": "nope"}, BUSY_INPUT_FLAG) is False
+
+    def test_seen_dict_missing_flag(self):
+        assert is_seen({"onboarding": {"seen": {}}}, BUSY_INPUT_FLAG) is False
+
+    def test_seen_flag_true(self):
+        cfg = {"onboarding": {"seen": {BUSY_INPUT_FLAG: True}}}
+        assert is_seen(cfg, BUSY_INPUT_FLAG) is True
+
+    def test_seen_flag_falsy(self):
+        cfg = {"onboarding": {"seen": {BUSY_INPUT_FLAG: False}}}
+        assert is_seen(cfg, BUSY_INPUT_FLAG) is False
+
+    def test_other_flags_isolated(self):
+        cfg = {"onboarding": {"seen": {BUSY_INPUT_FLAG: True}}}
+        assert is_seen(cfg, TOOL_PROGRESS_FLAG) is False
+
+
+class TestMarkSeen:
+    def test_creates_missing_file_and_sets_flag(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+
+        loaded = yaml.safe_load(cfg_path.read_text())
+        assert loaded["onboarding"]["seen"][BUSY_INPUT_FLAG] is True
+
+    def test_preserves_other_config(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        cfg_path.write_text(yaml.safe_dump({
+            "model": {"default": "claude-sonnet-4.6"},
+            "display": {"skin": "default"},
+        }))
+
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+
+        assert loaded["model"]["default"] == "claude-sonnet-4.6"
+        assert loaded["display"]["skin"] == "default"
+        assert loaded["onboarding"]["seen"][BUSY_INPUT_FLAG] is True
+
+    def test_preserves_other_seen_flags(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        cfg_path.write_text(yaml.safe_dump({
+            "onboarding": {"seen": {TOOL_PROGRESS_FLAG: True}},
+        }))
+
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+
+        assert loaded["onboarding"]["seen"][TOOL_PROGRESS_FLAG] is True
+        assert loaded["onboarding"]["seen"][BUSY_INPUT_FLAG] is True
+
+    def test_idempotent(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        mark_seen(cfg_path, BUSY_INPUT_FLAG)
+        first = cfg_path.read_text()
+
+        # Second call must be a no-op on-disk content (file may be touched,
+        # but the YAML contents should be identical).
+        mark_seen(cfg_path, BUSY_INPUT_FLAG)
+        second = cfg_path.read_text()
+
+        assert yaml.safe_load(first) == yaml.safe_load(second)
+
+    def test_handles_non_dict_onboarding(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        cfg_path.write_text(yaml.safe_dump({"onboarding": "corrupted"}))
+
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+        assert loaded["onboarding"]["seen"][BUSY_INPUT_FLAG] is True
+
+    def test_handles_non_dict_seen(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        cfg_path.write_text(yaml.safe_dump({"onboarding": {"seen": "corrupted"}}))
+
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+        assert loaded["onboarding"]["seen"][BUSY_INPUT_FLAG] is True
+
+
+class TestHintMessages:
+    def test_busy_input_hint_gateway_interrupt(self):
+        msg = busy_input_hint_gateway("interrupt")
+        assert "/busy queue" in msg
+        assert "interrupted" in msg.lower()
+
+    def test_busy_input_hint_gateway_queue(self):
+        msg = busy_input_hint_gateway("queue")
+        assert "/busy interrupt" in msg
+        assert "queued" in msg.lower()
+
+    def test_busy_input_hint_gateway_steer(self):
+        msg = busy_input_hint_gateway("steer")
+        assert "/busy interrupt" in msg
+        assert "/busy queue" in msg
+        assert "steer" in msg.lower()
+
+    def test_busy_input_hint_cli_interrupt(self):
+        msg = busy_input_hint_cli("interrupt")
+        assert "/busy queue" in msg
+
+    def test_busy_input_hint_cli_queue(self):
+        msg = busy_input_hint_cli("queue")
+        assert "/busy interrupt" in msg
+
+    def test_busy_input_hint_cli_steer(self):
+        msg = busy_input_hint_cli("steer")
+        assert "/busy interrupt" in msg
+        assert "/busy queue" in msg
+        assert "steer" in msg.lower()
+
+    def test_tool_progress_hints_mention_verbose(self):
+        assert "/verbose" in tool_progress_hint_gateway()
+        assert "/verbose" in tool_progress_hint_cli()
+
+    def test_hints_are_not_empty(self):
+        for hint in (
+            busy_input_hint_gateway("queue"),
+            busy_input_hint_gateway("interrupt"),
+            busy_input_hint_gateway("steer"),
+            busy_input_hint_cli("queue"),
+            busy_input_hint_cli("interrupt"),
+            busy_input_hint_cli("steer"),
+            tool_progress_hint_gateway(),
+            tool_progress_hint_cli(),
+        ):
+            assert hint.strip()
+
+
+class TestRoundTrip:
+    """After mark_seen, is_seen on the re-loaded config must return True."""
+
+    def test_mark_then_is_seen(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+
+        assert mark_seen(cfg_path, BUSY_INPUT_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+
+        assert is_seen(loaded, BUSY_INPUT_FLAG) is True
+        assert is_seen(loaded, TOOL_PROGRESS_FLAG) is False
+
+    def test_mark_both_flags_independently(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+
+        mark_seen(cfg_path, BUSY_INPUT_FLAG)
+        mark_seen(cfg_path, TOOL_PROGRESS_FLAG)
+        loaded = yaml.safe_load(cfg_path.read_text())
+
+        assert is_seen(loaded, BUSY_INPUT_FLAG) is True
+        assert is_seen(loaded, TOOL_PROGRESS_FLAG) is True
+
+
+# ---------------------------------------------------------------------------
+# OpenClaw residue banner
+# ---------------------------------------------------------------------------
+
+
+class TestDetectOpenclawResidue:
+    def test_returns_true_when_openclaw_dir_present(self, tmp_path):
+        (tmp_path / ".openclaw").mkdir()
+        assert detect_openclaw_residue(home=tmp_path) is True
+
+    def test_returns_false_when_absent(self, tmp_path):
+        assert detect_openclaw_residue(home=tmp_path) is False
+
+    def test_returns_false_when_path_is_a_file(self, tmp_path):
+        # A stray file named ``.openclaw`` is NOT a workspace — skip the banner.
+        (tmp_path / ".openclaw").write_text("oops")
+        assert detect_openclaw_residue(home=tmp_path) is False
+
+    def test_default_home_does_not_crash(self):
+        # Smoke: real $HOME lookup must not raise regardless of state.
+        assert isinstance(detect_openclaw_residue(), bool)
+
+
+class TestOpenclawResidueHint:
+    def test_hint_mentions_cleanup_command(self):
+        msg = openclaw_residue_hint_cli()
+        assert "hermes claw cleanup" in msg
+        assert "~/.openclaw" in msg
+
+    def test_hint_not_empty(self):
+        assert openclaw_residue_hint_cli().strip()
+
+
+class TestOpenclawResidueSeenFlag:
+    def test_flag_independent_of_other_flags(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        mark_seen(cfg_path, BUSY_INPUT_FLAG)
+        loaded = yaml.safe_load(cfg_path.read_text())
+        assert is_seen(loaded, OPENCLAW_RESIDUE_FLAG) is False
+
+    def test_flag_round_trips(self, tmp_path):
+        cfg_path = tmp_path / "config.yaml"
+        assert mark_seen(cfg_path, OPENCLAW_RESIDUE_FLAG) is True
+        loaded = yaml.safe_load(cfg_path.read_text())
+        assert is_seen(loaded, OPENCLAW_RESIDUE_FLAG) is True

