fix(api): send tool progress as custom SSE event to prevent model corruption (#6972)

Tool progress markers (e.g. `⏰ list`) were injected directly into
SSE delta.content chunks. OpenAI-compatible frontends (Open WebUI,
LobeChat, etc.) store delta.content verbatim as the assistant message
and send it back on subsequent requests. After enough turns, the model
learns to emit these markers as plain text instead of issuing real tool
calls — silently hallucinating tool results without ever running them.

Fix: Send tool progress as a custom `event: hermes.tool.progress` SSE
event instead of mixing it into delta.content. Per the SSE spec, clients
that don't understand a custom event type silently ignore it, so this is
backward-compatible. Frontends that want to render progress indicators
can listen for the custom event without persisting it to conversation
history.

The /v1/runs endpoint already uses structured events — this aligns the
/v1/chat/completions streaming path with the same principle.

Closes #6972

Dockerfile

@@ -13,7 +13,8 @@ COPY . /opt/hermes
 WORKDIR /opt/hermes
 
 # Install Python and Node dependencies in one layer, no cache
-RUN pip install --no-cache-dir -e ".[all]" --break-system-packages && \
+RUN pip install --no-cache-dir uv --break-system-packages && \
+    uv pip install --system --break-system-packages --no-cache -e ".[all]" && \
     npm install --prefer-offline --no-audit && \
     npx playwright install --with-deps chromium --only-shell && \
     cd /opt/hermes/scripts/whatsapp-bridge && \

README.md

@@ -33,8 +33,10 @@ Use any model you want — [Nous Portal](https://portal.nousresearch.com), [Open
 curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
 ```
 
-Works on Linux, macOS, and WSL2. The installer handles everything — Python, Node.js, dependencies, and the `hermes` command. No prerequisites except git.
+Works on Linux, macOS, WSL2, and Android via Termux. The installer handles the platform-specific setup for you.
 
+> **Android / Termux:** The tested manual path is documented in the [Termux guide](https://hermes-agent.nousresearch.com/docs/getting-started/termux). On Termux, Hermes installs a curated `.[termux]` extra because the full `.[all]` extra currently pulls Android-incompatible voice dependencies.
+>
 > **Windows:** Native Windows is not supported. Please install [WSL2](https://learn.microsoft.com/en-us/windows/wsl/install) and run the command above.
 
 After installation:

acp_adapter/server.py

@@ -36,6 +36,7 @@ from acp.schema import (
     SessionCapabilities,
     SessionForkCapabilities,
     SessionListCapabilities,
+    SessionResumeCapabilities,
     SessionInfo,
     TextContentBlock,
     UnstructuredCommandInput,
@@ -245,9 +246,11 @@ class HermesACPAgent(acp.Agent):
             protocol_version=acp.PROTOCOL_VERSION,
             agent_info=Implementation(name="hermes-agent", version=HERMES_VERSION),
             agent_capabilities=AgentCapabilities(
+                load_session=True,
                 session_capabilities=SessionCapabilities(
                     fork=SessionForkCapabilities(),
                     list=SessionListCapabilities(),
+                    resume=SessionResumeCapabilities(),
                 ),
             ),
             auth_methods=auth_methods,
@@ -451,14 +454,13 @@ class HermesACPAgent(acp.Agent):
             await conn.session_update(session_id, update)
 
         usage = None
-        usage_data = result.get("usage")
-        if usage_data and isinstance(usage_data, dict):
+        if any(result.get(key) is not None for key in ("prompt_tokens", "completion_tokens", "total_tokens")):
             usage = Usage(
-                input_tokens=usage_data.get("prompt_tokens", 0),
-                output_tokens=usage_data.get("completion_tokens", 0),
-                total_tokens=usage_data.get("total_tokens", 0),
-                thought_tokens=usage_data.get("reasoning_tokens"),
-                cached_read_tokens=usage_data.get("cached_tokens"),
+                input_tokens=result.get("prompt_tokens", 0),
+                output_tokens=result.get("completion_tokens", 0),
+                total_tokens=result.get("total_tokens", 0),
+                thought_tokens=result.get("reasoning_tokens"),
+                cached_read_tokens=result.get("cache_read_tokens"),
             )
 
         stop_reason = "cancelled" if state.cancel_event and state.cancel_event.is_set() else "end_turn"

agent/anthropic_adapter.py

@@ -74,8 +74,11 @@ def _get_anthropic_max_output(model: str) -> int:
     model IDs (claude-sonnet-4-5-20250929) and variant suffixes (:1m, :fast)
     resolve correctly.  Longest-prefix match wins to avoid e.g. "claude-3-5"
     matching before "claude-3-5-sonnet".
+
+    Normalizes dots to hyphens so that model names like
+    ``anthropic/claude-opus-4.6`` match the ``claude-opus-4-6`` table key.
     """
-    m = model.lower()
+    m = model.lower().replace(".", "-")
     best_key = ""
     best_val = _ANTHROPIC_DEFAULT_OUTPUT_LIMIT
     for key, val in _ANTHROPIC_OUTPUT_LIMITS.items():
@@ -95,6 +98,15 @@ _COMMON_BETAS = [
     "interleaved-thinking-2025-05-14",
     "fine-grained-tool-streaming-2025-05-14",
 ]
+# MiniMax's Anthropic-compatible endpoints fail tool-use requests when
+# the fine-grained tool streaming beta is present.  Omit it so tool calls
+# fall back to the provider's default response path.
+_TOOL_STREAMING_BETA = "fine-grained-tool-streaming-2025-05-14"
+
+# Fast mode beta — enables the ``speed: "fast"`` request parameter for
+# significantly higher output token throughput on Opus 4.6 (~2.5x).
+# See https://platform.claude.com/docs/en/build-with-claude/fast-mode
+_FAST_MODE_BETA = "fast-mode-2026-02-01"
 
 # Additional beta headers required for OAuth/subscription auth.
 # Matches what Claude Code (and pi-ai / OpenCode) send.
@@ -204,6 +216,19 @@ def _requires_bearer_auth(base_url: str | None) -> bool:
     return normalized.startswith(("https://api.minimax.io/anthropic", "https://api.minimaxi.com/anthropic"))
 
 
+def _common_betas_for_base_url(base_url: str | None) -> list[str]:
+    """Return the beta headers that are safe for the configured endpoint.
+
+    MiniMax's Anthropic-compatible endpoints (Bearer-auth) reject requests
+    that include Anthropic's ``fine-grained-tool-streaming`` beta — every
+    tool-use message triggers a connection error.  Strip that beta for
+    Bearer-auth endpoints while keeping all other betas intact.
+    """
+    if _requires_bearer_auth(base_url):
+        return [b for b in _COMMON_BETAS if b != _TOOL_STREAMING_BETA]
+    return _COMMON_BETAS
+
+
 def build_anthropic_client(api_key: str, base_url: str = None):
     """Create an Anthropic client, auto-detecting setup-tokens vs API keys.
 
@@ -222,6 +247,7 @@ def build_anthropic_client(api_key: str, base_url: str = None):
     }
     if normalized_base_url:
         kwargs["base_url"] = normalized_base_url
+    common_betas = _common_betas_for_base_url(normalized_base_url)
 
     if _requires_bearer_auth(normalized_base_url):
         # Some Anthropic-compatible providers (e.g. MiniMax) expect the API key in
@@ -231,21 +257,21 @@ def build_anthropic_client(api_key: str, base_url: str = None):
         # not use Anthropic's sk-ant-api prefix and would otherwise be misread as
         # Anthropic OAuth/setup tokens.
         kwargs["auth_token"] = api_key
-        if _COMMON_BETAS:
-            kwargs["default_headers"] = {"anthropic-beta": ",".join(_COMMON_BETAS)}
+        if common_betas:
+            kwargs["default_headers"] = {"anthropic-beta": ",".join(common_betas)}
     elif _is_third_party_anthropic_endpoint(base_url):
         # Third-party proxies (Azure AI Foundry, AWS Bedrock, etc.) use their
         # own API keys with x-api-key auth. Skip OAuth detection — their keys
         # don't follow Anthropic's sk-ant-* prefix convention and would be
         # misclassified as OAuth tokens.
         kwargs["api_key"] = api_key
-        if _COMMON_BETAS:
-            kwargs["default_headers"] = {"anthropic-beta": ",".join(_COMMON_BETAS)}
+        if common_betas:
+            kwargs["default_headers"] = {"anthropic-beta": ",".join(common_betas)}
     elif _is_oauth_token(api_key):
         # OAuth access token / setup-token → Bearer auth + Claude Code identity.
         # Anthropic routes OAuth requests based on user-agent and headers;
         # without Claude Code's fingerprint, requests get intermittent 500s.
-        all_betas = _COMMON_BETAS + _OAUTH_ONLY_BETAS
+        all_betas = common_betas + _OAUTH_ONLY_BETAS
         kwargs["auth_token"] = api_key
         kwargs["default_headers"] = {
             "anthropic-beta": ",".join(all_betas),
@@ -255,8 +281,8 @@ def build_anthropic_client(api_key: str, base_url: str = None):
     else:
         # Regular API key → x-api-key header + common betas
         kwargs["api_key"] = api_key
-        if _COMMON_BETAS:
-            kwargs["default_headers"] = {"anthropic-beta": ",".join(_COMMON_BETAS)}
+        if common_betas:
+            kwargs["default_headers"] = {"anthropic-beta": ",".join(common_betas)}
 
     return _anthropic_sdk.Anthropic(**kwargs)
 
@@ -1158,6 +1184,7 @@ def build_anthropic_kwargs(
     preserve_dots: bool = False,
     context_length: Optional[int] = None,
     base_url: str | None = None,
+    fast_mode: bool = False,
 ) -> Dict[str, Any]:
     """Build kwargs for anthropic.messages.create().
 
@@ -1191,6 +1218,10 @@ def build_anthropic_kwargs(
 
     When *base_url* points to a third-party Anthropic-compatible endpoint,
     thinking block signatures are stripped (they are Anthropic-proprietary).
+
+    When *fast_mode* is True, adds ``speed: "fast"`` and the fast-mode beta
+    header for ~2.5x faster output throughput on Opus 4.6.  Currently only
+    supported on native Anthropic endpoints (not third-party compatible ones).
     """
     system, anthropic_messages = convert_messages_to_anthropic(messages, base_url=base_url)
     anthropic_tools = convert_tools_to_anthropic(tools) if tools else []
@@ -1289,6 +1320,20 @@ def build_anthropic_kwargs(
                 kwargs["temperature"] = 1
                 kwargs["max_tokens"] = max(effective_max_tokens, budget + 4096)
 
+    # ── Fast mode (Opus 4.6 only) ────────────────────────────────────
+    # Adds speed:"fast" + the fast-mode beta header for ~2.5x output speed.
+    # Only for native Anthropic endpoints — third-party providers would
+    # reject the unknown beta header and speed parameter.
+    if fast_mode and not _is_third_party_anthropic_endpoint(base_url):
+        kwargs["speed"] = "fast"
+        # Build extra_headers with ALL applicable betas (the per-request
+        # extra_headers override the client-level anthropic-beta header).
+        betas = list(_common_betas_for_base_url(base_url))
+        if is_oauth:
+            betas.extend(_OAUTH_ONLY_BETAS)
+        betas.append(_FAST_MODE_BETA)
+        kwargs["extra_headers"] = {"anthropic-beta": ",".join(betas)}
+
     return kwargs
 
 
@@ -1350,4 +1395,4 @@ def normalize_anthropic_response(
             reasoning_details=reasoning_details or None,
         ),
         finish_reason,
-    )
+    )

agent/auxiliary_client.py

@@ -687,6 +687,15 @@ def _resolve_api_key_provider() -> Tuple[Optional[OpenAI], Optional[str]]:
         if pconfig.auth_type != "api_key":
             continue
         if provider_id == "anthropic":
+            # Only try anthropic when the user has explicitly configured it.
+            # Without this gate, Claude Code credentials get silently used
+            # as auxiliary fallback when the user's primary provider fails.
+            try:
+                from hermes_cli.auth import is_provider_explicitly_configured
+                if not is_provider_explicitly_configured("anthropic"):
+                    continue
+            except ImportError:
+                pass
             return _try_anthropic()
 
         pool_present, entry = _select_pool_entry(provider_id)
@@ -702,7 +711,7 @@ def _resolve_api_key_provider() -> Tuple[Optional[OpenAI], Optional[str]]:
             logger.debug("Auxiliary text client: %s (%s) via pool", pconfig.name, model)
             extra = {}
             if "api.kimi.com" in base_url.lower():
-                extra["default_headers"] = {"User-Agent": "KimiCLI/1.3"}
+                extra["default_headers"] = {"User-Agent": "KimiCLI/1.30.0"}
             elif "api.githubcopilot.com" in base_url.lower():
                 from hermes_cli.models import copilot_default_headers
 
@@ -721,7 +730,7 @@ def _resolve_api_key_provider() -> Tuple[Optional[OpenAI], Optional[str]]:
         logger.debug("Auxiliary text client: %s (%s)", pconfig.name, model)
         extra = {}
         if "api.kimi.com" in base_url.lower():
-            extra["default_headers"] = {"User-Agent": "KimiCLI/1.3"}
+            extra["default_headers"] = {"User-Agent": "KimiCLI/1.30.0"}
         elif "api.githubcopilot.com" in base_url.lower():
             from hermes_cli.models import copilot_default_headers
 
@@ -848,7 +857,7 @@ def _read_main_provider() -> str:
     return ""
 
 
-def _resolve_custom_runtime() -> Tuple[Optional[str], Optional[str]]:
+def _resolve_custom_runtime() -> Tuple[Optional[str], Optional[str], Optional[str]]:
     """Resolve the active custom/main endpoint the same way the main CLI does.
 
     This covers both env-driven OPENAI_BASE_URL setups and config-saved custom
@@ -861,18 +870,29 @@ def _resolve_custom_runtime() -> Tuple[Optional[str], Optional[str]]:
         runtime = resolve_runtime_provider(requested="custom")
     except Exception as exc:
         logger.debug("Auxiliary client: custom runtime resolution failed: %s", exc)
-        return None, None
+        runtime = None
+
+    if not isinstance(runtime, dict):
+        openai_base = os.getenv("OPENAI_BASE_URL", "").strip().rstrip("/")
+        openai_key = os.getenv("OPENAI_API_KEY", "").strip()
+        if not openai_base:
+            return None, None, None
+        runtime = {
+            "base_url": openai_base,
+            "api_key": openai_key,
+        }
 
     custom_base = runtime.get("base_url")
     custom_key = runtime.get("api_key")
+    custom_mode = runtime.get("api_mode")
     if not isinstance(custom_base, str) or not custom_base.strip():
-        return None, None
+        return None, None, None
 
     custom_base = custom_base.strip().rstrip("/")
     if "openrouter.ai" in custom_base.lower():
         # requested='custom' falls back to OpenRouter when no custom endpoint is
         # configured. Treat that as "no custom endpoint" for auxiliary routing.
-        return None, None
+        return None, None, None
 
     # Local servers (Ollama, llama.cpp, vLLM, LM Studio) don't require auth.
     # Use a placeholder key — the OpenAI SDK requires a non-empty string but
@@ -881,20 +901,33 @@ def _resolve_custom_runtime() -> Tuple[Optional[str], Optional[str]]:
     if not isinstance(custom_key, str) or not custom_key.strip():
         custom_key = "no-key-required"
 
-    return custom_base, custom_key.strip()
+    if not isinstance(custom_mode, str) or not custom_mode.strip():
+        custom_mode = None
+
+    return custom_base, custom_key.strip(), custom_mode
 
 
 def _current_custom_base_url() -> str:
-    custom_base, _ = _resolve_custom_runtime()
+    custom_base, _, _ = _resolve_custom_runtime()
     return custom_base or ""
 
 
 def _try_custom_endpoint() -> Tuple[Optional[OpenAI], Optional[str]]:
-    custom_base, custom_key = _resolve_custom_runtime()
+    runtime = _resolve_custom_runtime()
+    if len(runtime) == 2:
+        custom_base, custom_key = runtime
+        custom_mode = None
+    else:
+        custom_base, custom_key, custom_mode = runtime
     if not custom_base or not custom_key:
         return None, None
+    if custom_base.lower().startswith(_CODEX_AUX_BASE_URL.lower()):
+        return None, None
     model = _read_main_model() or "gpt-4o-mini"
-    logger.debug("Auxiliary client: custom endpoint (%s)", model)
+    logger.debug("Auxiliary client: custom endpoint (%s, api_mode=%s)", model, custom_mode or "chat_completions")
+    if custom_mode == "codex_responses":
+        real_client = OpenAI(api_key=custom_key, base_url=custom_base)
+        return CodexAuxiliaryClient(real_client, model), model
     return OpenAI(api_key=custom_key, base_url=custom_base), model
 
 
@@ -1161,10 +1194,22 @@ def _to_async_client(sync_client, model: str):
 
         async_kwargs["default_headers"] = copilot_default_headers()
     elif "api.kimi.com" in base_lower:
-        async_kwargs["default_headers"] = {"User-Agent": "KimiCLI/1.3"}
+        async_kwargs["default_headers"] = {"User-Agent": "KimiCLI/1.30.0"}
     return AsyncOpenAI(**async_kwargs), model
 
 
+def _normalize_resolved_model(model_name: Optional[str], provider: str) -> Optional[str]:
+    """Normalize a resolved model for the provider that will receive it."""
+    if not model_name:
+        return model_name
+    try:
+        from hermes_cli.model_normalize import normalize_model_for_provider
+
+        return normalize_model_for_provider(model_name, provider)
+    except Exception:
+        return model_name
+
+
 def resolve_provider_client(
     provider: str,
     model: str = None,
@@ -1227,7 +1272,7 @@ def resolve_provider_client(
             logger.warning("resolve_provider_client: openrouter requested "
                            "but OPENROUTER_API_KEY not set")
             return None, None
-        final_model = model or default
+        final_model = _normalize_resolved_model(model or default, provider)
         return (_to_async_client(client, final_model) if async_mode
                 else (client, final_model))
 
@@ -1238,7 +1283,7 @@ def resolve_provider_client(
             logger.warning("resolve_provider_client: nous requested "
                            "but Nous Portal not configured (run: hermes auth)")
             return None, None
-        final_model = model or default
+        final_model = _normalize_resolved_model(model or default, provider)
         return (_to_async_client(client, final_model) if async_mode
                 else (client, final_model))
 
@@ -1252,7 +1297,7 @@ def resolve_provider_client(
                 logger.warning("resolve_provider_client: openai-codex requested "
                                "but no Codex OAuth token found (run: hermes model)")
                 return None, None
-            final_model = model or _CODEX_AUX_MODEL
+            final_model = _normalize_resolved_model(model or _CODEX_AUX_MODEL, provider)
             raw_client = OpenAI(api_key=codex_token, base_url=_CODEX_AUX_BASE_URL)
             return (raw_client, final_model)
         # Standard path: wrap in CodexAuxiliaryClient adapter
@@ -1261,7 +1306,7 @@ def resolve_provider_client(
             logger.warning("resolve_provider_client: openai-codex requested "
                            "but no Codex OAuth token found (run: hermes model)")
             return None, None
-        final_model = model or default
+        final_model = _normalize_resolved_model(model or default, provider)
         return (_to_async_client(client, final_model) if async_mode
                 else (client, final_model))
 
@@ -1280,10 +1325,13 @@ def resolve_provider_client(
                     "but base_url is empty"
                 )
                 return None, None
-            final_model = model or _read_main_model() or "gpt-4o-mini"
+            final_model = _normalize_resolved_model(
+                model or _read_main_model() or "gpt-4o-mini",
+                provider,
+            )
             extra = {}
             if "api.kimi.com" in custom_base.lower():
-                extra["default_headers"] = {"User-Agent": "KimiCLI/1.3"}
+                extra["default_headers"] = {"User-Agent": "KimiCLI/1.30.0"}
             elif "api.githubcopilot.com" in custom_base.lower():
                 from hermes_cli.models import copilot_default_headers
                 extra["default_headers"] = copilot_default_headers()
@@ -1295,7 +1343,7 @@ def resolve_provider_client(
                        _resolve_api_key_provider):
             client, default = try_fn()
             if client is not None:
-                final_model = model or default
+                final_model = _normalize_resolved_model(model or default, provider)
                 return (_to_async_client(client, final_model) if async_mode
                         else (client, final_model))
         logger.warning("resolve_provider_client: custom/main requested "
@@ -1310,7 +1358,10 @@ def resolve_provider_client(
             custom_base = custom_entry.get("base_url", "").strip()
             custom_key = custom_entry.get("api_key", "").strip() or "no-key-required"
             if custom_base:
-                final_model = model or _read_main_model() or "gpt-4o-mini"
+                final_model = _normalize_resolved_model(
+                    model or _read_main_model() or "gpt-4o-mini",
+                    provider,
+                )
                 client = OpenAI(api_key=custom_key, base_url=custom_base)
                 logger.debug(
                     "resolve_provider_client: named custom provider %r (%s)",
@@ -1342,7 +1393,7 @@ def resolve_provider_client(
             if client is None:
                 logger.warning("resolve_provider_client: anthropic requested but no Anthropic credentials found")
                 return None, None
-            final_model = model or default_model
+            final_model = _normalize_resolved_model(model or default_model, provider)
             return (_to_async_client(client, final_model) if async_mode else (client, final_model))
 
         creds = resolve_api_key_provider_credentials(provider)
@@ -1361,12 +1412,12 @@ def resolve_provider_client(
         )
 
         default_model = _API_KEY_PROVIDER_AUX_MODELS.get(provider, "")
-        final_model = model or default_model
+        final_model = _normalize_resolved_model(model or default_model, provider)
 
         # Provider-specific headers
         headers = {}
         if "api.kimi.com" in base_url.lower():
-            headers["User-Agent"] = "KimiCLI/1.3"
+            headers["User-Agent"] = "KimiCLI/1.30.0"
         elif "api.githubcopilot.com" in base_url.lower():
             from hermes_cli.models import copilot_default_headers
 

agent/context_references.py

@@ -13,8 +13,9 @@ from typing import Awaitable, Callable
 
 from agent.model_metadata import estimate_tokens_rough
 
+_QUOTED_REFERENCE_VALUE = r'(?:`[^`\n]+`|"[^"\n]+"|\'[^\'\n]+\')'
 REFERENCE_PATTERN = re.compile(
-    r"(?<![\w/])@(?:(?P<simple>diff|staged)\b|(?P<kind>file|folder|git|url):(?P<value>\S+))"
+    rf"(?<![\w/])@(?:(?P<simple>diff|staged)\b|(?P<kind>file|folder|git|url):(?P<value>{_QUOTED_REFERENCE_VALUE}(?::\d+(?:-\d+)?)?|\S+))"
 )
 TRAILING_PUNCTUATION = ",.;!?"
 _SENSITIVE_HOME_DIRS = (".ssh", ".aws", ".gnupg", ".kube", ".docker", ".azure", ".config/gh")
@@ -81,14 +82,10 @@ def parse_context_references(message: str) -> list[ContextReference]:
         value = _strip_trailing_punctuation(match.group("value") or "")
         line_start = None
         line_end = None
-        target = value
+        target = _strip_reference_wrappers(value)
 
         if kind == "file":
-            range_match = re.match(r"^(?P<path>.+?):(?P<start>\d+)(?:-(?P<end>\d+))?$", value)
-            if range_match:
-                target = range_match.group("path")
-                line_start = int(range_match.group("start"))
-                line_end = int(range_match.group("end") or range_match.group("start"))
+            target, line_start, line_end = _parse_file_reference_value(value)
 
         refs.append(
             ContextReference(
@@ -375,6 +372,38 @@ def _strip_trailing_punctuation(value: str) -> str:
     return stripped
 
 
+def _strip_reference_wrappers(value: str) -> str:
+    if len(value) >= 2 and value[0] == value[-1] and value[0] in "`\"'":
+        return value[1:-1]
+    return value
+
+
+def _parse_file_reference_value(value: str) -> tuple[str, int | None, int | None]:
+    quoted_match = re.match(
+        r'^(?P<quote>`|"|\')(?P<path>.+?)(?P=quote)(?::(?P<start>\d+)(?:-(?P<end>\d+))?)?$',
+        value,
+    )
+    if quoted_match:
+        line_start = quoted_match.group("start")
+        line_end = quoted_match.group("end")
+        return (
+            quoted_match.group("path"),
+            int(line_start) if line_start is not None else None,
+            int(line_end or line_start) if line_start is not None else None,
+        )
+
+    range_match = re.match(r"^(?P<path>.+?):(?P<start>\d+)(?:-(?P<end>\d+))?$", value)
+    if range_match:
+        line_start = int(range_match.group("start"))
+        return (
+            range_match.group("path"),
+            line_start,
+            int(range_match.group("end") or range_match.group("start")),
+        )
+
+    return _strip_reference_wrappers(value), None, None
+
+
 def _remove_reference_tokens(message: str, refs: list[ContextReference]) -> str:
     pieces: list[str] = []
     cursor = 0

agent/credential_pool.py

@@ -20,6 +20,7 @@ from hermes_cli.auth import (
     DEFAULT_AGENT_KEY_MIN_TTL_SECONDS,
     KIMI_CODE_BASE_URL,
     PROVIDER_REGISTRY,
+    _auth_store_lock,
     _codex_access_token_is_expiring,
     _decode_jwt_claims,
     _import_codex_cli_tokens,
@@ -27,6 +28,8 @@ from hermes_cli.auth import (
     _load_provider_state,
     _resolve_kimi_base_url,
     _resolve_zai_base_url,
+    _save_auth_store,
+    _save_provider_state,
     read_credential_pool,
     write_credential_pool,
 )
@@ -479,6 +482,67 @@ class CredentialPool:
             logger.debug("Failed to sync from ~/.codex/auth.json: %s", exc)
         return entry
 
+    def _sync_device_code_entry_to_auth_store(self, entry: PooledCredential) -> None:
+        """Write refreshed pool entry tokens back to auth.json providers.