tests/agent/test_shell_hooks_consent.py

@@ -240,3 +240,74 @@ class TestAllowlistOps:
             and e.get("command") == str(script)
         ]
         assert len(matching) == 1
+
+
+# ── hooks_auto_accept config parsing ──────────────────────────────────────
+
+
+class TestHooksAutoAcceptParsing:
+    """Regression guard: YAML-string values must not silently auto-accept.
+
+    ``bool("false")`` is ``True`` in Python, so the old ``return bool(cfg_val)``
+    path treated ``hooks_auto_accept: "false"`` (quoted YAML string) as a
+    truthy opt-in, silently bypassing user consent for every shell hook.
+    """
+
+    def test_bool_true_accepts(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": True}, accept_hooks_arg=False,
+        ) is True
+
+    def test_bool_false_rejects(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": False}, accept_hooks_arg=False,
+        ) is False
+
+    def test_string_false_rejects(self):
+        # The bug: bool("false") is True. Must be parsed, not coerced.
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": "false"}, accept_hooks_arg=False,
+        ) is False
+
+    def test_string_no_rejects(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": "no"}, accept_hooks_arg=False,
+        ) is False
+
+    def test_string_true_accepts(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": "true"}, accept_hooks_arg=False,
+        ) is True
+
+    def test_string_true_case_insensitive(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": "  TRUE  "}, accept_hooks_arg=False,
+        ) is True
+
+    def test_string_yes_on_one_accept(self):
+        for val in ("yes", "on", "1"):
+            assert shell_hooks._resolve_effective_accept(
+                {"hooks_auto_accept": val}, accept_hooks_arg=False,
+            ) is True, val
+
+    def test_missing_key_rejects(self):
+        assert shell_hooks._resolve_effective_accept(
+            {}, accept_hooks_arg=False,
+        ) is False
+
+    def test_none_rejects(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": None}, accept_hooks_arg=False,
+        ) is False
+
+    def test_integer_ignored(self):
+        # Only bool and str are honored; anything else (including 1) is False.
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": 1}, accept_hooks_arg=False,
+        ) is False
+
+    def test_cli_arg_overrides_config(self):
+        assert shell_hooks._resolve_effective_accept(
+            {"hooks_auto_accept": "false"}, accept_hooks_arg=True,
+        ) is True
+