+
+        After a pool-level refresh, the pool entry has fresh tokens but
+        auth.json's ``providers.<id>`` still holds the pre-refresh state.
+        On the next ``load_pool()``, ``_seed_from_singletons()`` reads that
+        stale state and can overwrite the fresh pool entry — potentially
+        re-seeding a consumed single-use refresh token.
+
+        Applies to any OAuth provider whose singleton lives in auth.json
+        (currently Nous and OpenAI Codex).
+        """
+        if entry.source != "device_code":
+            return
+        try:
+            with _auth_store_lock():
+                auth_store = _load_auth_store()
+                if self.provider == "nous":
+                    state = _load_provider_state(auth_store, "nous")
+                    if state is None:
+                        return
+                    state["access_token"] = entry.access_token
+                    if entry.refresh_token:
+                        state["refresh_token"] = entry.refresh_token
+                    if entry.expires_at:
+                        state["expires_at"] = entry.expires_at
+                    if entry.agent_key:
+                        state["agent_key"] = entry.agent_key
+                    if entry.agent_key_expires_at:
+                        state["agent_key_expires_at"] = entry.agent_key_expires_at
+                    for extra_key in ("obtained_at", "expires_in", "agent_key_id",
+                                      "agent_key_expires_in", "agent_key_reused",
+                                      "agent_key_obtained_at"):
+                        val = entry.extra.get(extra_key)
+                        if val is not None:
+                            state[extra_key] = val
+                    if entry.inference_base_url:
+                        state["inference_base_url"] = entry.inference_base_url
+                    _save_provider_state(auth_store, "nous", state)
+
+                elif self.provider == "openai-codex":
+                    state = _load_provider_state(auth_store, "openai-codex")
+                    if not isinstance(state, dict):
+                        return
+                    tokens = state.get("tokens")
+                    if not isinstance(tokens, dict):
+                        return
+                    tokens["access_token"] = entry.access_token
+                    if entry.refresh_token:
+                        tokens["refresh_token"] = entry.refresh_token
+                    if entry.last_refresh:
+                        state["last_refresh"] = entry.last_refresh
+                    _save_provider_state(auth_store, "openai-codex", state)
+
+                else:
+                    return
+
+                _save_auth_store(auth_store)
+        except Exception as exc:
+            logger.debug("Failed to sync %s pool entry back to auth store: %s", self.provider, exc)
+
     def _refresh_entry(self, entry: PooledCredential, *, force: bool) -> Optional[PooledCredential]:
         if entry.auth_type != AUTH_TYPE_OAUTH or not entry.refresh_token:
             if force:
@@ -513,6 +577,13 @@ class CredentialPool:
                     except Exception as wexc:
                         logger.debug("Failed to write refreshed token to credentials file: %s", wexc)
             elif self.provider == "openai-codex":
+                # Proactively sync from ~/.codex/auth.json before refresh.
+                # The Codex CLI (or another Hermes profile) may have already
+                # consumed our refresh_token.  Syncing first avoids a
+                # "refresh_token_reused" error when the CLI has a newer pair.
+                synced = self._sync_codex_entry_from_cli(entry)
+                if synced is not entry:
+                    entry = synced
                 refreshed = auth_mod.refresh_codex_oauth_pure(
                     entry.access_token,
                     entry.refresh_token,
@@ -598,6 +669,37 @@ class CredentialPool:
                     # Credentials file had a valid (non-expired) token — use it directly
                     logger.debug("Credentials file has valid token, using without refresh")
                     return synced
+            # For openai-codex: the refresh_token may have been consumed by
+            # the Codex CLI between our proactive sync and the refresh call.
+            # Re-sync and retry once.
+            if self.provider == "openai-codex":
+                synced = self._sync_codex_entry_from_cli(entry)
+                if synced.refresh_token != entry.refresh_token:
+                    logger.debug("Retrying Codex refresh with synced token from ~/.codex/auth.json")
+                    try:
+                        refreshed = auth_mod.refresh_codex_oauth_pure(
+                            synced.access_token,
+                            synced.refresh_token,
+                        )
+                        updated = replace(
+                            synced,
+                            access_token=refreshed["access_token"],
+                            refresh_token=refreshed["refresh_token"],
+                            last_refresh=refreshed.get("last_refresh"),
+                            last_status=STATUS_OK,
+                            last_status_at=None,
+                            last_error_code=None,
+                        )
+                        self._replace_entry(synced, updated)
+                        self._persist()
+                        self._sync_device_code_entry_to_auth_store(updated)
+                        return updated
+                    except Exception as retry_exc:
+                        logger.debug("Codex retry refresh also failed: %s", retry_exc)
+                elif not self._entry_needs_refresh(synced):
+                    logger.debug("Codex CLI has valid token, using without refresh")
+                    self._sync_device_code_entry_to_auth_store(synced)
+                    return synced
             self._mark_exhausted(entry, None)
             return None
 
@@ -612,6 +714,10 @@ class CredentialPool:
         )
         self._replace_entry(entry, updated)
         self._persist()
+        # Sync refreshed tokens back to auth.json providers so that
+        # _seed_from_singletons() on the next load_pool() sees fresh state
+        # instead of re-seeding stale/consumed tokens.
+        self._sync_device_code_entry_to_auth_store(updated)
         return updated
 
     def _entry_needs_refresh(self, entry: PooledCredential) -> bool:
@@ -953,6 +1059,17 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
     auth_store = _load_auth_store()
 
     if provider == "anthropic":
+        # Only auto-discover external credentials (Claude Code, Hermes PKCE)
+        # when the user has explicitly configured anthropic as their provider.
+        # Without this gate, auxiliary client fallback chains silently read
+        # ~/.claude/.credentials.json without user consent.  See PR #4210.
+        try:
+            from hermes_cli.auth import is_provider_explicitly_configured
+            if not is_provider_explicitly_configured("anthropic"):
+                return changed, active_sources
+        except ImportError:
+            pass
+
         from agent.anthropic_adapter import read_claude_code_credentials, read_hermes_oauth_credentials
 
         for source_name, creds in (
@@ -960,6 +1077,13 @@ def _seed_from_singletons(provider: str, entries: List[PooledCredential]) -> Tup
             ("claude_code", read_claude_code_credentials()),
         ):
             if creds and creds.get("accessToken"):
+                # Check if user explicitly removed this source
+                try:
+                    from hermes_cli.auth import is_source_suppressed
+                    if is_source_suppressed(provider, source_name):
+                        continue
+                except ImportError:
+                    pass
                 active_sources.add(source_name)
                 changed |= _upsert_entry(
                     entries,

agent/error_classifier.py

@@ -112,6 +112,7 @@ _RATE_LIMIT_PATTERNS = [
     "try again in",
     "please retry after",
     "resource_exhausted",
+    "rate increased too quickly",  # Alibaba/DashScope throttling
 ]
 
 # Usage-limit patterns that need disambiguation (could be billing OR rate_limit)
@@ -667,6 +668,27 @@ def _classify_by_message(
             should_compress=True,
         )
 
+    # Usage-limit patterns need the same disambiguation as 402: some providers
+    # surface "usage limit" errors without an HTTP status code.  A transient
+    # signal ("try again", "resets at", …) means it's a periodic quota, not
+    # billing exhaustion.
+    has_usage_limit = any(p in error_msg for p in _USAGE_LIMIT_PATTERNS)
+    if has_usage_limit:
+        has_transient_signal = any(p in error_msg for p in _USAGE_LIMIT_TRANSIENT_SIGNALS)
+        if has_transient_signal:
+            return result_fn(
+                FailoverReason.rate_limit,
+                retryable=True,
+                should_rotate_credential=True,
+                should_fallback=True,
+            )
+        return result_fn(
+            FailoverReason.billing,
+            retryable=False,
+            should_rotate_credential=True,
+            should_fallback=True,
+        )
+
     # Billing patterns
     if any(p in error_msg for p in _BILLING_PATTERNS):
         return result_fn(
@@ -694,11 +716,16 @@ def _classify_by_message(
         )
 
     # Auth patterns
+    # Auth errors should NOT be retried directly — the credential is invalid and
+    # retrying with the same key will always fail.  Set retryable=False so the
+    # caller triggers credential rotation (should_rotate_credential=True) or
+    # provider fallback rather than an immediate retry loop.
     if any(p in error_msg for p in _AUTH_PATTERNS):
         return result_fn(
             FailoverReason.auth,
-            retryable=True,
+            retryable=False,
             should_rotate_credential=True,
+            should_fallback=True,
         )
 
     # Model not found patterns

agent/model_metadata.py

@@ -126,6 +126,21 @@ DEFAULT_CONTEXT_LENGTHS = {
     "minimax": 1048576,
     # GLM
     "glm": 202752,
+    # xAI Grok — xAI /v1/models does not return context_length metadata,
+    # so these hardcoded fallbacks prevent Hermes from probing-down to
+    # the default 128k when the user points at https://api.x.ai/v1
+    # via a custom provider. Values sourced from models.dev (2026-04).
+    # Keys use substring matching (longest-first), so e.g. "grok-4.20"
+    # matches "grok-4.20-0309-reasoning" / "-non-reasoning" / "-multi-agent-0309".
+    "grok-code-fast": 256000,   # grok-code-fast-1
+    "grok-4-1-fast": 2000000,   # grok-4-1-fast-(non-)reasoning
+    "grok-2-vision": 8192,      # grok-2-vision, -1212, -latest
+    "grok-4-fast": 2000000,     # grok-4-fast-(non-)reasoning
+    "grok-4.20": 2000000,       # grok-4.20-0309-(non-)reasoning, -multi-agent-0309
+    "grok-4": 256000,           # grok-4, grok-4-0709
+    "grok-3": 131072,           # grok-3, grok-3-mini, grok-3-fast, grok-3-mini-fast
+    "grok-2": 131072,           # grok-2, grok-2-1212, grok-2-latest
+    "grok": 131072,             # catch-all (grok-beta, unknown grok-*)
     # Kimi
     "kimi": 262144,
     # Arcee
@@ -198,6 +213,7 @@ _URL_TO_PROVIDER: Dict[str, str] = {
     "models.github.ai": "copilot",
     "api.fireworks.ai": "fireworks",
     "opencode.ai": "opencode-go",
+    "api.x.ai": "xai",
 }
 
 

agent/prompt_builder.py

@@ -40,7 +40,7 @@ _CONTEXT_THREAT_PATTERNS = [
     (r'disregard\s+(your|all|any)\s+(instructions|rules|guidelines)', "disregard_rules"),
     (r'act\s+as\s+(if|though)\s+you\s+(have\s+no|don\'t\s+have)\s+(restrictions|limits|rules)', "bypass_restrictions"),
     (r'<!--[^>]*(?:ignore|override|system|secret|hidden)[^>]*-->', "html_comment_injection"),
-    (r'<\s*div\s+style\s*=\s*["\'].*display\s*:\s*none', "hidden_div"),
+    (r'<\s*div\s+style\s*=\s*["\'][\s\S]*?display\s*:\s*none', "hidden_div"),
     (r'translate\s+.*\s+into\s+.*\s+and\s+(execute|run|eval)', "translate_execute"),
     (r'curl\s+[^\n]*\$\{?\w*(KEY|TOKEN|SECRET|PASSWORD|CREDENTIAL|API)', "exfil_curl"),
     (r'cat\s+[^\n]*(\.env|credentials|\.netrc|\.pgpass)', "read_secrets"),
@@ -356,6 +356,14 @@ PLATFORM_HINTS = {
         "MEDIA:/absolute/path/to/file in your response. Images (.jpg, .png, "
         ".heic) appear as photos and other files arrive as attachments."
     ),
+    "weixin": (
+        "You are on Weixin/WeChat. Markdown formatting is supported, so you may use it when "
+        "it improves readability, but keep the message compact and chat-friendly. You can send media files natively: "
+        "include MEDIA:/absolute/path/to/file in your response. Images are sent as native "
+        "photos, videos play inline when supported, and other files arrive as downloadable "
+        "documents. You can also include image URLs in markdown format ![alt](url) and they "
+        "will be downloaded and sent as native media when possible."
+    ),
 }
 
 CONTEXT_FILE_MAX_CHARS = 20_000
@@ -479,7 +487,7 @@ def _parse_skill_file(skill_file: Path) -> tuple[bool, dict, str]:
     (True, {}, "") to err on the side of showing the skill.
     """
     try:
-        raw = skill_file.read_text(encoding="utf-8")[:2000]
+        raw = skill_file.read_text(encoding="utf-8")
         frontmatter, _ = parse_frontmatter(raw)
 
         if not skill_matches_platform(frontmatter):
@@ -487,7 +495,7 @@ def _parse_skill_file(skill_file: Path) -> tuple[bool, dict, str]:
 
         return True, frontmatter, extract_skill_description(frontmatter)
     except Exception as e:
-        logger.debug("Failed to parse skill file %s: %s", skill_file, e)
+        logger.warning("Failed to parse skill file %s: %s", skill_file, e)
         return True, {}, ""
 
 
@@ -550,9 +558,10 @@ def build_skills_system_prompt(
     # ── Layer 1: in-process LRU cache ─────────────────────────────────
     # Include the resolved platform so per-platform disabled-skill lists
     # produce distinct cache entries (gateway serves multiple platforms).
+    from gateway.session_context import get_session_env
     _platform_hint = (
         os.environ.get("HERMES_PLATFORM")
-        or os.environ.get("HERMES_SESSION_PLATFORM")
+        or get_session_env("HERMES_SESSION_PLATFORM")
         or ""
     )
     cache_key = (

agent/rate_limit_tracker.py

@@ -97,8 +97,12 @@ def parse_rate_limit_headers(
 
     Returns None if no rate limit headers are present.
     """
+    # Normalize to lowercase so lookups work regardless of how the server
+    # capitalises headers (HTTP header names are case-insensitive per RFC 7230).
+    lowered = {k.lower(): v for k, v in headers.items()}
+
     # Quick check: at least one rate limit header must exist
-    has_any = any(k.lower().startswith("x-ratelimit-") for k in headers)
+    has_any = any(k.startswith("x-ratelimit-") for k in lowered)
     if not has_any:
         return None
 
@@ -109,9 +113,9 @@ def parse_rate_limit_headers(
         #      resource="tokens", suffix="-1h" -> per-hour
         tag = f"{resource}{suffix}"
         return RateLimitBucket(
-            limit=_safe_int(headers.get(f"x-ratelimit-limit-{tag}")),
-            remaining=_safe_int(headers.get(f"x-ratelimit-remaining-{tag}")),
-            reset_seconds=_safe_float(headers.get(f"x-ratelimit-reset-{tag}")),
+            limit=_safe_int(lowered.get(f"x-ratelimit-limit-{tag}")),
+            remaining=_safe_int(lowered.get(f"x-ratelimit-remaining-{tag}")),
+            reset_seconds=_safe_float(lowered.get(f"x-ratelimit-reset-{tag}")),
             captured_at=now,
         )
 

agent/skill_utils.py

@@ -145,10 +145,11 @@ def get_disabled_skill_names(platform: str | None = None) -> Set[str]:
     if not isinstance(skills_cfg, dict):
         return set()
 
+    from gateway.session_context import get_session_env
     resolved_platform = (
         platform
         or os.getenv("HERMES_PLATFORM")
-        or os.getenv("HERMES_SESSION_PLATFORM")
+        or get_session_env("HERMES_SESSION_PLATFORM")
     )
     if resolved_platform:
         platform_disabled = (skills_cfg.get("platform_disabled") or {}).get(

agent/smart_model_routing.py

@@ -181,6 +181,7 @@ def resolve_turn_route(user_message: str, routing_config: Optional[Dict[str, Any
             "api_mode": runtime.get("api_mode"),
             "command": runtime.get("command"),
             "args": list(runtime.get("args") or []),
+            "credential_pool": runtime.get("credential_pool"),
         },
         "label": f"smart route → {route.get('model')} ({runtime.get('provider')})",
         "signature": (

cli-config.yaml.example

@@ -684,7 +684,11 @@ platform_toolsets:
 stt:
   enabled: true
   # provider: "local"          # auto-detected if omitted
-  model: "whisper-1"  # whisper-1 (cheapest) | gpt-4o-mini-transcribe | gpt-4o-transcribe
+  local:
+    model: "base"              # tiny | base | small | medium | large-v3 | turbo
+    # language: ""             # auto-detect; set to "en", "es", "fr", etc. to force
+  openai:
+    model: "whisper-1"         # whisper-1 | gpt-4o-mini-transcribe | gpt-4o-transcribe
   # mistral:
   #   model: "voxtral-mini-latest"  # voxtral-mini-latest | voxtral-mini-2602
 

constraints-termux.txt

@@ -0,0 +1,15 @@
+# Termux / Android dependency constraints for Hermes Agent.
+#
+# Usage:
+#   python -m pip install -e '.[termux]' -c constraints-termux.txt
+#
+# These pins keep the tested Android install path stable when upstream packages
+# move faster than Termux-compatible wheels / sdists.
+
+ipython<10
+jedi>=0.18.1,<0.20
+parso>=0.8.4,<0.9
+stack-data>=0.6,<0.7
+pexpect>4.3,<5
+matplotlib-inline>=0.1.7,<0.2
+asttokens>=2.1,<3

cron/jobs.py

@@ -31,7 +31,7 @@ except ImportError:
 # Configuration
 # =============================================================================
 
-HERMES_DIR = get_hermes_home()
+HERMES_DIR = get_hermes_home().resolve()
 CRON_DIR = HERMES_DIR / "cron"
 JOBS_FILE = CRON_DIR / "jobs.json"
 OUTPUT_DIR = CRON_DIR / "output"
@@ -338,10 +338,12 @@ def load_jobs() -> List[Dict[str, Any]]:
                     save_jobs(jobs)
                     logger.warning("Auto-repaired jobs.json (had invalid control characters)")
                 return jobs
-        except Exception:
-            return []
-    except IOError:
-        return []
+        except Exception as e:
+            logger.error("Failed to auto-repair jobs.json: %s", e)
+            raise RuntimeError(f"Cron database corrupted and unrepairable: {e}") from e
+    except IOError as e:
+        logger.error("IOError reading jobs.json: %s", e)
+        raise RuntimeError(f"Failed to read cron database: {e}") from e
 
 
 def save_jobs(jobs: List[Dict[str, Any]]):
@@ -452,6 +454,7 @@ def create_job(
         "last_run_at": None,
         "last_status": None,
         "last_error": None,
+        "last_delivery_error": None,
         # Delivery configuration
         "deliver": deliver,
         "origin": origin,  # Tracks where job was created for "origin" delivery
@@ -620,8 +623,8 @@ def mark_job_run(job_id: str, success: bool, error: Optional[str] = None,
 
             save_jobs(jobs)
             return
-    
-    save_jobs(jobs)
+
+    logger.warning("mark_job_run: job_id %s not found, skipping save", job_id)
 
 
 def advance_next_run(job_id: str) -> bool:

cron/scheduler.py

@@ -44,7 +44,7 @@ logger = logging.getLogger(__name__)
 _KNOWN_DELIVERY_PLATFORMS = frozenset({
     "telegram", "discord", "slack", "whatsapp", "signal",
     "matrix", "mattermost", "homeassistant", "dingtalk", "feishu",
-    "wecom", "sms", "email", "webhook", "bluebubbles",
+    "wecom", "weixin", "sms", "email", "webhook", "bluebubbles",
 })
 
 from cron.jobs import get_due_jobs, mark_job_run, save_job_output, advance_next_run
@@ -234,6 +234,7 @@ def _deliver_result(job: dict, content: str, adapters=None, loop=None) -> Option
         "dingtalk": Platform.DINGTALK,
         "feishu": Platform.FEISHU,
         "wecom": Platform.WECOM,
+        "weixin": Platform.WEIXIN,
         "email": Platform.EMAIL,
         "sms": Platform.SMS,
         "bluebubbles": Platform.BLUEBUBBLES,
@@ -346,7 +347,42 @@ def _deliver_result(job: dict, content: str, adapters=None, loop=None) -> Option
     return None
 
 
-_SCRIPT_TIMEOUT = 120  # seconds
+_DEFAULT_SCRIPT_TIMEOUT = 120  # seconds
+# Backward-compatible module override used by tests and emergency monkeypatches.
+_SCRIPT_TIMEOUT = _DEFAULT_SCRIPT_TIMEOUT
+
+
+def _get_script_timeout() -> int:
+    """Resolve cron pre-run script timeout from module/env/config with a safe default."""
+    if _SCRIPT_TIMEOUT != _DEFAULT_SCRIPT_TIMEOUT:
+        try:
+            timeout = int(float(_SCRIPT_TIMEOUT))
+            if timeout > 0:
+                return timeout
+        except Exception:
+            logger.warning("Invalid patched _SCRIPT_TIMEOUT=%r; using env/config/default", _SCRIPT_TIMEOUT)
+
+    env_value = os.getenv("HERMES_CRON_SCRIPT_TIMEOUT", "").strip()
+    if env_value:
+        try:
+            timeout = int(float(env_value))
+            if timeout > 0:
+                return timeout
+        except Exception:
+            logger.warning("Invalid HERMES_CRON_SCRIPT_TIMEOUT=%r; using config/default", env_value)
+
+    try:
+        cfg = load_config() or {}
+        cron_cfg = cfg.get("cron", {}) if isinstance(cfg, dict) else {}
+        configured = cron_cfg.get("script_timeout_seconds")
+        if configured is not None:
+            timeout = int(float(configured))
+            if timeout > 0:
+                return timeout
+    except Exception as exc:
+        logger.debug("Failed to load cron script timeout from config: %s", exc)
+
+    return _DEFAULT_SCRIPT_TIMEOUT
 
 
 def _run_job_script(script_path: str) -> tuple[bool, str]:
@@ -393,12 +429,14 @@ def _run_job_script(script_path: str) -> tuple[bool, str]:
     if not path.is_file():
         return False, f"Script path is not a file: {path}"
 
+    script_timeout = _get_script_timeout()
+
     try:
         result = subprocess.run(
             [sys.executable, str(path)],
             capture_output=True,
             text=True,
-            timeout=_SCRIPT_TIMEOUT,
+            timeout=script_timeout,
             cwd=str(path.parent),
         )
         stdout = (result.stdout or "").strip()
@@ -422,7 +460,7 @@ def _run_job_script(script_path: str) -> tuple[bool, str]:
         return True, stdout
 
     except subprocess.TimeoutExpired:
-        return False, f"Script timed out after {_SCRIPT_TIMEOUT}s: {path}"
+        return False, f"Script timed out after {script_timeout}s: {path}"
     except Exception as exc:
         return False, f"Script execution failed: {exc}"
 
@@ -646,6 +684,24 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
             },
         )
 
+        fallback_model = _cfg.get("fallback_providers") or _cfg.get("fallback_model") or None
+        credential_pool = None
+        runtime_provider = str(turn_route["runtime"].get("provider") or "").strip().lower()
+        if runtime_provider:
+            try:
+                from agent.credential_pool import load_pool
+                pool = load_pool(runtime_provider)
+                if pool.has_credentials():
+                    credential_pool = pool
+                    logger.info(
+                        "Job '%s': loaded credential pool for provider %s with %d entries",
+                        job_id,
+                        runtime_provider,
+                        len(pool.entries()),
+                    )
+            except Exception as e:
+                logger.debug("Job '%s': failed to load credential pool for %s: %s", job_id, runtime_provider, e)
+
         agent = AIAgent(
             model=turn_route["model"],
             api_key=turn_route["runtime"].get("api_key"),
@@ -657,6 +713,8 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
             max_iterations=max_iterations,
             reasoning_config=reasoning_config,
             prefill_messages=prefill_messages,
+            fallback_model=fallback_model,
+            credential_pool=credential_pool,
             providers_allowed=pr.get("only"),
             providers_ignored=pr.get("ignore"),
             providers_order=pr.get("order"),
@@ -711,7 +769,7 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
             _cron_pool.shutdown(wait=False, cancel_futures=True)
             raise
         finally:
-            _cron_pool.shutdown(wait=False)
+            _cron_pool.shutdown(wait=False, cancel_futures=True)
 
         if _inactivity_timeout:
             # Build diagnostic summary from the agent's activity tracker.

docker/entrypoint.sh

@@ -9,7 +9,10 @@ INSTALL_DIR="/opt/hermes"
 # (cache/images, cache/audio, platforms/whatsapp, etc.) are created on
 # demand by the application — don't pre-create them here so new installs
 # get the consolidated layout from get_hermes_dir().
-mkdir -p "$HERMES_HOME"/{cron,sessions,logs,hooks,memories,skills}
+# The "home/" subdirectory is a per-profile HOME for subprocesses (git,
+# ssh, gh, npm …).  Without it those tools write to /root which is
+# ephemeral and shared across profiles.  See issue #4426.
+mkdir -p "$HERMES_HOME"/{cron,sessions,logs,hooks,memories,skills,skins,plans,workspace,home}
 
 # .env
 if [ ! -f "$HERMES_HOME/.env" ]; then

gateway/channel_directory.py

@@ -76,10 +76,15 @@ def build_channel_directory(adapters: Dict[Any, Any]) -> Dict[str, Any]:
         except Exception as e:
             logger.warning("Channel directory: failed to build %s: %s", platform.value, e)
 
-    # Telegram, WhatsApp & Signal can't enumerate chats -- pull from session history
-    for plat_name in ("telegram", "whatsapp", "signal", "email", "sms", "bluebubbles"):
-        if plat_name not in platforms:
-            platforms[plat_name] = _build_from_sessions(plat_name)
+    # Platforms that don't support direct channel enumeration get session-based
+    # discovery automatically.  Skip infrastructure entries that aren't messaging
+    # platforms — everything else falls through to _build_from_sessions().