tests/agent/test_unsupported_parameter_retry.py

@@ -0,0 +1,201 @@
+"""Regression tests for the generic unsupported-parameter detector in
+``agent.auxiliary_client``.
+
+The original temperature-specific detector (PR #15621) was generalized so the
+same reactive-retry strategy covers any provider that rejects an arbitrary
+request parameter — ``max_tokens``, ``seed``, ``top_p``, future quirks — not
+just ``temperature``. Credit @nicholasrae (PR #15416) for the generalization
+pattern.
+
+These tests lock in:
+  * ``_is_unsupported_parameter_error(exc, param)`` across common phrasings
+  * the back-compat wrapper ``_is_unsupported_temperature_error`` still works
+  * the max_tokens retry branch no longer pops a key that was never set
+    (``max_tokens is None`` gate)
+  * the max_tokens retry branch matches via the generic helper on top of the
+    legacy ``"max_tokens"`` / ``"unsupported_parameter"`` substring checks
+"""
+
+from unittest.mock import patch, MagicMock, AsyncMock
+
+import pytest
+
+from agent.auxiliary_client import (
+    call_llm,
+    async_call_llm,
+    _is_unsupported_parameter_error,
+    _is_unsupported_temperature_error,
+)
+
+
+class TestIsUnsupportedParameterError:
+    """The generic detector must match real provider phrasings for any param."""
+
+    @pytest.mark.parametrize("param,message", [
+        # temperature phrasings (regression coverage via the generic API)
+        ("temperature", "HTTP 400: Unsupported parameter: temperature"),
+        ("temperature", "Error code: 400 - {'error': {'code': 'unsupported_parameter', 'param': 'temperature'}}"),
+        ("temperature", "this model does not support temperature"),
+        # max_tokens phrasings
+        ("max_tokens", "HTTP 400: Unsupported parameter: max_tokens"),
+        ("max_tokens", "Unknown parameter: max_tokens — use max_completion_tokens"),
+        ("max_tokens", "Invalid parameter: max_tokens is not supported"),
+        # arbitrary future params
+        ("seed", "HTTP 400: unrecognized parameter: seed"),
+        ("top_p", "Error: top_p is not supported for this model"),
+    ])
+    def test_matches_real_provider_messages(self, param, message):
+        assert _is_unsupported_parameter_error(RuntimeError(message), param) is True
+
+    @pytest.mark.parametrize("param,message", [
+        # Param not mentioned at all
+        ("temperature", "HTTP 400: max_tokens is too large"),
+        # Param mentioned but not flagged as unsupported
+        ("temperature", "temperature must be between 0 and 2"),
+        # Totally unrelated 400
+        ("max_tokens", "Rate limit exceeded"),
+        # Connection-level errors
+        ("temperature", "Connection reset by peer"),
+    ])
+    def test_does_not_match_unrelated_errors(self, param, message):
+        assert _is_unsupported_parameter_error(RuntimeError(message), param) is False
+
+    def test_empty_param_returns_false(self):
+        assert _is_unsupported_parameter_error(
+            RuntimeError("HTTP 400: Unsupported parameter: temperature"), ""
+        ) is False
+
+    def test_temperature_wrapper_delegates_to_generic(self):
+        """Back-compat: ``_is_unsupported_temperature_error`` still routes through."""
+        msg = "HTTP 400: Unsupported parameter: temperature"
+        assert _is_unsupported_temperature_error(RuntimeError(msg)) is True
+        # And the unrelated-case still holds
+        assert _is_unsupported_temperature_error(
+            RuntimeError("max_tokens is too large")) is False
+
+
+def _dummy_response():
+    """Sentinel — real code calls ``_validate_llm_response`` which we patch out."""
+    return {"ok": True}
+
+
+class TestMaxTokensRetryHardening:
+    """The max_tokens retry branch now (a) gates on ``max_tokens is not None``
+    and (b) also matches the generic phrasings via the helper.
+    """
+
+    def test_sync_max_tokens_retry_skipped_when_max_tokens_is_none(self):
+        """No max_tokens kwarg → must not pop/retry even if the error mentions it.
+
+        Before the hardening, ``kwargs.pop("max_tokens", None)`` was safe but
+        ``kwargs["max_completion_tokens"] = max_tokens`` would set a None
+        value and hit the provider again. The gate skips the whole branch.
+        """
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        err = RuntimeError("HTTP 400: Unsupported parameter: max_tokens")
+        client.chat.completions.create.side_effect = err
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            with pytest.raises(RuntimeError):
+                call_llm(
+                    task="session_search",
+                    messages=[{"role": "user", "content": "hi"}],
+                    temperature=0.3,
+                    # max_tokens omitted on purpose
+                )
+
+        # Only the initial attempt — no retry because the gate blocked it
+        assert client.chat.completions.create.call_count == 1
+
+    def test_sync_max_tokens_retry_matches_generic_phrasing(self):
+        """A 400 saying "Unknown parameter: max_tokens" (not the legacy
+        substring ``"max_tokens"`` bare + no ``unsupported_parameter`` token)
+        now triggers the retry via the generic helper.
+        """
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        err = RuntimeError("Unknown parameter: max_tokens")
+        response = _dummy_response()
+        client.chat.completions.create.side_effect = [err, response]
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            result = call_llm(
+                task="session_search",
+                messages=[{"role": "user", "content": "hi"}],
+                temperature=0.3,
+                max_tokens=512,
+            )
+
+        assert result is response
+        assert client.chat.completions.create.call_count == 2
+        second_call = client.chat.completions.create.call_args_list[1]
+        assert "max_tokens" not in second_call.kwargs
+        assert second_call.kwargs["max_completion_tokens"] == 512
+
+    @pytest.mark.asyncio
+    async def test_async_max_tokens_retry_skipped_when_max_tokens_is_none(self):
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        err = RuntimeError("HTTP 400: Unsupported parameter: max_tokens")
+        client.chat.completions.create = AsyncMock(side_effect=err)
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            with pytest.raises(RuntimeError):
+                await async_call_llm(
+                    task="session_search",
+                    messages=[{"role": "user", "content": "hi"}],
+                    temperature=0.3,
+                )
+
+        assert client.chat.completions.create.call_count == 1
+
+    @pytest.mark.asyncio
+    async def test_async_max_tokens_retry_matches_generic_phrasing(self):
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        err = RuntimeError("Unknown parameter: max_tokens")
+        response = _dummy_response()
+        client.chat.completions.create = AsyncMock(side_effect=[err, response])
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            result = await async_call_llm(
+                task="session_search",
+                messages=[{"role": "user", "content": "hi"}],
+                temperature=0.3,
+                max_tokens=512,
+            )
+
+        assert result is response
+        assert client.chat.completions.create.await_count == 2
+        second_call = client.chat.completions.create.call_args_list[1]
+        assert "max_tokens" not in second_call.kwargs
+        assert second_call.kwargs["max_completion_tokens"] == 512