+    _SKIP_SESSION_DISCOVERY = frozenset({"local", "api_server", "webhook"})
+    for plat in Platform:
+        plat_name = plat.value
+        if plat_name in _SKIP_SESSION_DISCOVERY or plat_name in platforms:
+            continue
+        platforms[plat_name] = _build_from_sessions(plat_name)
 
     directory = {
         "updated_at": datetime.now().isoformat(),

gateway/config.py

@@ -63,6 +63,7 @@ class Platform(Enum):
     WEBHOOK = "webhook"
     FEISHU = "feishu"
     WECOM = "wecom"
+    WEIXIN = "weixin"
     BLUEBUBBLES = "bluebubbles"
 
 
@@ -261,6 +262,11 @@ class GatewayConfig:
         for platform, config in self.platforms.items():
             if not config.enabled:
                 continue
+            # Weixin requires both a token and an account_id
+            if platform == Platform.WEIXIN:
+                if config.extra.get("account_id") and (config.token or config.extra.get("token")):
+                    connected.append(platform)
+                continue
             # Platforms that use token/api_key auth
             if config.token or config.api_key:
                 connected.append(platform)
@@ -536,6 +542,8 @@ def load_gateway_config() -> GatewayConfig:
                     bridged["free_response_channels"] = platform_cfg["free_response_channels"]
                 if "mention_patterns" in platform_cfg:
                     bridged["mention_patterns"] = platform_cfg["mention_patterns"]
+                if plat == Platform.DISCORD and "channel_skill_bindings" in platform_cfg:
+                    bridged["channel_skill_bindings"] = platform_cfg["channel_skill_bindings"]
                 if not bridged:
                     continue
                 plat_data = platforms_data.setdefault(plat.value, {})
@@ -581,6 +589,12 @@ def load_gateway_config() -> GatewayConfig:
                     if isinstance(ic, list):
                         ic = ",".join(str(v) for v in ic)
                     os.environ["DISCORD_IGNORED_CHANNELS"] = str(ic)
+                # allowed_channels: if set, bot ONLY responds in these channels (whitelist)
+                ac = discord_cfg.get("allowed_channels")
+                if ac is not None and not os.getenv("DISCORD_ALLOWED_CHANNELS"):
+                    if isinstance(ac, list):
+                        ac = ",".join(str(v) for v in ac)
+                    os.environ["DISCORD_ALLOWED_CHANNELS"] = str(ac)
                 # no_thread_channels: channels where bot responds directly without creating thread
                 ntc = discord_cfg.get("no_thread_channels")
                 if ntc is not None and not os.getenv("DISCORD_NO_THREAD_CHANNELS"):
@@ -628,6 +642,8 @@ def load_gateway_config() -> GatewayConfig:
                     os.environ["MATRIX_FREE_RESPONSE_ROOMS"] = str(frc)
                 if "auto_thread" in matrix_cfg and not os.getenv("MATRIX_AUTO_THREAD"):
                     os.environ["MATRIX_AUTO_THREAD"] = str(matrix_cfg["auto_thread"]).lower()
+                if "dm_mention_threads" in matrix_cfg and not os.getenv("MATRIX_DM_MENTION_THREADS"):
+                    os.environ["MATRIX_DM_MENTION_THREADS"] = str(matrix_cfg["dm_mention_threads"]).lower()
 
     except Exception as e:
         logger.warning(
@@ -666,6 +682,7 @@ def load_gateway_config() -> GatewayConfig:
         Platform.SLACK: "SLACK_BOT_TOKEN",
         Platform.MATTERMOST: "MATTERMOST_TOKEN",
         Platform.MATRIX: "MATRIX_ACCESS_TOKEN",
+        Platform.WEIXIN: "WEIXIN_TOKEN",
     }
     for platform, pconfig in config.platforms.items():
         if not pconfig.enabled:
@@ -901,6 +918,9 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
                 pass
         if api_server_host:
             config.platforms[Platform.API_SERVER].extra["host"] = api_server_host
+        api_server_model_name = os.getenv("API_SERVER_MODEL_NAME", "")
+        if api_server_model_name:
+            config.platforms[Platform.API_SERVER].extra["model_name"] = api_server_model_name
 
     # Webhook platform
     webhook_enabled = os.getenv("WEBHOOK_ENABLED", "").lower() in ("true", "1", "yes")
@@ -967,6 +987,44 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
                 name=os.getenv("WECOM_HOME_CHANNEL_NAME", "Home"),
             )
 
+    # Weixin (personal WeChat via iLink Bot API)
+    weixin_token = os.getenv("WEIXIN_TOKEN")
+    weixin_account_id = os.getenv("WEIXIN_ACCOUNT_ID")
+    if weixin_token or weixin_account_id:
+        if Platform.WEIXIN not in config.platforms:
+            config.platforms[Platform.WEIXIN] = PlatformConfig()
+        config.platforms[Platform.WEIXIN].enabled = True
+        if weixin_token:
+            config.platforms[Platform.WEIXIN].token = weixin_token
+        extra = config.platforms[Platform.WEIXIN].extra
+        if weixin_account_id:
+            extra["account_id"] = weixin_account_id
+        weixin_base_url = os.getenv("WEIXIN_BASE_URL", "").strip()
+        if weixin_base_url:
+            extra["base_url"] = weixin_base_url.rstrip("/")
+        weixin_cdn_base_url = os.getenv("WEIXIN_CDN_BASE_URL", "").strip()
+        if weixin_cdn_base_url:
+            extra["cdn_base_url"] = weixin_cdn_base_url.rstrip("/")
+        weixin_dm_policy = os.getenv("WEIXIN_DM_POLICY", "").strip().lower()
+        if weixin_dm_policy:
+            extra["dm_policy"] = weixin_dm_policy
+        weixin_group_policy = os.getenv("WEIXIN_GROUP_POLICY", "").strip().lower()
+        if weixin_group_policy:
+            extra["group_policy"] = weixin_group_policy
+        weixin_allowed_users = os.getenv("WEIXIN_ALLOWED_USERS", "").strip()
+        if weixin_allowed_users:
+            extra["allow_from"] = weixin_allowed_users
+        weixin_group_allowed_users = os.getenv("WEIXIN_GROUP_ALLOWED_USERS", "").strip()
+        if weixin_group_allowed_users:
+            extra["group_allow_from"] = weixin_group_allowed_users
+        weixin_home = os.getenv("WEIXIN_HOME_CHANNEL", "").strip()
+        if weixin_home:
+            config.platforms[Platform.WEIXIN].home_channel = HomeChannel(
+                platform=Platform.WEIXIN,
+                chat_id=weixin_home,
+                name=os.getenv("WEIXIN_HOME_CHANNEL_NAME", "Home"),
+            )
+
     # BlueBubbles (iMessage)
     bluebubbles_server_url = os.getenv("BLUEBUBBLES_SERVER_URL")
     bluebubbles_password = os.getenv("BLUEBUBBLES_PASSWORD")

gateway/platforms/api_server.py

@@ -20,10 +20,13 @@ Requires:
 """
 
 import asyncio
+import hashlib
 import hmac
 import json
 import logging
 import os
+import socket as _socket
+import re
 import sqlite3
 import time
 import uuid
@@ -40,6 +43,7 @@ from gateway.config import Platform, PlatformConfig
 from gateway.platforms.base import (
     BasePlatformAdapter,
     SendResult,
+    is_network_accessible,
 )
 
 logger = logging.getLogger(__name__)
@@ -282,6 +286,24 @@ def _make_request_fingerprint(body: Dict[str, Any], keys: List[str]) -> str:
     return sha256(repr(subset).encode("utf-8")).hexdigest()
 
 
+def _derive_chat_session_id(
+    system_prompt: Optional[str],
+    first_user_message: str,
+) -> str:
+    """Derive a stable session ID from the conversation's first user message.
+
+    OpenAI-compatible frontends (Open WebUI, LibreChat, etc.) send the full
+    conversation history with every request.  The system prompt and first user
+    message are constant across all turns of the same conversation, so hashing
+    them produces a deterministic session ID that lets the API server reuse
+    the same Hermes session (and therefore the same Docker container sandbox
+    directory) across turns.
+    """
+    seed = f"{system_prompt or ''}\n{first_user_message}"
+    digest = hashlib.sha256(seed.encode("utf-8")).hexdigest()[:16]
+    return f"api-{digest}"
+
+
 class APIServerAdapter(BasePlatformAdapter):
     """
     OpenAI-compatible HTTP API server adapter.
@@ -299,6 +321,9 @@ class APIServerAdapter(BasePlatformAdapter):
         self._cors_origins: tuple[str, ...] = self._parse_cors_origins(
             extra.get("cors_origins", os.getenv("API_SERVER_CORS_ORIGINS", "")),
         )
+        self._model_name: str = self._resolve_model_name(
+            extra.get("model_name", os.getenv("API_SERVER_MODEL_NAME", "")),
+        )
         self._app: Optional["web.Application"] = None
         self._runner: Optional["web.AppRunner"] = None
         self._site: Optional["web.TCPSite"] = None
@@ -324,6 +349,26 @@ class APIServerAdapter(BasePlatformAdapter):
 
         return tuple(str(item).strip() for item in items if str(item).strip())
 
+    @staticmethod
+    def _resolve_model_name(explicit: str) -> str:
+        """Derive the advertised model name for /v1/models.
+
+        Priority:
+        1. Explicit override (config extra or API_SERVER_MODEL_NAME env var)
+        2. Active profile name (so each profile advertises a distinct model)
+        3. Fallback: "hermes-agent"
+        """
+        if explicit and explicit.strip():
+            return explicit.strip()
+        try:
+            from hermes_cli.profiles import get_active_profile_name
+            profile = get_active_profile_name()
+            if profile and profile not in ("default", "custom"):
+                return profile
+        except Exception:
+            pass
+        return "hermes-agent"
+
     def _cors_headers_for_origin(self, origin: str) -> Optional[Dict[str, str]]:
         """Return CORS headers for an allowed browser origin."""
         if not origin or not self._cors_origins:
@@ -363,7 +408,8 @@ class APIServerAdapter(BasePlatformAdapter):
         Validate Bearer token from Authorization header.
 
         Returns None if auth is OK, or a 401 web.Response on failure.
-        If no API key is configured, all requests are allowed.
+        If no API key is configured, all requests are allowed (only when API
+        server is local).
         """
         if not self._api_key:
             return None  # No key configured — allow all (local-only use)
@@ -468,12 +514,12 @@ class APIServerAdapter(BasePlatformAdapter):
             "object": "list",
             "data": [
                 {
-                    "id": "hermes-agent",
+                    "id": self._model_name,
                     "object": "model",
                     "created": int(time.time()),
                     "owned_by": "hermes",
                     "permission": [],
-                    "root": "hermes-agent",
+                    "root": self._model_name,
                     "parent": None,
                 }
             ],
@@ -531,8 +577,32 @@ class APIServerAdapter(BasePlatformAdapter):
 
         # Allow caller to continue an existing session by passing X-Hermes-Session-Id.
         # When provided, history is loaded from state.db instead of from the request body.
+        #
+        # Security: session continuation exposes conversation history, so it is
+        # only allowed when the API key is configured and the request is
+        # authenticated.  Without this gate, any unauthenticated client could
+        # read arbitrary session history by guessing/enumerating session IDs.
         provided_session_id = request.headers.get("X-Hermes-Session-Id", "").strip()
         if provided_session_id:
+            if not self._api_key:
+                logger.warning(
+                    "Session continuation via X-Hermes-Session-Id rejected: "
+                    "no API key configured.  Set API_SERVER_KEY to enable "
+                    "session continuity."
+                )
+                return web.json_response(
+                    _openai_error(
+                        "Session continuation requires API key authentication. "
+                        "Configure API_SERVER_KEY to enable this feature."
+                    ),
+                    status=403,
+                )
+            # Sanitize: reject control characters that could enable header injection.
+            if re.search(r'[\r\n\x00]', provided_session_id):
+                return web.json_response(
+                    {"error": {"message": "Invalid session ID", "type": "invalid_request_error"}},
+                    status=400,
+                )
             session_id = provided_session_id
             try:
                 db = self._ensure_session_db()
@@ -542,11 +612,20 @@ class APIServerAdapter(BasePlatformAdapter):
                 logger.warning("Failed to load session history for %s: %s", session_id, e)
                 history = []
         else:
-            session_id = str(uuid.uuid4())
+            # Derive a stable session ID from the conversation fingerprint so
+            # that consecutive messages from the same Open WebUI (or similar)
+            # conversation map to the same Hermes session.  The first user
+            # message + system prompt are constant across all turns.
+            first_user = ""
+            for cm in conversation_messages:
+                if cm.get("role") == "user":
+                    first_user = cm.get("content", "")
+                    break
+            session_id = _derive_chat_session_id(system_prompt, first_user)
             # history already set from request body above
 
         completion_id = f"chatcmpl-{uuid.uuid4().hex[:29]}"
-        model_name = body.get("model", "hermes-agent")
+        model_name = body.get("model", self._model_name)
         created = int(time.time())
 
         if stream:
@@ -565,15 +644,35 @@ class APIServerAdapter(BasePlatformAdapter):
                     _stream_q.put(delta)
 
             def _on_tool_progress(event_type, name, preview, args, **kwargs):
-                """Inject tool progress into the SSE stream for Open WebUI."""
+                """Send tool progress as a separate SSE event.
+
+                Previously, progress markers like ``⏰ list`` were injected
+                directly into ``delta.content``.  OpenAI-compatible frontends
+                (Open WebUI, LobeChat, …) store ``delta.content`` verbatim as
+                the assistant message and send it back on subsequent requests.
+                After enough turns the model learns to *emit* the markers as
+                plain text instead of issuing real tool calls — silently
+                hallucinating tool results.  See #6972.
+
+                The fix: push a tagged tuple ``("__tool_progress__", payload)``
+                onto the stream queue.  The SSE writer emits it as a custom
+                ``event: hermes.tool.progress`` line that compliant frontends
+                can render for UX but will *not* persist into conversation
+                history.  Clients that don't understand the custom event type
+                silently ignore it per the SSE specification.
+                """
                 if event_type != "tool.started":
-                    return  # Only show tool start events in chat stream
+                    return
                 if name.startswith("_"):
-                    return  # Skip internal events (_thinking)
+                    return
                 from agent.display import get_tool_emoji
                 emoji = get_tool_emoji(name)
                 label = preview or name
-                _stream_q.put(f"\n`{emoji} {label}`\n")
+                _stream_q.put(("__tool_progress__", {
+                    "tool": name,
+                    "emoji": emoji,
+                    "label": label,
+                }))
 
             # Start agent in background.  agent_ref is a mutable container
             # so the SSE writer can interrupt the agent on client disconnect.
@@ -684,6 +783,29 @@ class APIServerAdapter(BasePlatformAdapter):
             }
             await response.write(f"data: {json.dumps(role_chunk)}\n\n".encode())
 
+            # Helper — route a queue item to the correct SSE event.
+            async def _emit(item):
+                """Write a single queue item to the SSE stream.
+
+                Plain strings are sent as normal ``delta.content`` chunks.
+                Tagged tuples ``("__tool_progress__", payload)`` are sent
+                as a custom ``event: hermes.tool.progress`` SSE event so
+                frontends can display them without storing the markers in
+                conversation history.  See #6972.
+                """
+                if isinstance(item, tuple) and len(item) == 2 and item[0] == "__tool_progress__":
+                    event_data = json.dumps(item[1])
+                    await response.write(
+                        f"event: hermes.tool.progress\ndata: {event_data}\n\n".encode()
+                    )
+                else:
+                    content_chunk = {
+                        "id": completion_id, "object": "chat.completion.chunk",
+                        "created": created, "model": model,
+                        "choices": [{"index": 0, "delta": {"content": item}, "finish_reason": None}],
+                    }
+                    await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+
             # Stream content chunks as they arrive from the agent
             loop = asyncio.get_event_loop()
             while True:
@@ -697,12 +819,7 @@ class APIServerAdapter(BasePlatformAdapter):
                                 delta = stream_q.get_nowait()
                                 if delta is None:
                                     break
-                                content_chunk = {
-                                    "id": completion_id, "object": "chat.completion.chunk",
-                                    "created": created, "model": model,
-                                    "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
-                                }
-                                await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+                                await _emit(delta)
                             except _q.Empty:
                                 break
                         break
@@ -711,12 +828,7 @@ class APIServerAdapter(BasePlatformAdapter):
                 if delta is None:  # End of stream sentinel
                     break
 
-                content_chunk = {
-                    "id": completion_id, "object": "chat.completion.chunk",
-                    "created": created, "model": model,
-                    "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
-                }
-                await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+                await _emit(delta)
 
             # Get usage from completed agent
             usage = {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}
@@ -923,7 +1035,7 @@ class APIServerAdapter(BasePlatformAdapter):
             "object": "response",
             "status": "completed",
             "created_at": created_at,
-            "model": body.get("model", "hermes-agent"),
+            "model": body.get("model", self._model_name),
             "output": output_items,
             "usage": {
                 "input_tokens": usage.get("input_tokens", 0),
@@ -1318,6 +1430,7 @@ class APIServerAdapter(BasePlatformAdapter):
             result = agent.run_conversation(
                 user_message=user_message,
                 conversation_history=conversation_history,
+                task_id="default",
             )
             usage = {
                 "input_tokens": getattr(agent, "session_prompt_tokens", 0) or 0,
@@ -1484,6 +1597,7 @@ class APIServerAdapter(BasePlatformAdapter):
                     r = agent.run_conversation(
                         user_message=user_message,
                         conversation_history=conversation_history,
+                        task_id="default",
                     )
                     u = {
                         "input_tokens": getattr(agent, "session_prompt_tokens", 0) or 0,
@@ -1635,8 +1749,16 @@ class APIServerAdapter(BasePlatformAdapter):
             if hasattr(sweep_task, "add_done_callback"):
                 sweep_task.add_done_callback(self._background_tasks.discard)
 
+            # Refuse to start network-accessible without authentication
+            if is_network_accessible(self._host) and not self._api_key:
+                logger.error(
+                    "[%s] Refusing to start: binding to %s requires API_SERVER_KEY. "
+                    "Set API_SERVER_KEY or use the default 127.0.0.1.",
+                    self.name, self._host,
+                )
+                return False
+
             # Port conflict detection — fail fast if port is already in use
-            import socket as _socket
             try:
                 with _socket.socket(_socket.AF_INET, _socket.SOCK_STREAM) as _s:
                     _s.settimeout(1)
@@ -1652,9 +1774,17 @@ class APIServerAdapter(BasePlatformAdapter):
             await self._site.start()
 
             self._mark_connected()
+            if not self._api_key:
+                logger.warning(
+                    "[%s] ⚠️  No API key configured (API_SERVER_KEY / platforms.api_server.key). "
+                    "All requests will be accepted without authentication. "
+                    "Set an API key for production deployments to prevent "
+                    "unauthorized access to sessions, responses, and cron jobs.",
+                    self.name,
+                )
             logger.info(
-                "[%s] API server listening on http://%s:%d",
-                self.name, self._host, self._port,
+                "[%s] API server listening on http://%s:%d (model: %s)",
+                self.name, self._host, self._port, self._model_name,
             )
             return True
 

gateway/platforms/base.py

@@ -6,10 +6,12 @@ and implement the required methods.
 """
 
 import asyncio
+import ipaddress
 import logging
 import os
 import random
 import re
+import socket as _socket
 import subprocess
 import sys
 import uuid
@@ -19,6 +21,41 @@ from urllib.parse import urlsplit
 logger = logging.getLogger(__name__)
 
 
+def is_network_accessible(host: str) -> bool:
+    """Return True if *host* would expose the server beyond loopback.
+
+    Loopback addresses (127.0.0.1, ::1, IPv4-mapped ::ffff:127.0.0.1)
+    are local-only.  Unspecified addresses (0.0.0.0, ::) bind all
+    interfaces.  Hostnames are resolved; DNS failure fails closed.
+    """
+    try:
+        addr = ipaddress.ip_address(host)
+        if addr.is_loopback:
+            return False
+        # ::ffff:127.0.0.1 — Python reports is_loopback=False for mapped
+        # addresses, so check the underlying IPv4 explicitly.
+        if getattr(addr, "ipv4_mapped", None) and addr.ipv4_mapped.is_loopback:
+            return False
+        return True
+    except ValueError:
+        # when host variable is a hostname, we should try to resolve below
+        pass
+
+    try:
+        resolved = _socket.getaddrinfo(
+            host, None, _socket.AF_UNSPEC, _socket.SOCK_STREAM,
+        )
+        # if the hostname resolves into at least one non-loopback address,
+        # then we consider it to be network accessible
+        for _family, _type, _proto, _canonname, sockaddr in resolved:
+            addr = ipaddress.ip_address(sockaddr[0])
+            if not addr.is_loopback:
+                return True
+        return False
+    except (_socket.gaierror, OSError):
+        return True
+
+
 def _detect_macos_system_proxy() -> str | None:
     """Read the macOS system HTTP(S) proxy via ``scutil --proxy``.
 
@@ -160,7 +197,7 @@ GATEWAY_SECRET_CAPTURE_UNSUPPORTED_MESSAGE = (
 )
 
 
-def _safe_url_for_log(url: str, max_len: int = 80) -> str:
+def safe_url_for_log(url: str, max_len: int = 80) -> str:
     """Return a URL string safe for logs (no query/fragment/userinfo)."""
     if max_len <= 0:
         return ""
@@ -197,6 +234,23 @@ def _safe_url_for_log(url: str, max_len: int = 80) -> str:
     return f"{safe[:max_len - 3]}..."
 
 
+async def _ssrf_redirect_guard(response):
+    """Re-validate each redirect target to prevent redirect-based SSRF.
+
+    Without this, an attacker can host a public URL that 302-redirects to
+    http://169.254.169.254/ and bypass the pre-flight is_safe_url() check.
+
+    Must be async because httpx.AsyncClient awaits response event hooks.
+    """
+    if response.is_redirect and response.next_request:
+        redirect_url = str(response.next_request.url)
+        from tools.url_safety import is_safe_url
+        if not is_safe_url(redirect_url):
+            raise ValueError(
+                f"Blocked redirect to private/internal address: {safe_url_for_log(redirect_url)}"
+            )
+
+
 # ---------------------------------------------------------------------------
 # Image cache utilities
 #
@@ -216,6 +270,23 @@ def get_image_cache_dir() -> Path:
     return IMAGE_CACHE_DIR
 
 
+def _looks_like_image(data: bytes) -> bool:
+    """Return True if *data* starts with a known image magic-byte sequence."""
+    if len(data) < 4:
+        return False
+    if data[:8] == b"\x89PNG\r\n\x1a\n":
+        return True
+    if data[:3] == b"\xff\xd8\xff":
+        return True
+    if data[:6] in (b"GIF87a", b"GIF89a"):
+        return True
+    if data[:2] == b"BM":
+        return True
+    if data[:4] == b"RIFF" and len(data) >= 12 and data[8:12] == b"WEBP":
+        return True
+    return False
+
+
 def cache_image_from_bytes(data: bytes, ext: str = ".jpg") -> str:
     """
     Save raw image bytes to the cache and return the absolute file path.
@@ -226,7 +297,17 @@ def cache_image_from_bytes(data: bytes, ext: str = ".jpg") -> str:
 
     Returns:
         Absolute path to the cached image file as a string.
+
+    Raises:
+        ValueError: If *data* does not look like a valid image (e.g. an HTML
+            error page returned by the upstream server).
     """
+    if not _looks_like_image(data):
+        snippet = data[:80].decode("utf-8", errors="replace")
+        raise ValueError(
+            f"Refusing to cache non-image data as {ext} "
+            f"(starts with: {snippet!r})"
+        )
     cache_dir = get_image_cache_dir()
     filename = f"img_{uuid.uuid4().hex[:12]}{ext}"
     filepath = cache_dir / filename
@@ -254,7 +335,7 @@ async def cache_image_from_url(url: str, ext: str = ".jpg", retries: int = 2) ->
     """
     from tools.url_safety import is_safe_url
     if not is_safe_url(url):
-        raise ValueError(f"Blocked unsafe URL (SSRF protection): {_safe_url_for_log(url)}")
+        raise ValueError(f"Blocked unsafe URL (SSRF protection): {safe_url_for_log(url)}")
 
     import asyncio
     import httpx
@@ -262,7 +343,11 @@ async def cache_image_from_url(url: str, ext: str = ".jpg", retries: int = 2) ->
     _log = _logging.getLogger(__name__)
 
     last_exc = None
-    async with httpx.AsyncClient(timeout=30.0, follow_redirects=True) as client:
+    async with httpx.AsyncClient(
+        timeout=30.0,
+        follow_redirects=True,
+        event_hooks={"response": [_ssrf_redirect_guard]},
+    ) as client:
         for attempt in range(retries + 1):
             try:
                 response = await client.get(
@@ -284,7 +369,7 @@ async def cache_image_from_url(url: str, ext: str = ".jpg", retries: int = 2) ->
                         "Media cache retry %d/%d for %s (%.1fs): %s",
                         attempt + 1,
                         retries,
-                        _safe_url_for_log(url),
+                        safe_url_for_log(url),
                         wait,
                         exc,
                     )
@@ -369,7 +454,7 @@ async def cache_audio_from_url(url: str, ext: str = ".ogg", retries: int = 2) ->
     """
     from tools.url_safety import is_safe_url
     if not is_safe_url(url):
-        raise ValueError(f"Blocked unsafe URL (SSRF protection): {_safe_url_for_log(url)}")
+        raise ValueError(f"Blocked unsafe URL (SSRF protection): {safe_url_for_log(url)}")
 
     import asyncio
     import httpx
@@ -377,7 +462,11 @@ async def cache_audio_from_url(url: str, ext: str = ".ogg", retries: int = 2) ->
     _log = _logging.getLogger(__name__)
 
     last_exc = None
-    async with httpx.AsyncClient(timeout=30.0, follow_redirects=True) as client:
+    async with httpx.AsyncClient(
+        timeout=30.0,
+        follow_redirects=True,
+        event_hooks={"response": [_ssrf_redirect_guard]},
+    ) as client:
         for attempt in range(retries + 1):
             try:
                 response = await client.get(
@@ -399,7 +488,7 @@ async def cache_audio_from_url(url: str, ext: str = ".ogg", retries: int = 2) ->
                         "Audio cache retry %d/%d for %s (%.1fs): %s",
                         attempt + 1,
                         retries,
-                        _safe_url_for_log(url),
+                        safe_url_for_log(url),
                         wait,
                         exc,
                     )
@@ -502,6 +591,14 @@ class MessageType(Enum):
     COMMAND = "command"  # /command style
 
 
+class ProcessingOutcome(Enum):
+    """Result classification for message-processing lifecycle hooks."""