tests/agent/test_unsupported_temperature_retry.py

@@ -0,0 +1,237 @@
+"""Regression tests for the universal "unsupported temperature" retry in
+``agent.auxiliary_client``.
+
+Auxiliary callers (context compression, session search,
+web extract summarisation, etc.) hardcode ``temperature=0.3`` for historical
+reasons. Several provider/model combinations reject ``temperature`` with a
+400:
+
+  * OpenAI Responses (gpt-5/o-series reasoning models)
+  * Copilot Responses (reasoning models)
+  * OpenRouter reasoning models (gpt-5.5, some anthropic via OAI-compat)
+  * Anthropic Opus 4.7+ via OpenAI-compat endpoints
+  * Kimi/Moonshot (server-managed)
+
+``_fixed_temperature_for_model`` catches Kimi up front, and
+``build_chat_completion_kwargs`` drops temperature for Anthropic Opus 4.7+,
+but the same backend can accept ``temperature`` for some models and reject
+it for others (for example gpt-5.4 accepts but gpt-5.5 rejects on the same
+endpoint). An allow/deny-list is not maintainable across providers.
+
+The universal fix is reactive: when a call returns an
+``Unsupported parameter: temperature`` 400, retry once without temperature.
+These tests lock in that behaviour for both sync and async paths.
+"""
+
+from unittest.mock import patch, MagicMock, AsyncMock
+
+import pytest
+
+from agent.auxiliary_client import (
+    call_llm,
+    async_call_llm,
+    _is_unsupported_temperature_error,
+)
+
+
+class TestIsUnsupportedTemperatureError:
+    """The detector must match the phrasings providers actually return."""
+
+    @pytest.mark.parametrize("message", [
+        # OpenAI / Codex Responses
+        "HTTP 400: Unsupported parameter: temperature",
+        "Error code: 400 - {'error': {'message': \"Unsupported parameter: 'temperature'\"}}",
+        # Copilot / OpenAI error-code form
+        "Error code: 400 - {'error': {'code': 'unsupported_parameter', 'param': 'temperature'}}",
+        # OpenRouter-style
+        "Provider returned error: temperature is not supported for this model",
+        "this model does not support temperature",
+        # Anthropic-style via OAI-compat
+        "temperature: unknown parameter",
+        # Some gateways
+        "unrecognized request argument supplied: temperature",
+    ])
+    def test_matches_real_provider_messages(self, message):
+        assert _is_unsupported_temperature_error(RuntimeError(message)) is True
+
+    @pytest.mark.parametrize("message", [
+        # Unrelated 400s must NOT trigger a silent-retry
+        "HTTP 400: Invalid value: 'tool'. Supported values are: 'assistant'...",
+        "max_tokens is too large for this model",
+        "Rate limit exceeded",
+        "Connection reset by peer",
+        # Temperature value error is a different class of problem
+        "temperature must be between 0 and 2",
+    ])
+    def test_does_not_match_unrelated_errors(self, message):
+        assert _is_unsupported_temperature_error(RuntimeError(message)) is False
+
+
+def _dummy_response():
+    # The real code calls _validate_llm_response which inspects
+    # response.choices[0].message.  The tests here patch that out, so
+    # any sentinel object is fine.
+    return {"ok": True}
+
+
+class TestCallLlmUnsupportedTemperatureRetry:
+    """``call_llm`` retries once without temperature and returns on success."""
+
+    def _setup(self, first_exc):
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        client.chat.completions.create.side_effect = [first_exc, _dummy_response()]
+        return client
+
+    @pytest.mark.parametrize("error_message", [
+        "HTTP 400: Unsupported parameter: temperature",
+        "Error code: 400 - {'error': {'code': 'unsupported_parameter', 'param': 'temperature'}}",
+        "Provider error: this model does not support temperature",
+    ])
+    def test_retries_once_without_temperature(self, error_message):
+        client = self._setup(RuntimeError(error_message))
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            result = call_llm(
+                task="compression",
+                messages=[{"role": "user", "content": "remember this"}],
+                temperature=0.3,
+                max_tokens=500,
+            )
+
+        assert result == {"ok": True}
+        assert client.chat.completions.create.call_count == 2
+        first_kwargs = client.chat.completions.create.call_args_list[0].kwargs
+        retry_kwargs = client.chat.completions.create.call_args_list[1].kwargs
+        assert first_kwargs["temperature"] == 0.3
+        assert "temperature" not in retry_kwargs