+
+    SUCCESS = "success"
+    FAILURE = "failure"
+    CANCELLED = "cancelled"
+
+
 @dataclass
 class MessageEvent:
     """
@@ -529,8 +626,9 @@ class MessageEvent:
     reply_to_message_id: Optional[str] = None
     reply_to_text: Optional[str] = None  # Text of the replied-to message (for context injection)
     
-    # Auto-loaded skill for topic/channel bindings (e.g., Telegram DM Topics)
-    auto_skill: Optional[str] = None
+    # Auto-loaded skill(s) for topic/channel bindings (e.g., Telegram DM Topics,
+    # Discord channel_skill_bindings).  A single name or ordered list.
+    auto_skill: Optional[str | list[str]] = None
     
     # Internal flag — set for synthetic events (e.g. background process
     # completion notifications) that must bypass user authorization checks.
@@ -552,6 +650,9 @@ class MessageEvent:
         raw = parts[0][1:].lower() if parts else None
         if raw and "@" in raw:
             raw = raw.split("@", 1)[0]
+        # Reject file paths: valid command names never contain /
+        if raw and "/" in raw:
+            return None
         return raw
     
     def get_command_args(self) -> str:
@@ -625,6 +726,7 @@ class BasePlatformAdapter(ABC):
         # Gateway shutdown cancels these so an old gateway instance doesn't keep
         # working on a task after --replace or manual restarts.
         self._background_tasks: set[asyncio.Task] = set()
+        self._expected_cancelled_tasks: set[asyncio.Task] = set()
         # Chats where auto-TTS on voice input is disabled (set by /voice off)
         self._auto_tts_disabled_chats: set = set()
         # Chats where typing indicator is paused (e.g. during approval waits).
@@ -1133,7 +1235,7 @@ class BasePlatformAdapter(ABC):
     async def on_processing_start(self, event: MessageEvent) -> None:
         """Hook called when background processing begins."""
 
-    async def on_processing_complete(self, event: MessageEvent, success: bool) -> None:
+    async def on_processing_complete(self, event: MessageEvent, outcome: ProcessingOutcome) -> None:
         """Hook called when background processing completes."""
 
     async def _run_processing_hook(self, hook_name: str, *args: Any, **kwargs: Any) -> None:
@@ -1294,7 +1396,7 @@ class BasePlatformAdapter(ABC):
             # session lifecycle and its cleanup races with the running task
             # (see PR #4926).
             cmd = event.get_command()
-            if cmd in ("approve", "deny", "status", "stop", "new", "reset"):
+            if cmd in ("approve", "deny", "status", "stop", "new", "reset", "background"):
                 logger.debug(
                     "[%s] Command '/%s' bypassing active-session guard for %s",
                     self.name, cmd, session_key,
@@ -1352,6 +1454,7 @@ class BasePlatformAdapter(ABC):
             return
         if hasattr(task, "add_done_callback"):
             task.add_done_callback(self._background_tasks.discard)
+            task.add_done_callback(self._expected_cancelled_tasks.discard)
     
     @staticmethod
     def _get_human_delay() -> float:
@@ -1488,7 +1591,7 @@ class BasePlatformAdapter(ABC):
                         logger.info(
                             "[%s] Sending image: %s (alt=%s)",
                             self.name,
-                            _safe_url_for_log(image_url),
+                            safe_url_for_log(image_url),
                             alt_text[:30] if alt_text else "",
                         )
                         # Route animated GIFs through send_animation for proper playback
@@ -1580,7 +1683,11 @@ class BasePlatformAdapter(ABC):
 
             # Determine overall success for the processing hook
             processing_ok = delivery_succeeded if delivery_attempted else not bool(response)
-            await self._run_processing_hook("on_processing_complete", event, processing_ok)
+            await self._run_processing_hook(
+                "on_processing_complete",
+                event,
+                ProcessingOutcome.SUCCESS if processing_ok else ProcessingOutcome.FAILURE,
+            )
 
             # Check if there's a pending message that was queued during our processing
             if session_key in self._pending_messages:
@@ -1599,10 +1706,14 @@ class BasePlatformAdapter(ABC):
                 return  # Already cleaned up
                 
         except asyncio.CancelledError:
-            await self._run_processing_hook("on_processing_complete", event, False)
+            current_task = asyncio.current_task()
+            outcome = ProcessingOutcome.CANCELLED
+            if current_task is None or current_task not in self._expected_cancelled_tasks:
+                outcome = ProcessingOutcome.FAILURE
+            await self._run_processing_hook("on_processing_complete", event, outcome)
             raise
         except Exception as e:
-            await self._run_processing_hook("on_processing_complete", event, False)
+            await self._run_processing_hook("on_processing_complete", event, ProcessingOutcome.FAILURE)
             logger.error("[%s] Error handling message: %s", self.name, e, exc_info=True)
             # Send the error to the user so they aren't left with radio silence
             try:
@@ -1646,10 +1757,12 @@ class BasePlatformAdapter(ABC):
         """
         tasks = [task for task in self._background_tasks if not task.done()]
         for task in tasks:
+            self._expected_cancelled_tasks.add(task)
             task.cancel()
         if tasks:
             await asyncio.gather(*tasks, return_exceptions=True)
         self._background_tasks.clear()
+        self._expected_cancelled_tasks.clear()
         self._pending_messages.clear()
         self._active_sessions.clear()
 

gateway/platforms/bluebubbles.py

@@ -207,9 +207,17 @@ class BlueBubblesAdapter(BasePlatformAdapter):
             self.webhook_port,
             self.webhook_path,
         )
+
+        # Register webhook with BlueBubbles server
+        # This is required for the server to know where to send events
+        await self._register_webhook()
+
         return True
 
     async def disconnect(self) -> None:
+        # Unregister webhook before cleaning up
+        await self._unregister_webhook()
+
         if self.client:
             await self.client.aclose()
             self.client = None
@@ -218,6 +226,105 @@ class BlueBubblesAdapter(BasePlatformAdapter):
             self._runner = None
         self._mark_disconnected()
 
+    @property
+    def _webhook_url(self) -> str:
+        """Compute the external webhook URL for BlueBubbles registration."""
+        host = self.webhook_host
+        if host in ("0.0.0.0", "127.0.0.1", "localhost", "::"):
+            host = "localhost"
+        return f"http://{host}:{self.webhook_port}{self.webhook_path}"
+
+    async def _find_registered_webhooks(self, url: str) -> list:
+        """Return list of BB webhook entries matching *url*."""
+        try:
+            res = await self._api_get("/api/v1/webhook")
+            data = res.get("data")
+            if isinstance(data, list):
+                return [wh for wh in data if wh.get("url") == url]
+        except Exception:
+            pass
+        return []
+
+    async def _register_webhook(self) -> bool:
+        """Register this webhook URL with the BlueBubbles server.
+
+        BlueBubbles requires webhooks to be registered via API before
+        it will send events.  Checks for an existing registration first
+        to avoid duplicates (e.g. after a crash without clean shutdown).
+        """
+        if not self.client:
+            return False
+
+        webhook_url = self._webhook_url
+
+        # Crash resilience — reuse an existing registration if present
+        existing = await self._find_registered_webhooks(webhook_url)
+        if existing:
+            logger.info(
+                "[bluebubbles] webhook already registered: %s", webhook_url
+            )
+            return True
+
+        payload = {
+            "url": webhook_url,
+            "events": ["new-message", "updated-message", "message"],
+        }
+
+        try:
+            res = await self._api_post("/api/v1/webhook", payload)
+            status = res.get("status", 0)
+            if 200 <= status < 300:
+                logger.info(
+                    "[bluebubbles] webhook registered with server: %s",
+                    webhook_url,
+                )
+                return True
+            else:
+                logger.warning(
+                    "[bluebubbles] webhook registration returned status %s: %s",
+                    status,
+                    res.get("message"),
+                )
+                return False
+        except Exception as exc:
+            logger.warning(
+                "[bluebubbles] failed to register webhook with server: %s",
+                exc,
+            )
+            return False
+
+    async def _unregister_webhook(self) -> bool:
+        """Unregister this webhook URL from the BlueBubbles server.
+
+        Removes *all* matching registrations to clean up any duplicates
+        left by prior crashes.
+        """
+        if not self.client:
+            return False
+
+        webhook_url = self._webhook_url
+        removed = False
+
+        try:
+            for wh in await self._find_registered_webhooks(webhook_url):
+                wh_id = wh.get("id")
+                if wh_id:
+                    res = await self.client.delete(
+                        self._api_url(f"/api/v1/webhook/{wh_id}")
+                    )
+                    res.raise_for_status()
+                    removed = True
+            if removed:
+                logger.info(
+                    "[bluebubbles] webhook unregistered: %s", webhook_url
+                )
+        except Exception as exc:
+            logger.debug(
+                "[bluebubbles] failed to unregister webhook (non-critical): %s",
+                exc,
+            )
+        return removed
+
     # ------------------------------------------------------------------
     # Chat GUID resolution
     # ------------------------------------------------------------------
@@ -826,3 +933,4 @@ class BlueBubblesAdapter(BasePlatformAdapter):
             asyncio.create_task(self.mark_read(session_chat_id))
 
         return web.Response(text="ok")
+

gateway/platforms/dingtalk.py

@@ -20,6 +20,7 @@ Configuration in config.yaml:
 import asyncio
 import logging
 import os
+import re
 import time
 import uuid
 from datetime import datetime, timezone
@@ -54,6 +55,8 @@ MAX_MESSAGE_LENGTH = 20000
 DEDUP_WINDOW_SECONDS = 300
 DEDUP_MAX_SIZE = 1000
 RECONNECT_BACKOFF = [2, 5, 10, 30, 60]
+_SESSION_WEBHOOKS_MAX = 500
+_DINGTALK_WEBHOOK_RE = re.compile(r'^https://api\.dingtalk\.com/')
 
 
 def check_dingtalk_requirements() -> bool:
@@ -195,9 +198,15 @@ class DingTalkAdapter(BasePlatformAdapter):
         chat_id = conversation_id or sender_id
         chat_type = "group" if is_group else "dm"
 
-        # Store session webhook for reply routing
+        # Store session webhook for reply routing (validate origin to prevent SSRF)
         session_webhook = getattr(message, "session_webhook", None) or ""
-        if session_webhook and chat_id:
+        if session_webhook and chat_id and _DINGTALK_WEBHOOK_RE.match(session_webhook):
+            if len(self._session_webhooks) >= _SESSION_WEBHOOKS_MAX:
+                # Evict oldest entry to cap memory growth
+                try:
+                    self._session_webhooks.pop(next(iter(self._session_webhooks)))
+                except StopIteration:
+                    pass
             self._session_webhooks[chat_id] = session_webhook
 
         source = self.build_source(

gateway/platforms/discord.py

@@ -49,6 +49,7 @@ from gateway.platforms.base import (
     BasePlatformAdapter,
     MessageEvent,
     MessageType,
+    ProcessingOutcome,
     SendResult,
     cache_image_from_url,
     cache_audio_from_url,
@@ -422,6 +423,7 @@ class DiscordAdapter(BasePlatformAdapter):
 
     # Discord message limits
     MAX_MESSAGE_LENGTH = 2000
+    _SPLIT_THRESHOLD = 1900  # near the 2000-char split point
 
     # Auto-disconnect from voice channel after this many seconds of inactivity
     VOICE_TIMEOUT = 300
@@ -433,6 +435,11 @@ class DiscordAdapter(BasePlatformAdapter):
         self._allowed_user_ids: set = set()  # For button approval authorization
         # Voice channel state (per-guild)
         self._voice_clients: Dict[int, Any] = {}  # guild_id -> VoiceClient
+        # Text batching: merge rapid successive messages (Telegram-style)
+        self._text_batch_delay_seconds = float(os.getenv("HERMES_DISCORD_TEXT_BATCH_DELAY_SECONDS", "0.6"))
+        self._text_batch_split_delay_seconds = float(os.getenv("HERMES_DISCORD_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0"))
+        self._pending_text_batches: Dict[str, MessageEvent] = {}
+        self._pending_text_batch_tasks: Dict[str, asyncio.Task] = {}
         self._voice_text_channels: Dict[int, int] = {}  # guild_id -> text_channel_id
         self._voice_timeout_tasks: Dict[int, asyncio.Task] = {}  # guild_id -> timeout task
         # Phase 2: voice listening
@@ -599,22 +606,35 @@ class DiscordAdapter(BasePlatformAdapter):
                         if not self._client.user or self._client.user not in message.mentions:
                             return
                     # "all" falls through to handle_message
-
-                # If the message @mentions other users but NOT the bot, the
-                # sender is talking to someone else — stay silent.  Only
-                # applies in server channels; in DMs the user is always
-                # talking to the bot (mentions are just references).
-                # Controlled by DISCORD_IGNORE_NO_MENTION (default: true).
-                _ignore_no_mention = os.getenv(
-                    "DISCORD_IGNORE_NO_MENTION", "true"
-                ).lower() in ("true", "1", "yes")
-                if _ignore_no_mention and message.mentions and not isinstance(message.channel, discord.DMChannel):
-                    _bot_mentioned = (
+                
+                # Multi-agent filtering: if the message mentions specific bots
+                # but NOT this bot, the sender is talking to another agent —
+                # stay silent.  Messages with no bot mentions (general chat)
+                # still fall through to _handle_message for the existing
+                # DISCORD_REQUIRE_MENTION check.
+                #
+                # This replaces the older DISCORD_IGNORE_NO_MENTION logic
+                # with bot-aware filtering that works correctly when multiple
+                # agents share a channel.
+                if not isinstance(message.channel, discord.DMChannel) and message.mentions:
+                    _self_mentioned = (
                         self._client.user is not None
                         and self._client.user in message.mentions
                     )
-                    if not _bot_mentioned:
-                        return  # Talking to someone else, don't interrupt
+                    _other_bots_mentioned = any(
+                        m.bot and m != self._client.user
+                        for m in message.mentions
+                    )
+                    # If other bots are mentioned but we're not → not for us
+                    if _other_bots_mentioned and not _self_mentioned:
+                        return
+                    # If humans are mentioned but we're not → not for us
+                    # (preserves old DISCORD_IGNORE_NO_MENTION=true behavior)
+                    _ignore_no_mention = os.getenv(
+                        "DISCORD_IGNORE_NO_MENTION", "true"
+                    ).lower() in ("true", "1", "yes")
+                    if _ignore_no_mention and not _self_mentioned and not _other_bots_mentioned:
+                        return
 
                 await self._handle_message(message)
 
@@ -748,14 +768,17 @@ class DiscordAdapter(BasePlatformAdapter):
         if hasattr(message, "add_reaction"):
             await self._add_reaction(message, "👀")
 
-    async def on_processing_complete(self, event: MessageEvent, success: bool) -> None:
+    async def on_processing_complete(self, event: MessageEvent, outcome: ProcessingOutcome) -> None:
         """Swap the in-progress reaction for a final success/failure reaction."""
         if not self._reactions_enabled():
             return
         message = event.raw_message
         if hasattr(message, "add_reaction"):
             await self._remove_reaction(message, "👀")
-            await self._add_reaction(message, "✅" if success else "❌")
+            if outcome == ProcessingOutcome.SUCCESS:
+                await self._add_reaction(message, "✅")
+            elif outcome == ProcessingOutcome.FAILURE:
+                await self._add_reaction(message, "❌")
 
     async def send(
         self,
@@ -764,18 +787,34 @@ class DiscordAdapter(BasePlatformAdapter):
         reply_to: Optional[str] = None,
         metadata: Optional[Dict[str, Any]] = None
     ) -> SendResult:
-        """Send a message to a Discord channel."""
+        """Send a message to a Discord channel or thread.
+
+        When metadata contains a thread_id, the message is sent to that
+        thread instead of the parent channel identified by chat_id.
+        """
         if not self._client:
             return SendResult(success=False, error="Not connected")
 
         try:
-            # Get the channel
-            channel = self._client.get_channel(int(chat_id))
-            if not channel:
-                channel = await self._client.fetch_channel(int(chat_id))
+            # Determine target channel: thread_id in metadata takes precedence.
+            thread_id = None
+            if metadata and metadata.get("thread_id"):
+                thread_id = metadata["thread_id"]
 
-            if not channel:
-                return SendResult(success=False, error=f"Channel {chat_id} not found")
+            if thread_id:
+                # Fetch the thread directly — threads are addressed by their own ID.
+                channel = self._client.get_channel(int(thread_id))
+                if not channel:
+                    channel = await self._client.fetch_channel(int(thread_id))
+                if not channel:
+                    return SendResult(success=False, error=f"Thread {thread_id} not found")
+            else:
+                # Get the parent channel
+                channel = self._client.get_channel(int(chat_id))
+                if not channel:
+                    channel = await self._client.fetch_channel(int(chat_id))
+                if not channel:
+                    return SendResult(success=False, error=f"Channel {chat_id} not found")
 
             # Format and split message if needed
             formatted = self.format_message(content)
@@ -1238,9 +1277,8 @@ class DiscordAdapter(BasePlatformAdapter):
         try:
             await asyncio.to_thread(VoiceReceiver.pcm_to_wav, pcm_data, wav_path)
 
-            from tools.transcription_tools import transcribe_audio, get_stt_model_from_config
-            stt_model = get_stt_model_from_config()
-            result = await asyncio.to_thread(transcribe_audio, wav_path, model=stt_model)
+            from tools.transcription_tools import transcribe_audio
+            result = await asyncio.to_thread(transcribe_audio, wav_path)
 
             if not result.get("success"):
                 return
@@ -1867,14 +1905,42 @@ class DiscordAdapter(BasePlatformAdapter):
             chat_topic=chat_topic,
         )
 
+        _parent_id = str(getattr(getattr(interaction, "channel", None), "parent_id", "") or "")
+        _skills = self._resolve_channel_skills(thread_id, _parent_id or None)
         event = MessageEvent(
             text=text,
             message_type=MessageType.TEXT,
             source=source,
             raw_message=interaction,
+            auto_skill=_skills,
         )
         await self.handle_message(event)
 
+    def _resolve_channel_skills(self, channel_id: str, parent_id: str | None = None) -> list[str] | None:
+        """Look up auto-skill bindings for a Discord channel/forum thread.
+
+        Config format (in platform extra):
+            channel_skill_bindings:
+              - id: "123456"
+                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
+
     def _thread_parent_channel(self, channel: Any) -> Any:
         """Return the parent text channel when invoked from a thread."""
         return getattr(channel, "parent", None) or channel
@@ -2228,6 +2294,7 @@ class DiscordAdapter(BasePlatformAdapter):
         #   discord.require_mention: Require @mention in server channels (default: true)
         #   discord.free_response_channels: Channel IDs where bot responds without mention
         #   discord.ignored_channels: Channel IDs where bot NEVER responds (even when mentioned)
+        #   discord.allowed_channels: If set, bot ONLY responds in these channels (whitelist)
         #   discord.no_thread_channels: Channel IDs where bot responds directly without creating thread
         #   discord.auto_thread: Auto-create thread on @mention in channels (default: true)
 
@@ -2239,12 +2306,21 @@ class DiscordAdapter(BasePlatformAdapter):
             parent_channel_id = self._get_parent_channel_id(message.channel)
 
         if not isinstance(message.channel, discord.DMChannel):
-            # Check ignored channels first - never respond even when mentioned
-            ignored_channels_raw = os.getenv("DISCORD_IGNORED_CHANNELS", "")
-            ignored_channels = {ch.strip() for ch in ignored_channels_raw.split(",") if ch.strip()}
             channel_ids = {str(message.channel.id)}
             if parent_channel_id:
                 channel_ids.add(parent_channel_id)
+
+            # Check allowed channels - if set, only respond in these channels
+            allowed_channels_raw = os.getenv("DISCORD_ALLOWED_CHANNELS", "")
+            if allowed_channels_raw:
+                allowed_channels = {ch.strip() for ch in allowed_channels_raw.split(",") if ch.strip()}
+                if not (channel_ids & allowed_channels):
+                    logger.debug("[%s] Ignoring message in non-allowed channel: %s", self.name, channel_ids)
+                    return
+
+            # Check ignored channels - never respond even when mentioned
+            ignored_channels_raw = os.getenv("DISCORD_IGNORED_CHANNELS", "")
+            ignored_channels = {ch.strip() for ch in ignored_channels_raw.split(",") if ch.strip()}
             if channel_ids & ignored_channels:
                 logger.debug("[%s] Ignoring message in ignored channel: %s", self.name, channel_ids)
                 return
@@ -2449,6 +2525,10 @@ class DiscordAdapter(BasePlatformAdapter):
         if not event_text or not event_text.strip():
             event_text = "(The user sent a message with no text content)"
 
+        _chan = message.channel
+        _parent_id = str(getattr(_chan, "parent_id", "") or "")
+        _chan_id = str(getattr(_chan, "id", ""))
+        _skills = self._resolve_channel_skills(_chan_id, _parent_id or None)
         event = MessageEvent(
             text=event_text,
             message_type=msg_type,
@@ -2459,6 +2539,7 @@ class DiscordAdapter(BasePlatformAdapter):
             media_types=media_types,
             reply_to_message_id=str(message.reference.message_id) if message.reference else None,
             timestamp=message.created_at,
+            auto_skill=_skills,
         )
 
         # Track thread participation so the bot won't require @mention for
@@ -2466,7 +2547,80 @@ class DiscordAdapter(BasePlatformAdapter):
         if thread_id:
             self._track_thread(thread_id)
 
-        await self.handle_message(event)
+        # Only batch plain text messages — commands, media, etc. dispatch
+        # immediately since they won't be split by the Discord client.
+        if msg_type == MessageType.TEXT and self._text_batch_delay_seconds > 0:
+            self._enqueue_text_event(event)
+        else:
+            await self.handle_message(event)
+
+    # ------------------------------------------------------------------
+    # Text message aggregation (handles Discord client-side splits)
+    # ------------------------------------------------------------------
+
+    def _text_batch_key(self, event: MessageEvent) -> str:
+        """Session-scoped key for text message batching."""
+        from gateway.session import build_session_key
+        return build_session_key(
+            event.source,
+            group_sessions_per_user=self.config.extra.get("group_sessions_per_user", True),
+            thread_sessions_per_user=self.config.extra.get("thread_sessions_per_user", False),
+        )
+
+    def _enqueue_text_event(self, event: MessageEvent) -> None:
+        """Buffer a text event and reset the flush timer.
+
+        When Discord splits a long user message at 2000 chars, the chunks
+        arrive within a few hundred milliseconds.  This merges them into
+        a single event before dispatching.
+        """
+        key = self._text_batch_key(event)
+        existing = self._pending_text_batches.get(key)
+        chunk_len = len(event.text or "")
+        if existing is None:
+            event._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            self._pending_text_batches[key] = event
+        else:
+            if event.text:
+                existing.text = f"{existing.text}\n{event.text}" if existing.text else event.text
+            existing._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            if event.media_urls:
+                existing.media_urls.extend(event.media_urls)
+                existing.media_types.extend(event.media_types)
+
+        prior_task = self._pending_text_batch_tasks.get(key)
+        if prior_task and not prior_task.done():
+            prior_task.cancel()
+        self._pending_text_batch_tasks[key] = asyncio.create_task(
+            self._flush_text_batch(key)
+        )
+
+    async def _flush_text_batch(self, key: str) -> None:
+        """Wait for the quiet period then dispatch the aggregated text.
+
+        Uses a longer delay when the latest chunk is near Discord's 2000-char
+        split point, since a continuation chunk is almost certain.