+        # other kwargs preserved
+        assert retry_kwargs["max_tokens"] == 500
+
+    def test_non_temperature_400_does_not_retry_as_temperature(self):
+        """Unrelated 400s (e.g. bad tool role) must not silently drop temp."""
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        non_temp_err = RuntimeError(
+            "HTTP 400: Invalid value: 'tool'. Supported values are: 'assistant'..."
+        )
+        client.chat.completions.create.side_effect = non_temp_err
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+            patch("agent.auxiliary_client._try_payment_fallback",
+                  return_value=None),
+        ):
+            with pytest.raises(RuntimeError, match="Invalid value"):
+                call_llm(
+                    task="compression",
+                    messages=[{"role": "user", "content": "x"}],
+                    temperature=0.3,
+                    max_tokens=500,
+                )
+        # Should NOT have retried (non-temperature 400 doesn't match)
+        assert client.chat.completions.create.call_count == 1
+
+    def test_no_retry_when_temperature_not_in_kwargs(self):
+        """If caller didn't send temperature, don't invent a temperature-retry."""
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        # Provider complains about temperature even though we didn't send it.
+        # (Pathological but possible with misleading error text.)  The guard
+        # ``"temperature" in kwargs`` must prevent an unnecessary retry.
+        err = RuntimeError("HTTP 400: Unsupported parameter: temperature")
+        client.chat.completions.create.side_effect = err
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+            patch("agent.auxiliary_client._try_payment_fallback",
+                  return_value=None),
+        ):
+            with pytest.raises(RuntimeError):
+                call_llm(
+                    task="compression",
+                    messages=[{"role": "user", "content": "x"}],
+                    temperature=None,  # explicit: no temperature sent
+                    max_tokens=500,
+                )
+        assert client.chat.completions.create.call_count == 1
+
+
+class TestAsyncCallLlmUnsupportedTemperatureRetry:
+    """``async_call_llm`` mirror of the sync retry semantics."""
+
+    @pytest.mark.asyncio
+    async def test_async_retries_once_without_temperature(self):
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        client.chat.completions.create = AsyncMock(side_effect=[
+            RuntimeError("HTTP 400: Unsupported parameter: temperature"),
+            _dummy_response(),
+        ])
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+        ):
+            result = await async_call_llm(
+                task="session_search",
+                messages=[{"role": "user", "content": "query"}],
+                temperature=0.3,
+                max_tokens=500,
+            )
+
+        assert result == {"ok": True}
+        assert client.chat.completions.create.await_count == 2
+        first_kwargs = client.chat.completions.create.call_args_list[0].kwargs
+        retry_kwargs = client.chat.completions.create.call_args_list[1].kwargs
+        assert first_kwargs["temperature"] == 0.3
+        assert "temperature" not in retry_kwargs
+        assert retry_kwargs["max_tokens"] == 500
+
+    @pytest.mark.asyncio
+    async def test_async_non_temperature_400_does_not_retry(self):
+        client = MagicMock()
+        client.base_url = "https://api.openai.com/v1"
+        client.chat.completions.create = AsyncMock(
+            side_effect=RuntimeError("HTTP 400: Invalid value: 'tool'"),
+        )
+
+        with (
+            patch("agent.auxiliary_client._resolve_task_provider_model",
+                  return_value=("openai-codex", "gpt-5.5", None, None, None)),
+            patch("agent.auxiliary_client._get_cached_client",
+                  return_value=(client, "gpt-5.5")),
+            patch("agent.auxiliary_client._validate_llm_response",
+                  side_effect=lambda resp, _task: resp),
+            patch("agent.auxiliary_client._try_payment_fallback",
+                  return_value=None),
+        ):
+            with pytest.raises(RuntimeError, match="Invalid value"):
+                await async_call_llm(
+                    task="session_search",
+                    messages=[{"role": "user", "content": "x"}],
+                    temperature=0.3,
+                    max_tokens=500,
+                )
+        assert client.chat.completions.create.await_count == 1