+        """
+        current_task = asyncio.current_task()
+        try:
+            pending = self._pending_text_batches.get(key)
+            last_len = getattr(pending, "_last_chunk_len", 0) if pending else 0
+            if last_len >= self._SPLIT_THRESHOLD:
+                delay = self._text_batch_split_delay_seconds
+            else:
+                delay = self._text_batch_delay_seconds
+            await asyncio.sleep(delay)
+            event = self._pending_text_batches.pop(key, None)
+            if not event:
+                return
+            logger.info(
+                "[Discord] Flushing text batch %s (%d chars)",
+                key, len(event.text or ""),
+            )
+            await self.handle_message(event)
+        finally:
+            if self._pending_text_batch_tasks.get(key) is current_task:
+                self._pending_text_batch_tasks.pop(key, None)
 
 
 # ---------------------------------------------------------------------------

gateway/platforms/email.py

@@ -195,7 +195,11 @@ def _extract_attachments(
 
         ext = Path(filename).suffix.lower()
         if ext in _IMAGE_EXTS:
-            cached_path = cache_image_from_bytes(payload, ext)
+            try:
+                cached_path = cache_image_from_bytes(payload, ext)
+            except ValueError:
+                logger.debug("Skipping non-image attachment %s (invalid magic bytes)", filename)
+                continue
             attachments.append({
                 "path": cached_path,
                 "filename": filename,

gateway/platforms/feishu.py

@@ -264,6 +264,7 @@ class FeishuAdapterSettings:
     bot_name: str
     dedup_cache_size: int
     text_batch_delay_seconds: float
+    text_batch_split_delay_seconds: float
     text_batch_max_messages: int
     text_batch_max_chars: int
     media_batch_delay_seconds: float
@@ -972,7 +973,8 @@ def _run_official_feishu_ws_client(ws_client: Any, adapter: Any) -> None:
         return await original_connect(*args, **kwargs)
 
     def _configure_with_overrides(conf: Any) -> Any:
-        assert original_configure is not None
+        if original_configure is None:
+            raise RuntimeError("Feishu _configure_with_overrides called but original_configure is None")
         result = original_configure(conf)
         _apply_runtime_ws_overrides()
         return result
@@ -1014,6 +1016,10 @@ class FeishuAdapter(BasePlatformAdapter):
     """Feishu/Lark bot adapter."""
 
     MAX_MESSAGE_LENGTH = 8000
+    # Threshold for detecting Feishu client-side message splits.
+    # When a chunk is near the ~4096-char practical limit, a continuation
+    # is almost certain.
+    _SPLIT_THRESHOLD = 4000
 
     # =========================================================================
     # Lifecycle — init / settings / connect / disconnect
@@ -1105,6 +1111,9 @@ class FeishuAdapter(BasePlatformAdapter):
             text_batch_delay_seconds=float(
                 os.getenv("HERMES_FEISHU_TEXT_BATCH_DELAY_SECONDS", str(_DEFAULT_TEXT_BATCH_DELAY_SECONDS))
             ),
+            text_batch_split_delay_seconds=float(
+                os.getenv("HERMES_FEISHU_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0")
+            ),
             text_batch_max_messages=max(
                 1,
                 int(os.getenv("HERMES_FEISHU_TEXT_BATCH_MAX_MESSAGES", str(_DEFAULT_TEXT_BATCH_MAX_MESSAGES))),
@@ -1152,6 +1161,7 @@ class FeishuAdapter(BasePlatformAdapter):
         self._bot_name = settings.bot_name
         self._dedup_cache_size = settings.dedup_cache_size
         self._text_batch_delay_seconds = settings.text_batch_delay_seconds
+        self._text_batch_split_delay_seconds = settings.text_batch_split_delay_seconds
         self._text_batch_max_messages = settings.text_batch_max_messages
         self._text_batch_max_chars = settings.text_batch_max_chars
         self._media_batch_delay_seconds = settings.media_batch_delay_seconds
@@ -1180,6 +1190,8 @@ class FeishuAdapter(BasePlatformAdapter):
                 lambda data: self._on_reaction_event("im.message.reaction.deleted_v1", data)
             )
             .register_p2_card_action_trigger(self._on_card_action_trigger)
+            .register_p2_im_chat_member_bot_added_v1(self._on_bot_added_to_chat)
+            .register_p2_im_chat_member_bot_deleted_v1(self._on_bot_removed_from_chat)
             .build()
         )
 
@@ -1570,13 +1582,18 @@ class FeishuAdapter(BasePlatformAdapter):
             return SendResult(success=False, error=f"Image file not found: {image_path}")
 
         try:
-            with open(image_path, "rb") as image_file:
-                body = self._build_image_upload_body(
-                    image_type=_FEISHU_IMAGE_UPLOAD_TYPE,
-                    image=image_file,
-                )
-                request = self._build_image_upload_request(body)
-                upload_response = await asyncio.to_thread(self._client.im.v1.image.create, request)
+            import io as _io
+            with open(image_path, "rb") as f:
+                image_bytes = f.read()
+            # Wrap in BytesIO so lark SDK's MultipartEncoder can read .name and .tell()
+            image_file = _io.BytesIO(image_bytes)
+            image_file.name = os.path.basename(image_path)
+            body = self._build_image_upload_body(
+                image_type=_FEISHU_IMAGE_UPLOAD_TYPE,
+                image=image_file,
+            )
+            request = self._build_image_upload_request(body)
+            upload_response = await asyncio.to_thread(self._client.im.v1.image.create, request)
             image_key = self._extract_response_field(upload_response, "image_key")
             if not image_key:
                 return self._response_error_result(
@@ -2478,8 +2495,10 @@ class FeishuAdapter(BasePlatformAdapter):
     async def _enqueue_text_event(self, event: MessageEvent) -> None:
         """Debounce rapid Feishu text bursts into a single MessageEvent."""
         key = self._text_batch_key(event)
+        chunk_len = len(event.text or "")
         existing = self._pending_text_batches.get(key)
         if existing is None:
+            event._last_chunk_len = chunk_len  # type: ignore[attr-defined]
             self._pending_text_batches[key] = event
             self._pending_text_batch_counts[key] = 1
             self._schedule_text_batch_flush(key)
@@ -2504,6 +2523,7 @@ class FeishuAdapter(BasePlatformAdapter):
             return
 
         existing.text = next_text
+        existing._last_chunk_len = chunk_len  # type: ignore[attr-defined]
         existing.timestamp = event.timestamp
         if event.message_id:
             existing.message_id = event.message_id
@@ -2530,10 +2550,22 @@ class FeishuAdapter(BasePlatformAdapter):
         task_map[key] = asyncio.create_task(flush_fn(key))
 
     async def _flush_text_batch(self, key: str) -> None:
-        """Flush a pending text batch after the quiet period."""
+        """Flush a pending text batch after the quiet period.
+
+        Uses a longer delay when the latest chunk is near Feishu's ~4096-char
+        split point, since a continuation chunk is almost certain.
+        """
         current_task = asyncio.current_task()
         try:
-            await asyncio.sleep(self._text_batch_delay_seconds)
+            # Adaptive delay: if the latest chunk is near the split threshold,
+            # a continuation is almost certain — wait longer.
+            pending = self._pending_text_batches.get(key)
+            last_len = getattr(pending, "_last_chunk_len", 0) if pending else 0
+            if last_len >= self._SPLIT_THRESHOLD:
+                delay = self._text_batch_split_delay_seconds
+            else:
+                delay = self._text_batch_delay_seconds
+            await asyncio.sleep(delay)
             await self._flush_text_batch_now(key)
         finally:
             if self._pending_text_batch_tasks.get(key) is current_task:

gateway/platforms/matrix.py

@@ -18,6 +18,7 @@ Environment variables:
     MATRIX_REQUIRE_MENTION      Require @mention in rooms (default: true)
     MATRIX_FREE_RESPONSE_ROOMS  Comma-separated room IDs exempt from mention requirement
     MATRIX_AUTO_THREAD          Auto-create threads for room messages (default: true)
+    MATRIX_DM_MENTION_THREADS   Create a thread when bot is @mentioned in a DM (default: false)
 """
 
 from __future__ import annotations
@@ -40,6 +41,7 @@ from gateway.platforms.base import (
     BasePlatformAdapter,
     MessageEvent,
     MessageType,
+    ProcessingOutcome,
     SendResult,
 )
 
@@ -120,6 +122,11 @@ def check_matrix_requirements() -> bool:
 class MatrixAdapter(BasePlatformAdapter):
     """Gateway adapter for Matrix (any homeserver)."""
 
+    # Threshold for detecting Matrix client-side message splits.
+    # When a chunk is near the ~4000-char practical limit, a continuation
+    # is almost certain.
+    _SPLIT_THRESHOLD = 3900
+
     def __init__(self, config: PlatformConfig):
         super().__init__(config, Platform.MATRIX)
 
@@ -171,6 +178,16 @@ class MatrixAdapter(BasePlatformAdapter):
         self._reactions_enabled: bool = os.getenv(
             "MATRIX_REACTIONS", "true"
         ).lower() not in ("false", "0", "no")
+        # Tracks the reaction event_id for in-progress (eyes) reactions.
+        # Key: (room_id, message_event_id) → reaction_event_id (for the eyes reaction).
+        self._pending_reactions: dict[tuple[str, str], str] = {}
+
+        # Text batching: merge rapid successive messages (Telegram-style).
+        # Matrix clients split long messages around 4000 chars.
+        self._text_batch_delay_seconds = float(os.getenv("HERMES_MATRIX_TEXT_BATCH_DELAY_SECONDS", "0.6"))
+        self._text_batch_split_delay_seconds = float(os.getenv("HERMES_MATRIX_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0"))
+        self._pending_text_batches: Dict[str, MessageEvent] = {}
+        self._pending_text_batch_tasks: Dict[str, asyncio.Task] = {}
 
     def _is_duplicate_event(self, event_id) -> bool:
         """Return True if this event was already processed. Tracks the ID otherwise."""
@@ -1027,6 +1044,13 @@ class MatrixAdapter(BasePlatformAdapter):
                 if not self._is_bot_mentioned(body, formatted_body):
                     return
 
+        # DM mention-thread: when enabled, @mentioning bot in a DM creates a thread.
+        if is_dm and not thread_id:
+            dm_mention_threads = os.getenv("MATRIX_DM_MENTION_THREADS", "false").lower() in ("true", "1", "yes")
+            if dm_mention_threads and self._is_bot_mentioned(body, source_content.get("formatted_body")):
+                thread_id = event.event_id
+                self._track_thread(thread_id)
+
         # Strip mention from body when present (including in DMs).
         if self._is_bot_mentioned(body, source_content.get("formatted_body")):
             body = self._strip_mention(body)
@@ -1088,7 +1112,81 @@ class MatrixAdapter(BasePlatformAdapter):
         # Acknowledge receipt so the room shows as read (fire-and-forget).
         self._background_read_receipt(room.room_id, event.event_id)
 
-        await self.handle_message(msg_event)
+        # Only batch plain text messages — commands dispatch immediately.
+        if msg_type == MessageType.TEXT and self._text_batch_delay_seconds > 0:
+            self._enqueue_text_event(msg_event)
+        else:
+            await self.handle_message(msg_event)
+
+    # ------------------------------------------------------------------
+    # Text message aggregation (handles Matrix client-side splits)
+    # ------------------------------------------------------------------
+
+    def _text_batch_key(self, event: MessageEvent) -> str:
+        """Session-scoped key for text message batching."""
+        from gateway.session import build_session_key
+        return build_session_key(
+            event.source,
+            group_sessions_per_user=self.config.extra.get("group_sessions_per_user", True),
+            thread_sessions_per_user=self.config.extra.get("thread_sessions_per_user", False),
+        )
+
+    def _enqueue_text_event(self, event: MessageEvent) -> None:
+        """Buffer a text event and reset the flush timer.
+
+        When a Matrix client splits a long message, the chunks arrive within
+        a few hundred milliseconds.  This merges them into a single event
+        before dispatching.
+        """
+        key = self._text_batch_key(event)
+        existing = self._pending_text_batches.get(key)
+        chunk_len = len(event.text or "")
+        if existing is None:
+            event._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            self._pending_text_batches[key] = event
+        else:
+            if event.text:
+                existing.text = f"{existing.text}\n{event.text}" if existing.text else event.text
+            existing._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            # Merge any media that might be attached
+            if event.media_urls:
+                existing.media_urls.extend(event.media_urls)
+                existing.media_types.extend(event.media_types)
+
+        # Cancel any pending flush and restart the timer
+        prior_task = self._pending_text_batch_tasks.get(key)
+        if prior_task and not prior_task.done():
+            prior_task.cancel()
+        self._pending_text_batch_tasks[key] = asyncio.create_task(
+            self._flush_text_batch(key)
+        )
+
+    async def _flush_text_batch(self, key: str) -> None:
+        """Wait for the quiet period then dispatch the aggregated text.
+
+        Uses a longer delay when the latest chunk is near Matrix's ~4000-char
+        split point, since a continuation chunk is almost certain.
+        """
+        current_task = asyncio.current_task()
+        try:
+            pending = self._pending_text_batches.get(key)
+            last_len = getattr(pending, "_last_chunk_len", 0) if pending else 0
+            if last_len >= self._SPLIT_THRESHOLD:
+                delay = self._text_batch_split_delay_seconds
+            else:
+                delay = self._text_batch_delay_seconds
+            await asyncio.sleep(delay)
+            event = self._pending_text_batches.pop(key, None)
+            if not event:
+                return
+            logger.info(
+                "[Matrix] Flushing text batch %s (%d chars)",
+                key, len(event.text or ""),
+            )
+            await self.handle_message(event)
+        finally:
+            if self._pending_text_batch_tasks.get(key) is current_task:
+                self._pending_text_batch_tasks.pop(key, None)
 
     async def _on_room_message_media(self, room: Any, event: Any) -> None:
         """Handle incoming media messages (images, audio, video, files)."""
@@ -1270,6 +1368,13 @@ class MatrixAdapter(BasePlatformAdapter):
                 if not self._is_bot_mentioned(body, formatted_body):
                     return
 
+        # DM mention-thread: when enabled, @mentioning bot in a DM creates a thread.
+        if is_dm and not thread_id:
+            dm_mention_threads = os.getenv("MATRIX_DM_MENTION_THREADS", "false").lower() in ("true", "1", "yes")
+            if dm_mention_threads and self._is_bot_mentioned(body, source_content.get("formatted_body")):
+                thread_id = event.event_id
+                self._track_thread(thread_id)
+
         # Strip mention from body when present (including in DMs).
         if self._is_bot_mentioned(body, source_content.get("formatted_body")):
             body = self._strip_mention(body)
@@ -1350,12 +1455,14 @@ class MatrixAdapter(BasePlatformAdapter):
 
     async def _send_reaction(
         self, room_id: str, event_id: str, emoji: str,
-    ) -> bool:
-        """Send an emoji reaction to a message in a room."""
+    ) -> Optional[str]:
+        """Send an emoji reaction to a message in a room.
+        Returns the reaction event_id on success, None on failure.
+        """
         import nio
 
         if not self._client:
-            return False
+            return None
         content = {
             "m.relates_to": {
                 "rel_type": "m.annotation",
@@ -1370,12 +1477,12 @@ class MatrixAdapter(BasePlatformAdapter):
             )
             if isinstance(resp, nio.RoomSendResponse):
                 logger.debug("Matrix: sent reaction %s to %s", emoji, event_id)
-                return True
+                return resp.event_id
             logger.debug("Matrix: reaction send failed: %s", resp)
-            return False
+            return None
         except Exception as exc:
             logger.debug("Matrix: reaction send error: %s", exc)
-            return False
+            return None
 
     async def _redact_reaction(
         self, room_id: str, reaction_event_id: str, reason: str = "",
@@ -1390,10 +1497,12 @@ class MatrixAdapter(BasePlatformAdapter):
         msg_id = event.message_id
         room_id = event.source.chat_id
         if msg_id and room_id:
-            await self._send_reaction(room_id, msg_id, "\U0001f440")
+            reaction_event_id = await self._send_reaction(room_id, msg_id, "\U0001f440")
+            if reaction_event_id:
+                self._pending_reactions[(room_id, msg_id)] = reaction_event_id
 
     async def on_processing_complete(
-        self, event: MessageEvent, success: bool,
+        self, event: MessageEvent, outcome: ProcessingOutcome,
     ) -> None:
         """Replace eyes with checkmark (success) or cross (failure)."""
         if not self._reactions_enabled:
@@ -1402,11 +1511,18 @@ class MatrixAdapter(BasePlatformAdapter):
         room_id = event.source.chat_id
         if not msg_id or not room_id:
             return
-        # Note: Matrix doesn't support removing a specific reaction easily
-        # without tracking the reaction event_id. We send the new reaction;
-        # the eyes stays (acceptable UX — both are visible).
+        if outcome == ProcessingOutcome.CANCELLED:
+            return
+        # Remove the eyes reaction first, if we tracked its event_id.
+        reaction_key = (room_id, msg_id)
+        if reaction_key in self._pending_reactions:
+            eyes_event_id = self._pending_reactions.pop(reaction_key)
+            if not await self._redact_reaction(room_id, eyes_event_id):
+                logger.debug("Matrix: failed to redact eyes reaction %s", eyes_event_id)
         await self._send_reaction(
-            room_id, msg_id, "\u2705" if success else "\u274c",
+            room_id,
+            msg_id,
+            "\u2705" if outcome == ProcessingOutcome.SUCCESS else "\u274c",
         )
 
     async def _on_reaction(self, room: Any, event: Any) -> None:

gateway/platforms/slack.py

@@ -39,6 +39,7 @@ from gateway.platforms.base import (
     MessageType,
     SendResult,
     SUPPORTED_DOCUMENT_TYPES,
+    safe_url_for_log,
     cache_document_from_bytes,
 )
 
@@ -656,8 +657,19 @@ class SlackAdapter(BasePlatformAdapter):
         try:
             import httpx
 
+            async def _ssrf_redirect_guard(response):
+                """Re-check redirect targets so public URLs cannot bounce into private IPs."""
+                if response.is_redirect and response.next_request:
+                    redirect_url = str(response.next_request.url)
+                    if not is_safe_url(redirect_url):
+                        raise ValueError("Blocked redirect to private/internal address")
+
             # Download the image first
-            async with httpx.AsyncClient(timeout=30.0, follow_redirects=True) as client:
+            async with httpx.AsyncClient(
+                timeout=30.0,
+                follow_redirects=True,
+                event_hooks={"response": [_ssrf_redirect_guard]},
+            ) as client:
                 response = await client.get(image_url)
                 response.raise_for_status()
 
@@ -674,7 +686,7 @@ class SlackAdapter(BasePlatformAdapter):
         except Exception as e:  # pragma: no cover - defensive logging
             logger.warning(
                 "[Slack] Failed to upload image from URL %s, falling back to text: %s",
-                image_url,
+                safe_url_for_log(image_url),
                 e,
                 exc_info=True,
             )
@@ -1596,6 +1608,18 @@ class SlackAdapter(BasePlatformAdapter):
                     )
                     response.raise_for_status()
 
+                    # Slack may return an HTML sign-in/redirect page
+                    # instead of actual media bytes (e.g. expired token,
+                    # restricted file access).  Detect this early so we
+                    # don't cache bogus data and confuse downstream tools.
+                    ct = response.headers.get("content-type", "")
+                    if "text/html" in ct:
+                        raise ValueError(
+                            "Slack returned HTML instead of media "
+                            f"(content-type: {ct}); "
+                            "check bot token scopes and file permissions"
+                        )
+
                     if audio:
                         from gateway.platforms.base import cache_audio_from_bytes
                         return cache_audio_from_bytes(response.content, ext)

gateway/platforms/telegram.py

@@ -60,6 +60,7 @@ from gateway.platforms.base import (
     BasePlatformAdapter,
     MessageEvent,
     MessageType,
+    ProcessingOutcome,
     SendResult,
     cache_image_from_bytes,
     cache_audio_from_bytes,
@@ -121,6 +122,9 @@ class TelegramAdapter(BasePlatformAdapter):
     
     # Telegram message limits
     MAX_MESSAGE_LENGTH = 4096
+    # Threshold for detecting Telegram client-side message splits.
+    # When a chunk is near this limit, a continuation is almost certain.
+    _SPLIT_THRESHOLD = 4000
     MEDIA_GROUP_WAIT_SECONDS = 0.8
     
     def __init__(self, config: PlatformConfig):
@@ -140,6 +144,7 @@ class TelegramAdapter(BasePlatformAdapter):
         # Buffer rapid text messages so Telegram client-side splits of long
         # messages are aggregated into a single MessageEvent.
         self._text_batch_delay_seconds = float(os.getenv("HERMES_TELEGRAM_TEXT_BATCH_DELAY_SECONDS", "0.6"))
+        self._text_batch_split_delay_seconds = float(os.getenv("HERMES_TELEGRAM_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0"))
         self._pending_text_batches: Dict[str, MessageEvent] = {}
         self._pending_text_batch_tasks: Dict[str, asyncio.Task] = {}
         self._token_lock_identity: Optional[str] = None
@@ -513,6 +518,45 @@ class TelegramAdapter(BasePlatformAdapter):
 
             # Build the application
             builder = Application.builder().token(self.config.token)
+            custom_base_url = self.config.extra.get("base_url")
+            if custom_base_url:
+                builder = builder.base_url(custom_base_url)
+                builder = builder.base_file_url(
+                    self.config.extra.get("base_file_url", custom_base_url)
+                )
+                logger.info(
+                    "[%s] Using custom Telegram base_url: %s",
+                    self.name, custom_base_url,
+                )
+
+            # PTB defaults (pool_timeout=1s) are too aggressive on flaky networks and
+            # can trigger "Pool timeout: All connections in the connection pool are occupied"
+            # during reconnect/bootstrap. Use safer defaults and allow env overrides.
+            def _env_int(name: str, default: int) -> int:
+                try:
+                    return int(os.getenv(name, str(default)))
+                except (TypeError, ValueError):
+                    return default
+
+            def _env_float(name: str, default: float) -> float:
+                try:
+                    return float(os.getenv(name, str(default)))
+                except (TypeError, ValueError):
+                    return default
+
+            request_kwargs = {
+                "connection_pool_size": _env_int("HERMES_TELEGRAM_HTTP_POOL_SIZE", 512),
+                "pool_timeout": _env_float("HERMES_TELEGRAM_HTTP_POOL_TIMEOUT", 8.0),
+                "connect_timeout": _env_float("HERMES_TELEGRAM_HTTP_CONNECT_TIMEOUT", 10.0),
+                "read_timeout": _env_float("HERMES_TELEGRAM_HTTP_READ_TIMEOUT", 20.0),
+                "write_timeout": _env_float("HERMES_TELEGRAM_HTTP_WRITE_TIMEOUT", 20.0),
+            }
+
+            proxy_configured = any(
+                (os.getenv(k) or "").strip()
+                for k in ("HTTPS_PROXY", "HTTP_PROXY", "ALL_PROXY", "https_proxy", "http_proxy", "all_proxy")
+            )
+            disable_fallback = (os.getenv("HERMES_TELEGRAM_DISABLE_FALLBACK_IPS", "").strip().lower() in ("1", "true", "yes", "on"))
             fallback_ips = self._fallback_ips()
             if not fallback_ips:
                 fallback_ips = await discover_fallback_ips()
@@ -521,16 +565,32 @@ class TelegramAdapter(BasePlatformAdapter):
                     self.name,
                     ", ".join(fallback_ips),
                 )
-            if fallback_ips:
+
+            if fallback_ips and not proxy_configured and not disable_fallback:
                 logger.info(
                     "[%s] Telegram fallback IPs active: %s",
                     self.name,
                     ", ".join(fallback_ips),
                 )
-                transport = TelegramFallbackTransport(fallback_ips)
-                request = HTTPXRequest(httpx_kwargs={"transport": transport})
-                get_updates_request = HTTPXRequest(httpx_kwargs={"transport": transport})
-                builder = builder.request(request).get_updates_request(get_updates_request)
+                # Keep request/update pools separate to reduce contention during
+                # polling reconnect + bot API bootstrap/delete_webhook calls.
+                request = HTTPXRequest(
+                    **request_kwargs,
+                    httpx_kwargs={"transport": TelegramFallbackTransport(fallback_ips)},
+                )
+                get_updates_request = HTTPXRequest(
+                    **request_kwargs,
+                    httpx_kwargs={"transport": TelegramFallbackTransport(fallback_ips)},
+                )
+            else:
+                if proxy_configured:
+                    logger.info("[%s] Proxy configured; skipping Telegram fallback-IP transport", self.name)
+                elif disable_fallback:
+                    logger.info("[%s] Telegram fallback-IP transport disabled via env", self.name)
+                request = HTTPXRequest(**request_kwargs)
+                get_updates_request = HTTPXRequest(**request_kwargs)
+
+            builder = builder.request(request).get_updates_request(get_updates_request)
             self._app = builder.build()
             self._bot = self._app.bot
             
@@ -2160,12 +2220,15 @@ class TelegramAdapter(BasePlatformAdapter):
         """
         key = self._text_batch_key(event)
         existing = self._pending_text_batches.get(key)
+        chunk_len = len(event.text or "")
         if existing is None:
+            event._last_chunk_len = chunk_len  # type: ignore[attr-defined]
             self._pending_text_batches[key] = event
         else:
             # Append text from the follow-up chunk
             if event.text:
                 existing.text = f"{existing.text}\n{event.text}" if existing.text else event.text
+            existing._last_chunk_len = chunk_len  # type: ignore[attr-defined]
             # Merge any media that might be attached
             if event.media_urls:
                 existing.media_urls.extend(event.media_urls)
@@ -2180,10 +2243,22 @@ class TelegramAdapter(BasePlatformAdapter):
         )
 
     async def _flush_text_batch(self, key: str) -> None:
-        """Wait for the quiet period then dispatch the aggregated text."""
+        """Wait for the quiet period then dispatch the aggregated text.
+
+        Uses a longer delay when the latest chunk is near Telegram's 4096-char
+        split point, since a continuation chunk is almost certain.
+        """
         current_task = asyncio.current_task()
         try:
-            await asyncio.sleep(self._text_batch_delay_seconds)
+            # Adaptive delay: if the latest chunk is near Telegram's 4096-char
+            # split point, a continuation is almost certain — wait longer.
+            pending = self._pending_text_batches.get(key)
+            last_len = getattr(pending, "_last_chunk_len", 0) if pending else 0
+            if last_len >= self._SPLIT_THRESHOLD:
+                delay = self._text_batch_split_delay_seconds
+            else:
+                delay = self._text_batch_delay_seconds
+            await asyncio.sleep(delay)
             event = self._pending_text_batches.pop(key, None)
             if not event:
                 return
@@ -2713,7 +2788,7 @@ class TelegramAdapter(BasePlatformAdapter):
         if chat_id and message_id:
             await self._set_reaction(chat_id, message_id, "\U0001f440")
 
-    async def on_processing_complete(self, event: MessageEvent, success: bool) -> None:
+    async def on_processing_complete(self, event: MessageEvent, outcome: ProcessingOutcome) -> None:
         """Swap the in-progress reaction for a final success/failure reaction.
 
         Unlike Discord (additive reactions), Telegram's set_message_reaction
@@ -2723,5 +2798,9 @@ class TelegramAdapter(BasePlatformAdapter):
             return
         chat_id = getattr(event.source, "chat_id", None)
         message_id = getattr(event, "message_id", None)
-        if chat_id and message_id:
-            await self._set_reaction(chat_id, message_id, "\u2705" if success else "\u274c")
+        if chat_id and message_id and outcome != ProcessingOutcome.CANCELLED:
+            await self._set_reaction(
+                chat_id,
+                message_id,
+                "\U0001f44d" if outcome == ProcessingOutcome.SUCCESS else "\U0001f44e",
+            )

gateway/platforms/telegram_network.py

@@ -110,7 +110,8 @@ class TelegramFallbackTransport(httpx.AsyncBaseTransport):
                 logger.warning("[Telegram] Fallback IP %s failed: %s", ip, exc)
                 continue
 
-        assert last_error is not None
+        if last_error is None:
+            raise RuntimeError("All Telegram fallback IPs exhausted but no error was recorded")
         raise last_error
 
     async def aclose(self) -> None:

gateway/platforms/webhook.py

@@ -186,13 +186,23 @@ class WebhookAdapter(BasePlatformAdapter):
         if deliver_type == "github_comment":
             return await self._deliver_github_comment(content, delivery)
 
-        # Cross-platform delivery (telegram, discord, etc.)
+        # Cross-platform delivery — any platform with a gateway adapter
         if self.gateway_runner and deliver_type in (
             "telegram",
             "discord",
             "slack",
             "signal",
             "sms",
+            "whatsapp",
+            "matrix",
+            "mattermost",
+            "homeassistant",
+            "email",
+            "dingtalk",
+            "feishu",
+            "wecom",
+            "weixin",
+            "bluebubbles",
         ):
             return await self._deliver_cross_platform(
                 deliver_type, content, delivery
@@ -262,7 +272,7 @@ class WebhookAdapter(BasePlatformAdapter):
                 ", ".join(self._dynamic_routes.keys()) or "(none)",
             )
         except Exception as e:
-            logger.warning("[webhook] Failed to reload dynamic routes: %s", e)
+            logger.error("[webhook] Failed to reload dynamic routes: %s", e)
 
     async def _handle_webhook(self, request: "web.Request") -> "web.Response":
         """POST /webhooks/{route_name} — receive and process a webhook event."""

gateway/platforms/wecom.py

@@ -143,6 +143,9 @@ class WeComAdapter(BasePlatformAdapter):
     """WeCom AI Bot adapter backed by a persistent WebSocket connection."""
 
     MAX_MESSAGE_LENGTH = MAX_MESSAGE_LENGTH
+    # Threshold for detecting WeCom client-side message splits.
+    # When a chunk is near the 4000-char limit, a continuation is almost certain.
+    _SPLIT_THRESHOLD = 3900
 
     def __init__(self, config: PlatformConfig):
         super().__init__(config, Platform.WECOM)
@@ -172,6 +175,13 @@ class WeComAdapter(BasePlatformAdapter):
         self._seen_messages: Dict[str, float] = {}
         self._reply_req_ids: Dict[str, str] = {}
 
+        # Text batching: merge rapid successive messages (Telegram-style).
+        # WeCom clients split long messages around 4000 chars.
+        self._text_batch_delay_seconds = float(os.getenv("HERMES_WECOM_TEXT_BATCH_DELAY_SECONDS", "0.6"))
+        self._text_batch_split_delay_seconds = float(os.getenv("HERMES_WECOM_TEXT_BATCH_SPLIT_DELAY_SECONDS", "2.0"))
+        self._pending_text_batches: Dict[str, MessageEvent] = {}
+        self._pending_text_batch_tasks: Dict[str, asyncio.Task] = {}
+
     # ------------------------------------------------------------------
     # Connection lifecycle
     # ------------------------------------------------------------------
@@ -519,7 +529,82 @@ class WeComAdapter(BasePlatformAdapter):
             timestamp=datetime.now(tz=timezone.utc),
         )
 
-        await self.handle_message(event)
+        # Only batch plain text messages — commands, media, etc. dispatch
+        # immediately since they won't be split by the WeCom client.
+        if message_type == MessageType.TEXT and self._text_batch_delay_seconds > 0:
+            self._enqueue_text_event(event)
+        else:
+            await self.handle_message(event)
+
+    # ------------------------------------------------------------------
+    # Text message aggregation (handles WeCom client-side splits)
+    # ------------------------------------------------------------------
+
+    def _text_batch_key(self, event: MessageEvent) -> str:
+        """Session-scoped key for text message batching."""
+        from gateway.session import build_session_key
+        return build_session_key(
+            event.source,
+            group_sessions_per_user=self.config.extra.get("group_sessions_per_user", True),
+            thread_sessions_per_user=self.config.extra.get("thread_sessions_per_user", False),
+        )
+
+    def _enqueue_text_event(self, event: MessageEvent) -> None:
+        """Buffer a text event and reset the flush timer.
+
+        When WeCom splits a long user message at 4000 chars, the chunks
+        arrive within a few hundred milliseconds.  This merges them into
+        a single event before dispatching.
+        """
+        key = self._text_batch_key(event)
+        existing = self._pending_text_batches.get(key)
+        chunk_len = len(event.text or "")
+        if existing is None:
+            event._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            self._pending_text_batches[key] = event
+        else:
+            if event.text:
+                existing.text = f"{existing.text}\n{event.text}" if existing.text else event.text
+            existing._last_chunk_len = chunk_len  # type: ignore[attr-defined]
+            # Merge any media that might be attached
+            if event.media_urls:
+                existing.media_urls.extend(event.media_urls)
+                existing.media_types.extend(event.media_types)
+
+        # Cancel any pending flush and restart the timer
+        prior_task = self._pending_text_batch_tasks.get(key)
+        if prior_task and not prior_task.done():
+            prior_task.cancel()
+        self._pending_text_batch_tasks[key] = asyncio.create_task(
+            self._flush_text_batch(key)
+        )
+
+    async def _flush_text_batch(self, key: str) -> None:
+        """Wait for the quiet period then dispatch the aggregated text.
+
+        Uses a longer delay when the latest chunk is near WeCom's 4000-char
+        split point, since a continuation chunk is almost certain.
+        """
+        current_task = asyncio.current_task()
+        try:
+            pending = self._pending_text_batches.get(key)
+            last_len = getattr(pending, "_last_chunk_len", 0) if pending else 0
+            if last_len >= self._SPLIT_THRESHOLD:
+                delay = self._text_batch_split_delay_seconds
+            else:
+                delay = self._text_batch_delay_seconds
+            await asyncio.sleep(delay)
+            event = self._pending_text_batches.pop(key, None)
+            if not event:
+                return
+            logger.info(
+                "[WeCom] Flushing text batch %s (%d chars)",
+                key, len(event.text or ""),
+            )
+            await self.handle_message(event)
+        finally:
+            if self._pending_text_batch_tasks.get(key) is current_task:
+                self._pending_text_batch_tasks.pop(key, None)
 
     @staticmethod
     def _extract_text(body: Dict[str, Any]) -> Tuple[str, Optional[str]]:
@@ -611,7 +696,11 @@ class WeComAdapter(BasePlatformAdapter):
 
             if kind == "image":
                 ext = self._detect_image_ext(raw)
-                return cache_image_from_bytes(raw, ext), self._mime_for_ext(ext, fallback="image/jpeg")
+                try:
+                    return cache_image_from_bytes(raw, ext), self._mime_for_ext(ext, fallback="image/jpeg")
+                except ValueError as exc:
+                    logger.warning("[%s] Rejected non-image bytes: %s", self.name, exc)
+                    return None
 
             filename = str(media.get("filename") or media.get("name") or "wecom_file")
             return cache_document_from_bytes(raw, filename), mimetypes.guess_type(filename)[0] or "application/octet-stream"
@@ -637,7 +726,11 @@ class WeComAdapter(BasePlatformAdapter):
         content_type = str(headers.get("content-type") or "").split(";", 1)[0].strip() or "application/octet-stream"
         if kind == "image":
             ext = self._guess_extension(url, content_type, fallback=self._detect_image_ext(raw))
-            return cache_image_from_bytes(raw, ext), content_type or self._mime_for_ext(ext, fallback="image/jpeg")
+            try:
+                return cache_image_from_bytes(raw, ext), content_type or self._mime_for_ext(ext, fallback="image/jpeg")
+            except ValueError as exc:
+                logger.warning("[%s] Rejected non-image bytes from %s: %s", self.name, url, exc)
+                return None
 
         filename = self._guess_filename(url, headers.get("content-disposition"), content_type)
         return cache_document_from_bytes(raw, filename), content_type

gateway/run.py

@@ -481,6 +481,7 @@ class GatewayRunner:
         self._prefill_messages = self._load_prefill_messages()
         self._ephemeral_system_prompt = self._load_ephemeral_system_prompt()
         self._reasoning_config = self._load_reasoning_config()
+        self._service_tier = self._load_service_tier()
         self._show_reasoning = self._load_show_reasoning()
         self._provider_routing = self._load_provider_routing()
         self._fallback_model = self._load_fallback_model()
@@ -776,6 +777,7 @@ class GatewayRunner:
 
     def _resolve_turn_agent_config(self, user_message: str, model: str, runtime_kwargs: dict) -> dict:
         from agent.smart_model_routing import resolve_turn_route
+        from hermes_cli.models import resolve_fast_mode_overrides
 
         primary = {
             "model": model,
@@ -787,7 +789,19 @@ class GatewayRunner:
             "args": list(runtime_kwargs.get("args") or []),
             "credential_pool": runtime_kwargs.get("credential_pool"),
         }
-        return resolve_turn_route(user_message, getattr(self, "_smart_model_routing", {}), primary)
+        route = resolve_turn_route(user_message, getattr(self, "_smart_model_routing", {}), primary)
+
+        service_tier = getattr(self, "_service_tier", None)
+        if not service_tier:
+            route["request_overrides"] = None
+            return route
+
+        try:
+            overrides = resolve_fast_mode_overrides(route.get("model"))
+        except Exception:
+            overrides = None
+        route["request_overrides"] = overrides
+        return route
 
     async def _handle_adapter_fatal_error(self, adapter: BasePlatformAdapter) -> None:
         """React to an adapter failure after startup.
@@ -939,6 +953,33 @@ class GatewayRunner:
             logger.warning("Unknown reasoning_effort '%s', using default (medium)", effort)
         return result
 
+    @staticmethod
+    def _load_service_tier() -> str | None:
+        """Load Priority Processing setting from config.yaml.
+
+        Reads agent.service_tier from config.yaml. Accepted values mirror the CLI:
+        "fast"/"priority"/"on" => "priority", while "normal"/"off" disables it.
+        Returns None when unset or unsupported.
+        """
+        raw = ""
+        try:
+            import yaml as _y
+            cfg_path = _hermes_home / "config.yaml"
+            if cfg_path.exists():
+                with open(cfg_path, encoding="utf-8") as _f:
+                    cfg = _y.safe_load(_f) or {}
+                raw = str(cfg.get("agent", {}).get("service_tier", "") or "").strip()
+        except Exception:
+            pass
+
+        value = raw.lower()
+        if not value or value in {"normal", "default", "standard", "off", "none"}:
+            return None
+        if value in {"fast", "priority", "on"}:
+            return "priority"
+        logger.warning("Unknown service_tier '%s', ignoring", raw)
+        return None
+
     @staticmethod
     def _load_show_reasoning() -> bool:
         """Load show_reasoning toggle from config.yaml display section."""
@@ -1069,6 +1110,7 @@ class GatewayRunner:
                        "MATRIX_ALLOWED_USERS", "DINGTALK_ALLOWED_USERS",
                        "FEISHU_ALLOWED_USERS",
                        "WECOM_ALLOWED_USERS",
+                       "WEIXIN_ALLOWED_USERS",
                        "BLUEBUBBLES_ALLOWED_USERS",
                        "GATEWAY_ALLOWED_USERS")
         )
@@ -1081,6 +1123,7 @@ class GatewayRunner:
                        "MATRIX_ALLOW_ALL_USERS", "DINGTALK_ALLOW_ALL_USERS",
                        "FEISHU_ALLOW_ALL_USERS",
                        "WECOM_ALLOW_ALL_USERS",
+                       "WEIXIN_ALLOW_ALL_USERS",
                        "BLUEBUBBLES_ALLOW_ALL_USERS")
         )
         if not _any_allowlist and not _allow_all:
@@ -1305,12 +1348,28 @@ class GatewayRunner:
                 for key, entry in _expired_entries:
                     try:
                         await self._async_flush_memories(entry.session_id)
-                        # Shut down memory provider on the cached agent
-                        cached_agent = self._running_agents.get(key)
-                        if cached_agent and cached_agent is not _AGENT_PENDING_SENTINEL:
+                        # Shut down memory provider and close tool resources
+                        # on the cached agent.  Idle agents live in
+                        # _agent_cache (not _running_agents), so look there.
+                        _cached_agent = None
+                        _cache_lock = getattr(self, "_agent_cache_lock", None)
+                        if _cache_lock is not None:
+                            with _cache_lock:
+                                _cached = self._agent_cache.get(key)
+                                _cached_agent = _cached[0] if isinstance(_cached, tuple) else _cached if _cached else None
+                        # Fall back to _running_agents in case the agent is
+                        # still mid-turn when the expiry fires.
+                        if _cached_agent is None:
+                            _cached_agent = self._running_agents.get(key)
+                        if _cached_agent and _cached_agent is not _AGENT_PENDING_SENTINEL:
                             try:
-                                if hasattr(cached_agent, 'shutdown_memory_provider'):
-                                    cached_agent.shutdown_memory_provider()
+                                if hasattr(_cached_agent, 'shutdown_memory_provider'):
+                                    _cached_agent.shutdown_memory_provider()
+                            except Exception:
+                                pass
+                            try:
+                                if hasattr(_cached_agent, 'close'):
+                                    _cached_agent.close()
                             except Exception:
                                 pass
                         # Mark as flushed and persist to disk so the flag
@@ -1493,6 +1552,14 @@ class GatewayRunner:
                     agent.shutdown_memory_provider()
             except Exception:
                 pass
+            # Close tool resources (terminal sandboxes, browser daemons,
+            # background processes, httpx clients) to prevent zombie
+            # process accumulation.
+            try:
+                if hasattr(agent, 'close'):
+                    agent.close()
+            except Exception:
+                pass
 
         for platform, adapter in list(self.adapters.items()):
             try:
@@ -1515,7 +1582,25 @@ class GatewayRunner:
         self._pending_messages.clear()
         self._pending_approvals.clear()
         self._shutdown_event.set()
-        
+
+        # Global cleanup: kill any remaining tool subprocesses not tied
+        # to a specific agent (catch-all for zombie prevention).
+        try:
+            from tools.process_registry import process_registry
+            process_registry.kill_all()
+        except Exception:
+            pass
+        try:
+            from tools.terminal_tool import cleanup_all_environments
+            cleanup_all_environments()
+        except Exception:
+            pass
+        try:
+            from tools.browser_tool import cleanup_all_browsers
+            cleanup_all_browsers()
+        except Exception:
+            pass
+
         from gateway.status import remove_pid_file, write_runtime_status
         remove_pid_file()
         try:
@@ -1622,6 +1707,13 @@ class GatewayRunner:
                 return None
             return WeComAdapter(config)
 
+        elif platform == Platform.WEIXIN:
+            from gateway.platforms.weixin import WeixinAdapter, check_weixin_requirements
+            if not check_weixin_requirements():
+                logger.warning("Weixin: aiohttp/cryptography not installed")
+                return None
+            return WeixinAdapter(config)
+
         elif platform == Platform.MATTERMOST:
             from gateway.platforms.mattermost import MattermostAdapter, check_mattermost_requirements
             if not check_mattermost_requirements():
@@ -1697,6 +1789,7 @@ class GatewayRunner:
             Platform.DINGTALK: "DINGTALK_ALLOWED_USERS",
             Platform.FEISHU: "FEISHU_ALLOWED_USERS",
             Platform.WECOM: "WECOM_ALLOWED_USERS",
+            Platform.WEIXIN: "WEIXIN_ALLOWED_USERS",
             Platform.BLUEBUBBLES: "BLUEBUBBLES_ALLOWED_USERS",
         }
         platform_allow_all_map = {
@@ -1712,6 +1805,7 @@ class GatewayRunner:
             Platform.DINGTALK: "DINGTALK_ALLOW_ALL_USERS",
             Platform.FEISHU: "FEISHU_ALLOW_ALL_USERS",
             Platform.WECOM: "WECOM_ALLOW_ALL_USERS",
+            Platform.WEIXIN: "WEIXIN_ALLOW_ALL_USERS",
             Platform.BLUEBUBBLES: "BLUEBUBBLES_ALLOW_ALL_USERS",
         }
 
@@ -1991,6 +2085,11 @@ class GatewayRunner:
                     return await self._handle_approve_command(event)
                 return await self._handle_deny_command(event)
 
+            # /background must bypass the running-agent guard — it starts a
+            # parallel task and must never interrupt the active conversation.
+            if _cmd_def_inner and _cmd_def_inner.name == "background":
+                return await self._handle_background_command(event)
+
             if event.message_type == MessageType.PHOTO:
                 logger.debug("PRIORITY photo follow-up for session %s — queueing without interrupt", _quick_key[:20])
                 adapter = self.adapters.get(source.platform)
@@ -2072,6 +2171,9 @@ class GatewayRunner:
         if canonical == "reasoning":
             return await self._handle_reasoning_command(event)
 
+        if canonical == "fast":
+            return await self._handle_fast_command(event)
+
         if canonical == "verbose":
             return await self._handle_verbose_command(event)
 
@@ -2340,8 +2442,8 @@ class GatewayRunner:
         # Build session context
         context = build_session_context(source, self.config, session_entry)
         
-        # Set environment variables for tools
-        self._set_session_env(context)
+        # Set session context variables for tools (task-local, concurrency-safe)
+        _session_env_tokens = self._set_session_env(context)
         
         # Read privacy.redact_pii from config (re-read per message)
         _redact_pii = False
@@ -2414,37 +2516,41 @@ class GatewayRunner:
             session_entry.was_auto_reset = False
             session_entry.auto_reset_reason = None
 
-        # Auto-load skill for DM topic bindings (e.g., Telegram Private Chat Topics)
-        # Only inject on NEW sessions — for ongoing conversations the skill content
-        # is already in the conversation history from the first message.
-        if _is_new_session and getattr(event, "auto_skill", None):
+        # Auto-load skill(s) for topic/channel bindings (Telegram DM Topics,
+        # Discord channel_skill_bindings).  Supports a single name or ordered list.
+        # Only inject on NEW sessions — ongoing conversations already have the
+        # skill content in their conversation history from the first message.
+        _auto = getattr(event, "auto_skill", None)
+        if _is_new_session and _auto:
+            _skill_names = [_auto] if isinstance(_auto, str) else list(_auto)
             try:
                 from agent.skill_commands import _load_skill_payload, _build_skill_message
-                _skill_name = event.auto_skill
-                _loaded = _load_skill_payload(_skill_name, task_id=_quick_key)
-                if _loaded:
-                    _loaded_skill, _skill_dir, _display_name = _loaded
-                    _activation_note = (
-                        f'[SYSTEM: This conversation is in a topic with the "{_display_name}" skill '
-                        f"auto-loaded. Follow its instructions for the duration of this session.]"
-                    )
-                    _skill_msg = _build_skill_message(
-                        _loaded_skill, _skill_dir, _activation_note,
-                        user_instruction=event.text,
-                    )
-                    if _skill_msg:
-                        event.text = _skill_msg
-                        logger.info(
-                            "[Gateway] Auto-loaded skill '%s' for DM topic session %s",
-                            _skill_name, session_key,
+                _combined_parts: list[str] = []
+                _loaded_names: list[str] = []
+                for _sname in _skill_names:
+                    _loaded = _load_skill_payload(_sname, task_id=_quick_key)
+                    if _loaded:
+                        _loaded_skill, _skill_dir, _display_name = _loaded
+                        _note = (
+                            f'[SYSTEM: The "{_display_name}" skill is auto-loaded. '
+                            f"Follow its instructions for this session.]"
                         )
-                else:
-                    logger.warning(
-                        "[Gateway] DM topic skill '%s' not found in available skills",
-                        _skill_name,
+                        _part = _build_skill_message(_loaded_skill, _skill_dir, _note)
+                        if _part:
+                            _combined_parts.append(_part)
+                            _loaded_names.append(_sname)
+                    else:
+                        logger.warning("[Gateway] Auto-skill '%s' not found", _sname)
+                if _combined_parts:
+                    # Append the user's original text after all skill payloads
+                    _combined_parts.append(event.text)
+                    event.text = "\n\n".join(_combined_parts)
+                    logger.info(
+                        "[Gateway] Auto-loaded skill(s) %s for session %s",
+                        _loaded_names, session_key,
                     )
             except Exception as e:
-                logger.warning("[Gateway] Failed to auto-load topic skill '%s': %s", event.auto_skill, e)
+                logger.warning("[Gateway] Failed to auto-load skill(s) %s: %s", _skill_names, e)
 
         # Load conversation history from transcript
         history = self.session_store.load_transcript(session_entry.session_id)
@@ -3170,8 +3276,8 @@ class GatewayRunner:
                 "Try again or use /reset to start a fresh session."
             )
         finally:
-            # Clear session env
-            self._clear_session_env()
+            # Restore session context variables to their pre-handler state
+            self._clear_session_env(_session_env_tokens)
     
     def _format_session_info(self) -> str:
         """Resolve current model config and return a formatted info block.
@@ -3271,8 +3377,22 @@ class GatewayRunner:
                 _flush_task.add_done_callback(self._background_tasks.discard)
         except Exception as e:
             logger.debug("Gateway memory flush on reset failed: %s", e)
+        # Close tool resources on the old agent (terminal sandboxes, browser
+        # daemons, background processes) before evicting from cache.
+        # Guard with getattr because test fixtures may skip __init__.
+        _cache_lock = getattr(self, "_agent_cache_lock", None)
+        if _cache_lock is not None:
+            with _cache_lock:
+                _cached = self._agent_cache.get(session_key)
+                _old_agent = _cached[0] if isinstance(_cached, tuple) else _cached if _cached else None
+            if _old_agent is not None:
+                try:
+                    if hasattr(_old_agent, "close"):
+                        _old_agent.close()
+                except Exception:
+                    pass
         self._evict_cached_agent(session_key)
-        
+
         try:
             from tools.env_passthrough import clear_env_passthrough
             clear_env_passthrough()
@@ -3540,6 +3660,7 @@ class GatewayRunner:
         current_base_url = ""
         current_api_key = ""
         user_provs = None
+        custom_provs = None
         config_path = _hermes_home / "config.yaml"
         try:
             if config_path.exists():
@@ -3551,6 +3672,7 @@ class GatewayRunner:
                     current_provider = model_cfg.get("provider", current_provider)
                     current_base_url = model_cfg.get("base_url", "")
                 user_provs = cfg.get("providers")
+                custom_provs = cfg.get("custom_providers")
         except Exception:
             pass
 
@@ -3578,6 +3700,7 @@ class GatewayRunner:
                     providers = list_authenticated_providers(
                         current_provider=current_provider,
                         user_providers=user_provs,
+                        custom_providers=custom_provs,
                         max_models=50,
                     )
                 except Exception:
@@ -3605,6 +3728,8 @@ class GatewayRunner:
                             current_api_key=_cur_api_key,
                             is_global=False,
                             explicit_provider=provider_slug,
+                            user_providers=user_provs,
+                            custom_providers=custom_provs,
                         )
                         if not result.success:
                             return f"Error: {result.error_message}"
@@ -3683,6 +3808,7 @@ class GatewayRunner:
                 providers = list_authenticated_providers(
                     current_provider=current_provider,
                     user_providers=user_provs,
+                    custom_providers=custom_provs,
                     max_models=5,
                 )
                 for p in providers:
@@ -3712,6 +3838,8 @@ class GatewayRunner:
             current_api_key=current_api_key,
             is_global=persist_global,
             explicit_provider=explicit_provider,
+            user_providers=user_provs,
+            custom_providers=custom_provs,
         )
 
         if not result.success:
@@ -3833,6 +3961,7 @@ class GatewayRunner:
 
         # Resolve current provider from config
         current_provider = "openrouter"
+        model_cfg = {}
         config_path = _hermes_home / 'config.yaml'
         try:
             if config_path.exists():
@@ -4573,6 +4702,7 @@ class GatewayRunner:
             max_iterations = int(os.getenv("HERMES_MAX_ITERATIONS", "90"))
             reasoning_config = self._load_reasoning_config()
             self._reasoning_config = reasoning_config
+            self._service_tier = self._load_service_tier()
             turn_route = self._resolve_turn_agent_config(prompt, model, runtime_kwargs)
 
             def run_sync():
@@ -4584,6 +4714,8 @@ class GatewayRunner:
                     verbose_logging=False,
                     enabled_toolsets=enabled_toolsets,
                     reasoning_config=reasoning_config,
+                    service_tier=self._service_tier,
+                    request_overrides=turn_route.get("request_overrides"),
                     providers_allowed=pr.get("only"),
                     providers_ignored=pr.get("ignore"),
                     providers_order=pr.get("order"),
@@ -4733,6 +4865,7 @@ class GatewayRunner:
             model = _resolve_gateway_model(user_config)
             platform_key = _platform_config_key(source.platform)
             reasoning_config = self._load_reasoning_config()
+            self._service_tier = self._load_service_tier()
             turn_route = self._resolve_turn_agent_config(question, model, runtime_kwargs)
             pr = self._provider_routing
 
@@ -4759,6 +4892,8 @@ class GatewayRunner:
                     verbose_logging=False,
                     enabled_toolsets=[],
                     reasoning_config=reasoning_config,
+                    service_tier=self._service_tier,
+                    request_overrides=turn_route.get("request_overrides"),
                     providers_allowed=pr.get("only"),
                     providers_ignored=pr.get("ignore"),
                     providers_order=pr.get("order"),
@@ -4912,15 +5047,82 @@ class GatewayRunner:
         else:
             return f"🧠 ✓ Reasoning effort set to `{effort}` (this session only)"
 
-    async def _handle_yolo_command(self, event: MessageEvent) -> str:
-        """Handle /yolo — toggle dangerous command approval bypass."""