tests/agent/transports/test_chat_completions.py

@@ -33,15 +33,18 @@ class TestChatCompletionsBasic:
     def test_convert_messages_strips_codex_fields(self, transport):
         msgs = [
             {"role": "assistant", "content": "ok", "codex_reasoning_items": [{"id": "rs_1"}],
+             "codex_message_items": [{"id": "msg_1", "type": "message"}],
              "tool_calls": [{"id": "call_1", "call_id": "call_1", "response_item_id": "fc_1",
                             "type": "function", "function": {"name": "t", "arguments": "{}"}}]},
         ]
         result = transport.convert_messages(msgs)
         assert "codex_reasoning_items" not in result[0]
+        assert "codex_message_items" not in result[0]
         assert "call_id" not in result[0]["tool_calls"][0]
         assert "response_item_id" not in result[0]["tool_calls"][0]
         # Original list untouched (deepcopy-on-demand)
         assert "codex_reasoning_items" in msgs[0]
+        assert "codex_message_items" in msgs[0]
 
 
 class TestChatCompletionsBuildKwargs:

tests/agent/transports/test_codex_transport.py

@@ -194,6 +194,36 @@ class TestCodexNormalizeResponse:
         assert nr.content == "Hello world"
         assert nr.finish_reason == "stop"
 