-        current = bool(os.environ.get("HERMES_YOLO_MODE"))
-        if current:
-            os.environ.pop("HERMES_YOLO_MODE", None)
-            return "⚠️ YOLO mode **OFF** — dangerous commands will require approval."
+    async def _handle_fast_command(self, event: MessageEvent) -> str:
+        """Handle /fast — mirror the CLI Priority Processing toggle in gateway chats."""
+        import yaml
+        from hermes_cli.models import model_supports_fast_mode
+
+        args = event.get_command_args().strip().lower()
+        config_path = _hermes_home / "config.yaml"
+        self._service_tier = self._load_service_tier()
+
+        user_config = _load_gateway_config()
+        model = _resolve_gateway_model(user_config)
+        if not model_supports_fast_mode(model):
+            return "⚡ /fast is only available for OpenAI models that support Priority Processing."
+
+        def _save_config_key(key_path: str, value):
+            """Save a dot-separated key to config.yaml."""
+            try:
+                user_config = {}
+                if config_path.exists():
+                    with open(config_path, encoding="utf-8") as f:
+                        user_config = yaml.safe_load(f) or {}
+                keys = key_path.split(".")
+                current = user_config
+                for k in keys[:-1]:
+                    if k not in current or not isinstance(current[k], dict):
+                        current[k] = {}
+                    current = current[k]
+                current[keys[-1]] = value
+                atomic_yaml_write(config_path, user_config)
+                return True
+            except Exception as e:
+                logger.error("Failed to save config key %s: %s", key_path, e)
+                return False
+
+        if not args or args == "status":
+            status = "fast" if self._service_tier == "priority" else "normal"
+            return (
+                "⚡ Priority Processing\n\n"
+                f"Current mode: `{status}`\n\n"
+                "_Usage:_ `/fast <normal|fast|status>`"
+            )
+
+        if args in {"fast", "on"}:
+            self._service_tier = "priority"
+            saved_value = "fast"
+            label = "FAST"
+        elif args in {"normal", "off"}:
+            self._service_tier = None
+            saved_value = "normal"
+            label = "NORMAL"
         else:
-            os.environ["HERMES_YOLO_MODE"] = "1"
-            return "⚡ YOLO mode **ON** — all commands auto-approved. Use with caution."
+            return (
+                f"⚠️ Unknown argument: `{args}`\n\n"
+                "**Valid options:** normal, fast, status"
+            )
+
+        if _save_config_key("agent.service_tier", saved_value):
+            return f"⚡ ✓ Priority Processing: **{label}** (saved to config)\n_(takes effect on next message)_"
+        return f"⚡ ✓ Priority Processing: **{label}** (this session only)"
+
+    async def _handle_yolo_command(self, event: MessageEvent) -> str:
+        """Handle /yolo — toggle dangerous command approval bypass for this session only."""
+        from tools.approval import (
+            disable_session_yolo,
+            enable_session_yolo,
+            is_session_yolo_enabled,
+        )
+
+        session_key = self._session_key_for_source(event.source)
+        current = is_session_yolo_enabled(session_key)
+        if current:
+            disable_session_yolo(session_key)
+            return "⚠️ YOLO mode **OFF** for this session — dangerous commands will require approval."
+        else:
+            enable_session_yolo(session_key)
+            return "⚡ YOLO mode **ON** for this session — all commands auto-approved. Use with caution."
 
     async def _handle_verbose_command(self, event: MessageEvent) -> str:
         """Handle /verbose command — cycle tool progress display mode.
@@ -5268,27 +5470,76 @@ class GatewayRunner:
         )
 
     async def _handle_usage_command(self, event: MessageEvent) -> str:
-        """Handle /usage command -- show token usage for the session's last agent run."""
+        """Handle /usage command -- show token usage for the current session.
+
+        Checks both _running_agents (mid-turn) and _agent_cache (between turns)
+        so that rate limits, cost estimates, and detailed token breakdowns are
+        available whenever the user asks, not only while the agent is running.
+        """
         source = event.source
         session_key = self._session_key_for_source(source)
 
+        # Try running agent first (mid-turn), then cached agent (between turns)
         agent = self._running_agents.get(session_key)
+        if not agent or agent is _AGENT_PENDING_SENTINEL:
+            _cache_lock = getattr(self, "_agent_cache_lock", None)
+            _cache = getattr(self, "_agent_cache", None)
+            if _cache_lock and _cache is not None:
+                with _cache_lock:
+                    cached = _cache.get(session_key)
+                    if cached:
+                        agent = cached[0]
+
         if agent and hasattr(agent, "session_total_tokens") and agent.session_api_calls > 0:
             lines = []
 
-            # Rate limits first (when available from provider headers)
+            # Rate limits (when available from provider headers)
             rl_state = agent.get_rate_limit_state()
             if rl_state and rl_state.has_data:
                 from agent.rate_limit_tracker import format_rate_limit_compact
                 lines.append(f"⏱️ **Rate Limits:** {format_rate_limit_compact(rl_state)}")
                 lines.append("")
 
-            # Session token usage
+            # Session token usage — detailed breakdown matching CLI
+            input_tokens = getattr(agent, "session_input_tokens", 0) or 0
+            output_tokens = getattr(agent, "session_output_tokens", 0) or 0
+            cache_read = getattr(agent, "session_cache_read_tokens", 0) or 0
+            cache_write = getattr(agent, "session_cache_write_tokens", 0) or 0
+
             lines.append("📊 **Session Token Usage**")
-            lines.append(f"Prompt (input): {agent.session_prompt_tokens:,}")
-            lines.append(f"Completion (output): {agent.session_completion_tokens:,}")
+            lines.append(f"Model: `{agent.model}`")
+            lines.append(f"Input tokens: {input_tokens:,}")
+            if cache_read:
+                lines.append(f"Cache read tokens: {cache_read:,}")
+            if cache_write:
+                lines.append(f"Cache write tokens: {cache_write:,}")
+            lines.append(f"Output tokens: {output_tokens:,}")
             lines.append(f"Total: {agent.session_total_tokens:,}")
             lines.append(f"API calls: {agent.session_api_calls}")
+
+            # Cost estimation
+            try:
+                from agent.usage_pricing import CanonicalUsage, estimate_usage_cost
+                cost_result = estimate_usage_cost(
+                    agent.model,
+                    CanonicalUsage(
+                        input_tokens=input_tokens,
+                        output_tokens=output_tokens,
+                        cache_read_tokens=cache_read,
+                        cache_write_tokens=cache_write,
+                    ),
+                    provider=getattr(agent, "provider", None),
+                    base_url=getattr(agent, "base_url", None),
+                )
+                if cost_result.amount_usd is not None:
+                    prefix = "~" if cost_result.status == "estimated" else ""
+                    lines.append(f"Cost: {prefix}${float(cost_result.amount_usd):.4f}")
+                elif cost_result.status == "included":
+                    lines.append("Cost: included")
+            except Exception:
+                pass
+
+            # Context window and compressions
             ctx = agent.context_compressor
             if ctx.last_prompt_tokens:
                 pct = min(100, ctx.last_prompt_tokens / ctx.context_length * 100) if ctx.context_length else 0
@@ -5298,7 +5549,7 @@ class GatewayRunner:
 
             return "\n".join(lines)
 
-        # No running agent -- check session history for a rough count
+        # No agent at all -- check session history for a rough count
         session_entry = self.session_store.get_or_create_session(source)
         history = self.session_store.load_transcript(session_entry.session_id)
         if history:
@@ -5309,7 +5560,7 @@ class GatewayRunner:
                 f"📊 **Session Info**\n"
                 f"Messages: {len(msgs)}\n"
                 f"Estimated context: ~{approx:,} tokens\n"
-                f"_(Detailed usage available during active conversations)_"
+                f"_(Detailed usage available after the first agent response)_"
             )
         return "No usage data available for this session."
 
@@ -5537,7 +5788,7 @@ class GatewayRunner:
         Platform.TELEGRAM, Platform.DISCORD, Platform.SLACK, Platform.WHATSAPP,
         Platform.SIGNAL, Platform.MATTERMOST, Platform.MATRIX,
         Platform.HOMEASSISTANT, Platform.EMAIL, Platform.SMS, Platform.DINGTALK,
-        Platform.FEISHU, Platform.WECOM, Platform.BLUEBUBBLES, Platform.LOCAL,
+        Platform.FEISHU, Platform.WECOM, Platform.WEIXIN, Platform.BLUEBUBBLES, Platform.LOCAL,
     })
 
     async def _handle_update_command(self, event: MessageEvent) -> str:
@@ -5925,20 +6176,27 @@ class GatewayRunner:
 
         return True
 
-    def _set_session_env(self, context: SessionContext) -> None:
-        """Set environment variables for the current session."""
-        os.environ["HERMES_SESSION_PLATFORM"] = context.source.platform.value
-        os.environ["HERMES_SESSION_CHAT_ID"] = context.source.chat_id
-        if context.source.chat_name:
-            os.environ["HERMES_SESSION_CHAT_NAME"] = context.source.chat_name
-        if context.source.thread_id:
-            os.environ["HERMES_SESSION_THREAD_ID"] = str(context.source.thread_id)
-    
-    def _clear_session_env(self) -> None:
-        """Clear session environment variables."""
-        for var in ["HERMES_SESSION_PLATFORM", "HERMES_SESSION_CHAT_ID", "HERMES_SESSION_CHAT_NAME", "HERMES_SESSION_THREAD_ID"]:
-            if var in os.environ:
-                del os.environ[var]
+    def _set_session_env(self, context: SessionContext) -> list:
+        """Set session context variables for the current async task.
+
+        Uses ``contextvars`` instead of ``os.environ`` so that concurrent
+        gateway messages cannot overwrite each other's session state.
+
+        Returns a list of reset tokens; pass them to ``_clear_session_env``
+        in a ``finally`` block.
+        """
+        from gateway.session_context import set_session_vars
+        return set_session_vars(
+            platform=context.source.platform.value,
+            chat_id=context.source.chat_id,
+            chat_name=context.source.chat_name or "",
+            thread_id=str(context.source.thread_id) if context.source.thread_id else "",
+        )
+
+    def _clear_session_env(self, tokens: list) -> None:
+        """Restore session context variables to their pre-handler values."""
+        from gateway.session_context import clear_session_vars
+        clear_session_vars(tokens)
     
     async def _enrich_message_with_vision(
         self,
@@ -6036,16 +6294,14 @@ class GatewayRunner:
                 return f"{disabled_note}\n\n{user_text}"
             return disabled_note
 
-        from tools.transcription_tools import transcribe_audio, get_stt_model_from_config
+        from tools.transcription_tools import transcribe_audio
         import asyncio
 
-        stt_model = get_stt_model_from_config()
-
         enriched_parts = []
         for path in audio_paths:
             try:
                 logger.debug("Transcribing user voice: %s", path)
-                result = await asyncio.to_thread(transcribe_audio, path, model=stt_model)
+                result = await asyncio.to_thread(transcribe_audio, path)
                 if result["success"]:
                     transcript = result["transcript"]
                     enriched_parts.append(
@@ -6277,6 +6533,32 @@ class GatewayRunner:
         )
         return hashlib.sha256(blob.encode()).hexdigest()[:16]
 
+    def _apply_session_model_override(
+        self, session_key: str, model: str, runtime_kwargs: dict
+    ) -> tuple:
+        """Apply /model session overrides if present, returning (model, runtime_kwargs).
+
+        The gateway /model command stores per-session overrides in
+        ``_session_model_overrides``.  These must take precedence over
+        config.yaml defaults so the switched model is actually used for
+        subsequent messages.  Fields with ``None`` values are skipped so
+        partial overrides don't clobber valid config defaults.
+        """
+        override = self._session_model_overrides.get(session_key)
+        if not override:
+            return model, runtime_kwargs
+        model = override.get("model", model)
+        for key in ("provider", "api_key", "base_url", "api_mode"):
+            val = override.get(key)
+            if val is not None:
+                runtime_kwargs[key] = val
+        return model, runtime_kwargs
+
+    def _is_intentional_model_switch(self, session_key: str, agent_model: str) -> bool:
+        """Return True if *agent_model* matches an active /model session override."""
+        override = self._session_model_overrides.get(session_key)
+        return override is not None and override.get("model") == agent_model
+
     def _evict_cached_agent(self, session_key: str) -> None:
         """Remove a cached agent for a session (called on /new, /model, etc)."""
         _lock = getattr(self, "_agent_cache_lock", None)
@@ -6654,9 +6936,15 @@ class GatewayRunner:
                     "tools": [],
                 }
 
+            # /model overrides take precedence over config.yaml defaults.
+            model, runtime_kwargs = self._apply_session_model_override(
+                session_key, model, runtime_kwargs
+            )
+
             pr = self._provider_routing
             reasoning_config = self._load_reasoning_config()
             self._reasoning_config = reasoning_config
+            self._service_tier = self._load_service_tier()
             # Set up streaming consumer if enabled
             _stream_consumer = None
             _stream_delta_cb = None
@@ -6719,6 +7007,8 @@ class GatewayRunner:
                     ephemeral_system_prompt=combined_ephemeral or None,
                     prefill_messages=self._prefill_messages or None,
                     reasoning_config=reasoning_config,
+                    service_tier=self._service_tier,
+                    request_overrides=turn_route.get("request_overrides"),
                     providers_allowed=pr.get("only"),
                     providers_ignored=pr.get("ignore"),
                     providers_order=pr.get("order"),
@@ -6743,6 +7033,8 @@ class GatewayRunner:
             agent.stream_delta_callback = _stream_delta_cb
             agent.status_callback = _status_callback_sync
             agent.reasoning_config = reasoning_config
+            agent.service_tier = self._service_tier
+            agent.request_overrides = turn_route.get("request_overrides")
 
             # Background review delivery — send "💾 Memory updated" etc. to user
             def _bg_review_send(message: str) -> None:
@@ -7273,7 +7565,7 @@ class GatewayRunner:
             _agent = agent_holder[0]
             if _agent is not None and hasattr(_agent, 'model'):
                 _cfg_model = _resolve_gateway_model()
-                if _agent.model != _cfg_model:
+                if _agent.model != _cfg_model and not self._is_intentional_model_switch(session_key, _agent.model):
                     # Fallback activated — evict cached agent so the next
                     # message starts fresh and retries the primary model.
                     self._evict_cached_agent(session_key)
@@ -7484,7 +7776,7 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
     # setups (each profile using a distinct HERMES_HOME) will naturally
     # allow concurrent instances without tripping this guard.
     import time as _time
-    from gateway.status import get_running_pid, remove_pid_file
+    from gateway.status import get_running_pid, remove_pid_file, terminate_pid
     existing_pid = get_running_pid()
     if existing_pid is not None and existing_pid != os.getpid():
         if replace:
@@ -7493,10 +7785,10 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
                 existing_pid,
             )
             try:
-                os.kill(existing_pid, signal.SIGTERM)
+                terminate_pid(existing_pid, force=False)
             except ProcessLookupError:
                 pass  # Already gone
-            except PermissionError:
+            except (PermissionError, OSError):
                 logger.error(
                     "Permission denied killing PID %d. Cannot replace.",
                     existing_pid,
@@ -7516,9 +7808,9 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
                     existing_pid,
                 )
                 try:
-                    os.kill(existing_pid, signal.SIGKILL)
+                    terminate_pid(existing_pid, force=True)
                     _time.sleep(0.5)
-                except (ProcessLookupError, PermissionError):
+                except (ProcessLookupError, PermissionError, OSError):
                     pass
             remove_pid_file()
             # Also release all scoped locks left by the old process.

gateway/session.py

@@ -753,41 +753,6 @@ class SessionStore:
             except Exception as e:
                 print(f"[gateway] Warning: Failed to create SQLite session: {e}")
 
-        # Seed new DM thread sessions with parent DM session history.
-        # When a bot reply creates a Slack thread and the user responds in it,
-        # the thread gets a new session (keyed by thread_ts).  Without seeding,
-        # the thread session starts with zero context — the user's original
-        # question and the bot's answer are invisible.  Fix: copy the parent
-        # DM session's transcript into the new thread session so context carries
-        # over while still keeping threads isolated from each other.
-        if (
-            source.chat_type == "dm"
-            and source.thread_id
-            and entry.created_at == entry.updated_at  # brand-new session
-            and not was_auto_reset
-        ):
-            parent_source = SessionSource(
-                platform=source.platform,
-                chat_id=source.chat_id,
-                chat_type="dm",
-                user_id=source.user_id,
-                # no thread_id — this is the parent DM session
-            )
-            parent_key = self._generate_session_key(parent_source)
-            with self._lock:
-                parent_entry = self._entries.get(parent_key)
-            if parent_entry and parent_entry.session_id != entry.session_id:
-                try:
-                    parent_history = self.load_transcript(parent_entry.session_id)
-                    if parent_history:
-                        self.rewrite_transcript(entry.session_id, parent_history)
-                        logger.info(
-                            "[Session] Seeded DM thread session %s with %d messages from parent %s",
-                            entry.session_id, len(parent_history), parent_entry.session_id,
-                        )
-                except Exception as e:
-                    logger.warning("[Session] Failed to seed thread session: %s", e)
-
         return entry
 
     def update_session(

gateway/session_context.py

@@ -0,0 +1,113 @@
+"""
+Session-scoped context variables for the Hermes gateway.
+
+Replaces the previous ``os.environ``-based session state
+(``HERMES_SESSION_PLATFORM``, ``HERMES_SESSION_CHAT_ID``, etc.) with
+Python's ``contextvars.ContextVar``.
+
+**Why this matters**
+
+The gateway processes messages concurrently via ``asyncio``.  When two
+messages arrive at the same time the old code did:
+
+    os.environ["HERMES_SESSION_THREAD_ID"] = str(context.source.thread_id)
+
+Because ``os.environ`` is *process-global*, Message A's value was
+silently overwritten by Message B before Message A's agent finished
+running.  Background-task notifications and tool calls therefore routed
+to the wrong thread.
+
+``contextvars.ContextVar`` values are *task-local*: each ``asyncio``
+task (and any ``run_in_executor`` thread it spawns) gets its own copy,
+so concurrent messages never interfere.
+
+**Backward compatibility**
+
+The public helper ``get_session_env(name, default="")`` mirrors the old
+``os.getenv("HERMES_SESSION_*", ...)`` calls.  Existing tool code only
+needs to replace the import + call site:
+
+    # before
+    import os
+    platform = os.getenv("HERMES_SESSION_PLATFORM", "")
+
+    # after
+    from gateway.session_context import get_session_env
+    platform = get_session_env("HERMES_SESSION_PLATFORM", "")
+"""
+
+from contextvars import ContextVar
+
+# ---------------------------------------------------------------------------
+# Per-task session variables
+# ---------------------------------------------------------------------------
+
+_SESSION_PLATFORM: ContextVar[str] = ContextVar("HERMES_SESSION_PLATFORM", default="")
+_SESSION_CHAT_ID: ContextVar[str] = ContextVar("HERMES_SESSION_CHAT_ID", default="")
+_SESSION_CHAT_NAME: ContextVar[str] = ContextVar("HERMES_SESSION_CHAT_NAME", default="")
+_SESSION_THREAD_ID: ContextVar[str] = ContextVar("HERMES_SESSION_THREAD_ID", default="")
+
+_VAR_MAP = {
+    "HERMES_SESSION_PLATFORM": _SESSION_PLATFORM,
+    "HERMES_SESSION_CHAT_ID": _SESSION_CHAT_ID,
+    "HERMES_SESSION_CHAT_NAME": _SESSION_CHAT_NAME,
+    "HERMES_SESSION_THREAD_ID": _SESSION_THREAD_ID,
+}
+
+
+def set_session_vars(
+    platform: str = "",
+    chat_id: str = "",
+    chat_name: str = "",
+    thread_id: str = "",
+) -> list:
+    """Set all session context variables and return reset tokens.
+
+    Call ``clear_session_vars(tokens)`` in a ``finally`` block to restore
+    the previous values when the handler exits.
+
+    Returns a list of ``Token`` objects (one per variable) that can be
+    passed to ``clear_session_vars``.
+    """
+    tokens = [
+        _SESSION_PLATFORM.set(platform),
+        _SESSION_CHAT_ID.set(chat_id),
+        _SESSION_CHAT_NAME.set(chat_name),
+        _SESSION_THREAD_ID.set(thread_id),
+    ]
+    return tokens
+
+
+def clear_session_vars(tokens: list) -> None:
+    """Restore session context variables to their pre-handler values."""
+    if not tokens:
+        return
+    vars_in_order = [
+        _SESSION_PLATFORM,
+        _SESSION_CHAT_ID,
+        _SESSION_CHAT_NAME,
+        _SESSION_THREAD_ID,
+    ]
+    for var, token in zip(vars_in_order, tokens):
+        var.reset(token)
+
+
+def get_session_env(name: str, default: str = "") -> str:
+    """Read a session context variable by its legacy ``HERMES_SESSION_*`` name.
+
+    Drop-in replacement for ``os.getenv("HERMES_SESSION_*", default)``.
+
+    Resolution order:
+    1. Context variable (set by the gateway for concurrency-safe access)
+    2. ``os.environ`` (used by CLI, cron scheduler, and tests)
+    3. *default*
+    """
+    import os
+
+    var = _VAR_MAP.get(name)
+    if var is not None:
+        value = var.get()
+        if value:
+            return value
+    # Fall back to os.environ for CLI, cron, and test compatibility
+    return os.getenv(name, default)

gateway/status.py

@@ -14,6 +14,8 @@ concurrently under distinct configurations).
 import hashlib
 import json
 import os
+import signal
+import subprocess
 import sys
 from datetime import datetime, timezone
 from pathlib import Path
@@ -23,6 +25,7 @@ from typing import Any, Optional
 _GATEWAY_KIND = "hermes-gateway"
 _RUNTIME_STATUS_FILE = "gateway_state.json"
 _LOCKS_DIRNAME = "gateway-locks"
+_IS_WINDOWS = sys.platform == "win32"
 
 
 def _get_pid_path() -> Path:
@@ -49,6 +52,33 @@ def _utc_now_iso() -> str:
     return datetime.now(timezone.utc).isoformat()
 
 
+def terminate_pid(pid: int, *, force: bool = False) -> None:
+    """Terminate a PID with platform-appropriate force semantics.
+
+    POSIX uses SIGTERM/SIGKILL. Windows uses taskkill /T /F for true force-kill
+    because os.kill(..., SIGTERM) is not equivalent to a tree-killing hard stop.
+    """
+    if force and _IS_WINDOWS:
+        try:
+            result = subprocess.run(
+                ["taskkill", "/PID", str(pid), "/T", "/F"],
+                capture_output=True,
+                text=True,
+                timeout=10,
+            )
+        except FileNotFoundError:
+            os.kill(pid, signal.SIGTERM)
+            return
+
+        if result.returncode != 0:
+            details = (result.stderr or result.stdout or "").strip()
+            raise OSError(details or f"taskkill failed for PID {pid}")
+        return
+
+    sig = signal.SIGTERM if not force else getattr(signal, "SIGKILL", signal.SIGTERM)
+    os.kill(pid, sig)
+
+
 def _scope_hash(identity: str) -> str:
     return hashlib.sha256(identity.encode("utf-8")).hexdigest()[:16]
 

gateway/stream_consumer.py

@@ -205,11 +205,20 @@ class GatewayStreamConsumer:
                             await self._send_or_edit(self._accumulated)
                     return
 
-                # Tool boundary: the should_edit block above already flushed
-                # accumulated text without a cursor.  Reset state so the next
-                # text chunk creates a fresh message below any tool-progress
-                # messages the gateway sent in between.
-                if got_segment_break:
+                # Tool boundary: reset message state so the next text chunk
+                # creates a fresh message below any tool-progress messages.
+                #
+                # Exception: when _message_id is "__no_edit__" the platform
+                # never returned a real message ID (e.g. Signal, webhook with
+                # github_comment delivery).  Resetting to None would re-enter
+                # the "first send" path on every tool boundary and post one
+                # platform message per tool call — that is what caused 155
+                # comments under a single PR.  Instead, keep all state so the
+                # full continuation is delivered once via _send_fallback_final.
+                # (When editing fails mid-stream due to flood control the id is
+                # a real string like "msg_1", not "__no_edit__", so that case
+                # still resets and creates a fresh segment as intended.)