+    def test_message_items_preserved_in_provider_data(self, transport):
+        """Codex assistant message item ids/phases must survive transport normalization."""
+        r = SimpleNamespace(
+            output=[
+                SimpleNamespace(
+                    type="message",
+                    role="assistant",
+                    id="msg_abc",
+                    phase="final_answer",
+                    content=[SimpleNamespace(type="output_text", text="Hello world")],
+                    status="completed",
+                ),
+            ],
+            status="completed",
+            incomplete_details=None,
+            usage=SimpleNamespace(input_tokens=10, output_tokens=5,
+                                  input_tokens_details=None, output_tokens_details=None),
+        )
+        nr = transport.normalize_response(r)
+        assert nr.codex_message_items == [
+            {
+                "type": "message",
+                "role": "assistant",
+                "status": "completed",
+                "content": [{"type": "output_text", "text": "Hello world"}],
+                "id": "msg_abc",
+                "phase": "final_answer",
+            }
+        ]
+
     def test_tool_call_response(self, transport):
         """Normalize a Codex response with tool calls."""
         r = SimpleNamespace(

tests/agent/transports/test_transport.py

@@ -60,6 +60,13 @@ class TestTransportRegistry:
         assert t is not None
         assert t.api_mode == "anthropic_messages"
 
+    def test_discovers_missing_transport_when_registry_partially_populated(self):
+        """Importing one transport directly must not hide other valid api_modes."""
+        import agent.transports.chat_completions  # noqa: F401
+        t = get_transport("codex_responses")
+        assert t is not None
+        assert t.api_mode == "codex_responses"
+
     def test_register_and_get(self):
         class DummyTransport(ProviderTransport):
             @property

tests/agent/transports/test_types.py

@@ -270,3 +270,15 @@ class TestNormalizedResponseBackwardCompat:
     def test_codex_reasoning_items_none_when_absent(self):
         nr = NormalizedResponse(content="hi", tool_calls=None, finish_reason="stop")
         assert nr.codex_reasoning_items is None
+
+    def test_codex_message_items_from_provider_data(self):
+        items = [{"id": "msg_1", "type": "message"}]
+        nr = NormalizedResponse(
+            content="hi", tool_calls=None, finish_reason="stop",
+            provider_data={"codex_message_items": items},
+        )
+        assert nr.codex_message_items == items
+
+    def test_codex_message_items_none_when_absent(self):
+        nr = NormalizedResponse(content="hi", tool_calls=None, finish_reason="stop")
+        assert nr.codex_message_items is None

tests/cli/test_branch_command.py

@@ -160,6 +160,30 @@ class TestBranchCommandCLI:
         assert agent.reset_session_state.called
         assert agent._last_flushed_db_idx == 4  # len(conversation_history)
 
+    def test_branch_updates_agent_session_log_file(self, cli_instance, session_db, tmp_path):
+        """Branching must redirect the agent's session_log_file to the new session's path."""
+        from cli import HermesCLI
+        from pathlib import Path
+
+        logs_dir = tmp_path / "sessions"
+        logs_dir.mkdir()
+
+        agent = MagicMock()
+        agent._last_flushed_db_idx = 0
+        agent.logs_dir = logs_dir
+        agent.session_log_file = logs_dir / f"session_{cli_instance.session_id}.json"
+        cli_instance.agent = agent
+
+        old_log_file = agent.session_log_file
+        HermesCLI._handle_branch_command(cli_instance, "/branch")
+
+        new_session_id = cli_instance.session_id
+        expected_log = logs_dir / f"session_{new_session_id}.json"
+        assert agent.session_log_file == expected_log, (
+            "session_log_file must point to the branch session, not the original"
+        )
+        assert agent.session_log_file != old_log_file
+
     def test_branch_sets_resumed_flag(self, cli_instance, session_db):
         """Branch should set _resumed=True to prevent auto-title generation."""
         from cli import HermesCLI