+                if got_segment_break and self._message_id != "__no_edit__":
                     self._message_id = None
                     self._accumulated = ""
                     self._last_sent_text = ""

hermes_cli/auth.py

@@ -198,6 +198,14 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         api_key_env_vars=("DEEPSEEK_API_KEY",),
         base_url_env_var="DEEPSEEK_BASE_URL",
     ),
+    "xai": ProviderConfig(
+        id="xai",
+        name="xAI",
+        auth_type="api_key",
+        inference_base_url="https://api.x.ai/v1",
+        api_key_env_vars=("XAI_API_KEY",),
+        base_url_env_var="XAI_BASE_URL",
+    ),
     "ai-gateway": ProviderConfig(
         id="ai-gateway",
         name="AI Gateway",
@@ -704,6 +712,27 @@ def write_credential_pool(provider_id: str, entries: List[Dict[str, Any]]) -> Pa
         return _save_auth_store(auth_store)
 
 
+def suppress_credential_source(provider_id: str, source: str) -> None:
+    """Mark a credential source as suppressed so it won't be re-seeded."""
+    with _auth_store_lock():
+        auth_store = _load_auth_store()
+        suppressed = auth_store.setdefault("suppressed_sources", {})
+        provider_list = suppressed.setdefault(provider_id, [])
+        if source not in provider_list:
+            provider_list.append(source)
+        _save_auth_store(auth_store)
+
+
+def is_source_suppressed(provider_id: str, source: str) -> bool:
+    """Check if a credential source has been suppressed by the user."""
+    try:
+        auth_store = _load_auth_store()
+        suppressed = auth_store.get("suppressed_sources", {})
+        return source in suppressed.get(provider_id, [])
+    except Exception:
+        return False
+
+
 def get_provider_auth_state(provider_id: str) -> Optional[Dict[str, Any]]:
     """Return persisted auth state for a provider, or None."""
     auth_store = _load_auth_store()
@@ -716,6 +745,57 @@ def get_active_provider() -> Optional[str]:
     return auth_store.get("active_provider")
 
 
+def is_provider_explicitly_configured(provider_id: str) -> bool:
+    """Return True only if the user has explicitly configured this provider.
+
+    Checks:
+      1. active_provider in auth.json matches
+      2. model.provider in config.yaml matches
+      3. Provider-specific env vars are set (e.g. ANTHROPIC_API_KEY)
+
+    This is used to gate auto-discovery of external credentials (e.g.
+    Claude Code's ~/.claude/.credentials.json) so they are never used
+    without the user's explicit choice.  See PR #4210 for the same
+    pattern applied to the setup wizard gate.
+    """
+    normalized = (provider_id or "").strip().lower()
+
+    # 1. Check auth.json active_provider
+    try:
+        auth_store = _load_auth_store()
+        active = (auth_store.get("active_provider") or "").strip().lower()
+        if active and active == normalized:
+            return True
+    except Exception:
+        pass
+
+    # 2. Check config.yaml model.provider
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config()
+        model_cfg = cfg.get("model")
+        if isinstance(model_cfg, dict):
+            cfg_provider = (model_cfg.get("provider") or "").strip().lower()
+            if cfg_provider == normalized:
+                return True
+    except Exception:
+        pass
+
+    # 3. Check provider-specific env vars
+    # Exclude CLAUDE_CODE_OAUTH_TOKEN — it's set by Claude Code itself,
+    # not by the user explicitly configuring anthropic in Hermes.
+    _IMPLICIT_ENV_VARS = {"CLAUDE_CODE_OAUTH_TOKEN"}
+    pconfig = PROVIDER_REGISTRY.get(normalized)
+    if pconfig and pconfig.auth_type == "api_key":
+        for env_var in pconfig.api_key_env_vars:
+            if env_var in _IMPLICIT_ENV_VARS:
+                continue
+            if has_usable_secret(os.getenv(env_var, "")):
+                return True
+
+    return False
+
+
 def clear_provider_auth(provider_id: Optional[str] = None) -> bool:
     """
     Clear auth state for a provider. Used by `hermes logout`.
@@ -818,7 +898,7 @@ def resolve_provider(
     _PROVIDER_ALIASES = {
         "glm": "zai", "z-ai": "zai", "z.ai": "zai", "zhipu": "zai",
         "google": "gemini", "google-gemini": "gemini", "google-ai-studio": "gemini",
-        "kimi": "kimi-coding", "moonshot": "kimi-coding",
+        "kimi": "kimi-coding", "kimi-for-coding": "kimi-coding", "moonshot": "kimi-coding",
         "minimax-china": "minimax-cn", "minimax_cn": "minimax-cn",
         "claude": "anthropic", "claude-code": "anthropic",
         "github": "copilot", "github-copilot": "copilot",
@@ -1441,7 +1521,15 @@ def _resolve_verify(
     if effective_insecure:
         return False
     if effective_ca:
-        return str(effective_ca)
+        ca_path = str(effective_ca)
+        if not os.path.isfile(ca_path):
+            import logging
+            logging.getLogger("hermes.auth").warning(
+                "CA bundle path does not exist: %s — falling back to default certificates",
+                ca_path,
+            )
+            return True
+        return ca_path
     return True
 
 
@@ -2544,6 +2632,8 @@ def _prompt_model_selection(
             title=effective_title,
         )
         idx = menu.show()
+        from hermes_cli.curses_ui import flush_stdin
+        flush_stdin()
         if idx is None:
             return None
         print()
@@ -2553,7 +2643,7 @@ def _prompt_model_selection(
             custom = input("Enter model name: ").strip()
             return custom if custom else None
         return None
-    except (ImportError, NotImplementedError):
+    except (ImportError, NotImplementedError, OSError, subprocess.SubprocessError):
         pass
 
     # Fallback: numbered list

hermes_cli/auth_commands.py

@@ -347,8 +347,11 @@ def auth_remove_command(args) -> None:
             print("Cleared Hermes Anthropic OAuth credentials")
 
     elif removed.source == "claude_code" and provider == "anthropic":
-        print("Note: Claude Code credentials live in ~/.claude/.credentials.json")
-        print("      Remove them manually if you want to deauthorize Claude Code.")
+        from hermes_cli.auth import suppress_credential_source
+        suppress_credential_source(provider, "claude_code")
+        print("Suppressed claude_code credential — it will not be re-seeded.")
+        print("Note: Claude Code credentials still live in ~/.claude/.credentials.json")
+        print("Run `hermes auth add anthropic` to re-enable if needed.")
 
 
 def auth_reset_command(args) -> None:

hermes_cli/commands.py

@@ -16,8 +16,18 @@ from collections.abc import Callable, Mapping
 from dataclasses import dataclass
 from typing import Any
 
-from prompt_toolkit.auto_suggest import AutoSuggest, Suggestion
-from prompt_toolkit.completion import Completer, Completion
+# prompt_toolkit is an optional CLI dependency — only needed for
+# SlashCommandCompleter and SlashCommandAutoSuggest.  Gateway and test
+# environments that lack it must still be able to import this module
+# for resolve_command, gateway_help_lines, and COMMAND_REGISTRY.
+try:
+    from prompt_toolkit.auto_suggest import AutoSuggest, Suggestion
+    from prompt_toolkit.completion import Completer, Completion
+except ImportError:  # pragma: no cover
+    AutoSuggest = object  # type: ignore[assignment,misc]
+    Completer = object    # type: ignore[assignment,misc]
+    Suggestion = None     # type: ignore[assignment]
+    Completion = None     # type: ignore[assignment]
 
 
 # ---------------------------------------------------------------------------
@@ -73,8 +83,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
                args_hint="<question>"),
     CommandDef("queue", "Queue a prompt for the next turn (doesn't interrupt)", "Session",
                aliases=("q",), args_hint="<prompt>"),
-    CommandDef("status", "Show session info", "Session",
-               gateway_only=True),
+    CommandDef("status", "Show session info", "Session"),
     CommandDef("profile", "Show active profile name and home directory", "Info"),
     CommandDef("sethome", "Set this chat as the home channel", "Session",
                gateway_only=True, aliases=("set-home",)),
@@ -100,6 +109,9 @@ COMMAND_REGISTRY: list[CommandDef] = [
     CommandDef("reasoning", "Manage reasoning effort and display", "Configuration",
                args_hint="[level|show|hide]",
                subcommands=("none", "minimal", "low", "medium", "high", "xhigh", "show", "hide", "on", "off")),
+    CommandDef("fast", "Toggle fast mode — OpenAI Priority Processing / Anthropic Fast Mode (Normal/Fast)", "Configuration",
+               args_hint="[normal|fast|status]",
+               subcommands=("normal", "fast", "status", "on", "off")),
     CommandDef("skin", "Show or change the display skin/theme", "Configuration",
                cli_only=True, args_hint="[name]"),
     CommandDef("voice", "Toggle voice mode", "Configuration",
@@ -135,6 +147,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
                cli_only=True, aliases=("gateway",)),
     CommandDef("paste", "Check clipboard for an image and attach it", "Info",
                cli_only=True),
+    CommandDef("image", "Attach a local image file for your next prompt", "Info",
+               cli_only=True, args_hint="<path>"),
     CommandDef("update", "Update Hermes Agent to the latest version", "Info",
                gateway_only=True),
 
@@ -631,8 +645,18 @@ class SlashCommandCompleter(Completer):
     def __init__(
         self,
         skill_commands_provider: Callable[[], Mapping[str, dict[str, Any]]] | None = None,
+        command_filter: Callable[[str], bool] | None = None,
     ) -> None:
         self._skill_commands_provider = skill_commands_provider
+        self._command_filter = command_filter
+
+    def _command_allowed(self, slash_command: str) -> bool:
+        if self._command_filter is None:
+            return True
+        try:
+            return bool(self._command_filter(slash_command))
+        except Exception:
+            return True
 
     def _iter_skill_commands(self) -> Mapping[str, dict[str, Any]]:
         if self._skill_commands_provider is None:
@@ -910,7 +934,7 @@ class SlashCommandCompleter(Completer):
                 return
 
             # Static subcommand completions
-            if " " not in sub_text and base_cmd in SUBCOMMANDS:
+            if " " not in sub_text and base_cmd in SUBCOMMANDS and self._command_allowed(base_cmd):
                 for sub in SUBCOMMANDS[base_cmd]:
                     if sub.startswith(sub_lower) and sub != sub_lower:
                         yield Completion(
@@ -923,6 +947,8 @@ class SlashCommandCompleter(Completer):
         word = text[1:]
 
         for cmd, desc in COMMANDS.items():
+            if not self._command_allowed(cmd):
+                continue
             cmd_name = cmd[1:]
             if cmd_name.startswith(word):
                 yield Completion(
@@ -981,6 +1007,8 @@ class SlashCommandAutoSuggest(AutoSuggest):
             # Still typing the command name: /upd → suggest "ate"
             word = text[1:].lower()
             for cmd in COMMANDS:
+                if self._completer is not None and not self._completer._command_allowed(cmd):
+                    continue
                 cmd_name = cmd[1:]  # strip leading /
                 if cmd_name.startswith(word) and cmd_name != word:
                     return Suggestion(cmd_name[len(word):])
@@ -991,6 +1019,8 @@ class SlashCommandAutoSuggest(AutoSuggest):
         sub_lower = sub_text.lower()
 
         # Static subcommands
+        if self._completer is not None and not self._completer._command_allowed(base_cmd):
+            return None
         if base_cmd in SUBCOMMANDS and SUBCOMMANDS[base_cmd]:
             if " " not in sub_text:
                 for sub in SUBCOMMANDS[base_cmd]:

hermes_cli/config.py

@@ -39,6 +39,9 @@ _EXTRA_ENV_KEYS = frozenset({
     "DINGTALK_CLIENT_ID", "DINGTALK_CLIENT_SECRET",
     "FEISHU_APP_ID", "FEISHU_APP_SECRET", "FEISHU_ENCRYPT_KEY", "FEISHU_VERIFICATION_TOKEN",
     "WECOM_BOT_ID", "WECOM_SECRET",
+    "WEIXIN_ACCOUNT_ID", "WEIXIN_TOKEN", "WEIXIN_BASE_URL", "WEIXIN_CDN_BASE_URL",
+    "WEIXIN_HOME_CHANNEL", "WEIXIN_HOME_CHANNEL_NAME", "WEIXIN_DM_POLICY", "WEIXIN_GROUP_POLICY",
+    "WEIXIN_ALLOWED_USERS", "WEIXIN_GROUP_ALLOWED_USERS", "WEIXIN_ALLOW_ALL_USERS",
     "BLUEBUBBLES_SERVER_URL", "BLUEBUBBLES_PASSWORD",
     "TERMINAL_ENV", "TERMINAL_SSH_KEY", "TERMINAL_SSH_PORT",
     "WHATSAPP_MODE", "WHATSAPP_ENABLED",
@@ -158,16 +161,27 @@ def get_project_root() -> Path:
     return Path(__file__).parent.parent.resolve()
 
 def _secure_dir(path):
-    """Set directory to owner-only access (0700). No-op on Windows.
+    """Set directory to owner-only access (0700 by default). No-op on Windows.
 
     Skipped in managed mode — the NixOS module sets group-readable
     permissions (0750) so interactive users in the hermes group can
     share state with the gateway service.
+
+    The mode can be overridden via the HERMES_HOME_MODE environment variable
+    (e.g. HERMES_HOME_MODE=0701) for deployments where a web server (nginx,
+    caddy, etc.) needs to traverse HERMES_HOME to reach a served subdirectory.
+    The execute-only bit on a directory permits cd-through without exposing
+    directory listings.
     """
     if is_managed():
         return
     try:
-        os.chmod(path, 0o700)
+        mode_str = os.environ.get("HERMES_HOME_MODE", "").strip()
+        mode = int(mode_str, 8) if mode_str else 0o700
+    except ValueError:
+        mode = 0o700
+    try:
+        os.chmod(path, mode)
     except (OSError, NotImplementedError):
         pass
 
@@ -255,6 +269,7 @@ DEFAULT_CONFIG = {
         # tools or receiving API responses.  Only fires when the agent has
         # been completely idle for this duration.  0 = unlimited.
         "gateway_timeout": 1800,
+        "service_tier": "",
         # Tool-use enforcement: injects system prompt guidance that tells the
         # model to actually call tools instead of describing intended actions.
         # Values: "auto" (default — applies to gpt/codex models), true/false
@@ -540,6 +555,7 @@ DEFAULT_CONFIG = {
     "discord": {
         "require_mention": True,       # Require @mention to respond in server channels
         "free_response_channels": "",  # Comma-separated channel IDs where bot responds without mention
+        "allowed_channels": "",        # If set, bot ONLY responds in these channel IDs (whitelist)
         "auto_thread": True,           # Auto-create threads on @mention in channels (like Slack)
         "reactions": True,             # Add 👀/✅/❌ reactions to messages during processing
     },
@@ -599,7 +615,7 @@ DEFAULT_CONFIG = {
     },
 
     # Config schema version - bump this when adding new required fields
-    "_config_version": 13,
+    "_config_version": 14,
 }
 
 # =============================================================================
@@ -1193,8 +1209,8 @@ OPTIONAL_ENV_VARS = {
         "advanced": True,
     },
     "API_SERVER_KEY": {
-        "description": "Bearer token for API server authentication. If empty, all requests are allowed (local use only).",
-        "prompt": "API server auth key (optional)",
+        "description": "Bearer token for API server authentication. Required for non-loopback binding; server refuses to start without it. On loopback (127.0.0.1), all requests are allowed if empty.",
+        "prompt": "API server auth key (required for network access)",
         "url": None,
         "password": True,
         "category": "messaging",
@@ -1209,13 +1225,21 @@ OPTIONAL_ENV_VARS = {
         "advanced": True,
     },
     "API_SERVER_HOST": {
-        "description": "Host/bind address for the API server (default: 127.0.0.1). Use 0.0.0.0 for network access — requires API_SERVER_KEY for security.",
+        "description": "Host/bind address for the API server (default: 127.0.0.1). Use 0.0.0.0 for network access — server refuses to start without API_SERVER_KEY.",
         "prompt": "API server host",
         "url": None,
         "password": False,
         "category": "messaging",
         "advanced": True,
     },
+    "API_SERVER_MODEL_NAME": {
+        "description": "Model name advertised on /v1/models. Defaults to the profile name (or 'hermes-agent' for the default profile). Useful for multi-user setups with OpenWebUI.",
+        "prompt": "API server model name",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+        "advanced": True,
+    },
     "WEBHOOK_ENABLED": {
         "description": "Enable the webhook platform adapter for receiving events from GitHub, GitLab, etc.",
         "prompt": "Enable webhooks (true/false)",
@@ -1746,6 +1770,56 @@ def migrate_config(interactive: bool = True, quiet: bool = False) -> Dict[str, A
             except Exception:
                 pass
 
+    # ── Version 13 → 14: migrate legacy flat stt.model to provider section ──
+    # Old configs (and cli-config.yaml.example) had a flat `stt.model` key
+    # that was provider-agnostic.  When the provider was "local" this caused
+    # OpenAI model names (e.g. "whisper-1") to be fed to faster-whisper,
+    # crashing with "Invalid model size".  Move the value into the correct
+    # provider-specific section and remove the flat key.
+    if current_ver < 14:
+        # Read raw config (no defaults merged) to check what the user actually
+        # wrote, then apply changes to the merged config for saving.
+        raw = read_raw_config()
+        raw_stt = raw.get("stt", {})
+        if isinstance(raw_stt, dict) and "model" in raw_stt:
+            legacy_model = raw_stt["model"]
+            provider = raw_stt.get("provider", "local")
+            config = load_config()
+            stt = config.get("stt", {})
+            # Remove the legacy flat key
+            stt.pop("model", None)
+            # Place it in the appropriate provider section only if the
+            # user didn't already set a model there
+            if provider in ("local", "local_command"):
+                # Don't migrate an OpenAI model name into the local section
+                _local_models = {
+                    "tiny.en", "tiny", "base.en", "base", "small.en", "small",
+                    "medium.en", "medium", "large-v1", "large-v2", "large-v3",
+                    "large", "distil-large-v2", "distil-medium.en",
+                    "distil-small.en", "distil-large-v3", "distil-large-v3.5",
+                    "large-v3-turbo", "turbo",
+                }
+                if legacy_model in _local_models:
+                    # Check raw config — only set if user didn't already
+                    # have a nested local.model
+                    raw_local = raw_stt.get("local", {})
+                    if not isinstance(raw_local, dict) or "model" not in raw_local:
+                        local_cfg = stt.setdefault("local", {})
+                        local_cfg["model"] = legacy_model
+                # else: drop it — it was an OpenAI model name, local section
+                # already defaults to "base" via DEFAULT_CONFIG
+            else:
+                # Cloud provider — put it in that provider's section only
+                # if user didn't already set a nested model
+                raw_provider = raw_stt.get(provider, {})
+                if not isinstance(raw_provider, dict) or "model" not in raw_provider:
+                    provider_cfg = stt.setdefault(provider, {})
+                    provider_cfg["model"] = legacy_model
+            config["stt"] = stt
+            save_config(config)
+            if not quiet:
+                print(f"  ✓ Migrated legacy stt.model to provider-specific config")
+
     if current_ver < latest_ver and not quiet:
         print(f"Config version: {current_ver} → {latest_ver}")
     

hermes_cli/copilot_auth.py

@@ -273,6 +273,7 @@ def copilot_request_headers(
     headers: dict[str, str] = {
         "Editor-Version": "vscode/1.104.1",
         "User-Agent": "HermesAgent/1.0",
+        "Copilot-Integration-Id": "vscode-chat",
         "Openai-Intent": "conversation-edits",
         "x-initiator": "agent" if is_agent_turn else "user",
     }

hermes_cli/curses_ui.py

@@ -10,6 +10,28 @@ from typing import Callable, List, Optional, Set
 from hermes_cli.colors import Colors, color
 
 
+def flush_stdin() -> None:
+    """Flush any stray bytes from the stdin input buffer.
+
+    Must be called after ``curses.wrapper()`` (or any terminal-mode library
+    like simple_term_menu) returns, **before** the next ``input()`` /
+    ``getpass.getpass()`` call.  ``curses.endwin()`` restores the terminal
+    but does NOT drain the OS input buffer — leftover escape-sequence bytes
+    (from arrow keys, terminal mode-switch responses, or rapid keypresses)
+    remain buffered and silently get consumed by the next ``input()`` call,
+    corrupting user data (e.g. writing ``^[^[`` into .env files).
+
+    On non-TTY stdin (piped, redirected) or Windows, this is a no-op.
+    """
+    try:
+        if not sys.stdin.isatty():
+            return
+        import termios
+        termios.tcflush(sys.stdin, termios.TCIFLUSH)
+    except Exception:
+        pass
+
+
 def curses_checklist(
     title: str,
     items: List[str],
@@ -131,6 +153,7 @@ def curses_checklist(
                     return
 
         curses.wrapper(_draw)
+        flush_stdin()
         return result_holder[0] if result_holder[0] is not None else cancel_returns
 
     except Exception:

hermes_cli/doctor.py

@@ -54,6 +54,32 @@ _PROVIDER_ENV_HINTS = (
 )
 
 
+from hermes_constants import is_termux as _is_termux
+
+
+def _python_install_cmd() -> str:
+    return "python -m pip install" if _is_termux() else "uv pip install"
+
+
+def _system_package_install_cmd(pkg: str) -> str:
+    if _is_termux():
+        return f"pkg install {pkg}"
+    if sys.platform == "darwin":
+        return f"brew install {pkg}"
+    return f"sudo apt install {pkg}"
+
+
+def _termux_browser_setup_steps(node_installed: bool) -> list[str]:
+    steps: list[str] = []
+    step = 1
+    if not node_installed:
+        steps.append(f"{step}) pkg install nodejs")
+        step += 1
+    steps.append(f"{step}) npm install -g agent-browser")
+    steps.append(f"{step + 1}) agent-browser install")
+    return steps
+
+
 def _has_provider_env_config(content: str) -> bool:
     """Return True when ~/.hermes/.env contains provider auth/base URL settings."""
     return any(key in content for key in _PROVIDER_ENV_HINTS)
@@ -200,7 +226,7 @@ def run_doctor(args):
             check_ok(name)
         except ImportError:
             check_fail(name, "(missing)")
-            issues.append(f"Install {name}: uv pip install {module}")
+            issues.append(f"Install {name}: {_python_install_cmd()} {module}")
     
     for module, name in optional_packages:
         try:
@@ -503,7 +529,7 @@ def run_doctor(args):
         check_ok("ripgrep (rg)", "(faster file search)")
     else:
         check_warn("ripgrep (rg) not found", "(file search uses grep fallback)")
-        check_info("Install for faster search: sudo apt install ripgrep")
+        check_info(f"Install for faster search: {_system_package_install_cmd('ripgrep')}")
     
     # Docker (optional)
     terminal_env = os.getenv("TERMINAL_ENV", "local")
@@ -526,7 +552,10 @@ def run_doctor(args):
         if shutil.which("docker"):
             check_ok("docker", "(optional)")
         else:
-            check_warn("docker not found", "(optional)")
+            if _is_termux():
+                check_info("Docker backend is not available inside Termux (expected on Android)")
+            else:
+                check_warn("docker not found", "(optional)")
     
     # SSH (if using ssh backend)
     if terminal_env == "ssh":
@@ -574,9 +603,23 @@ def run_doctor(args):
         if agent_browser_path.exists():
             check_ok("agent-browser (Node.js)", "(browser automation)")
         else:
-            check_warn("agent-browser not installed", "(run: npm install)")
+            if _is_termux():
+                check_info("agent-browser is not installed (expected in the tested Termux path)")
+                check_info("Install it manually later with: npm install -g agent-browser && agent-browser install")
+                check_info("Termux browser setup:")
+                for step in _termux_browser_setup_steps(node_installed=True):
+                    check_info(step)
+            else:
+                check_warn("agent-browser not installed", "(run: npm install)")
     else:
-        check_warn("Node.js not found", "(optional, needed for browser tools)")
+        if _is_termux():
+            check_info("Node.js not found (browser tools are optional in the tested Termux path)")
+            check_info("Install Node.js on Termux with: pkg install nodejs")
+            check_info("Termux browser setup:")
+            for step in _termux_browser_setup_steps(node_installed=False):
+                check_info(step)
+        else:
+            check_warn("Node.js not found", "(optional, needed for browser tools)")
     
     # npm audit for all Node.js packages
     if shutil.which("npm"):
@@ -709,7 +752,7 @@ def run_doctor(args):
                 _url = (_base.rstrip("/") + "/models") if _base else _default_url
                 _headers = {"Authorization": f"Bearer {_key}"}
                 if "api.kimi.com" in _url.lower():
-                    _headers["User-Agent"] = "KimiCLI/1.0"
+                    _headers["User-Agent"] = "KimiCLI/1.30.0"
                 _resp = httpx.get(
                     _url,
                     headers=_headers,
@@ -739,8 +782,9 @@ def run_doctor(args):
                 __import__("tinker_atropos")
                 check_ok("tinker-atropos", "(RL training backend)")
             except ImportError:
-                check_warn("tinker-atropos found but not installed", "(run: uv pip install -e ./tinker-atropos)")
-                issues.append("Install tinker-atropos: uv pip install -e ./tinker-atropos")
+                install_cmd = f"{_python_install_cmd()} -e ./tinker-atropos"
+                check_warn("tinker-atropos found but not installed", f"(run: {install_cmd})")
+                issues.append(f"Install tinker-atropos: {install_cmd}")
         else:
             check_warn("tinker-atropos requires Python 3.11+", f"(current: {py_version.major}.{py_version.minor})")
     else:

hermes_cli/dump.py

@@ -119,6 +119,7 @@ def _configured_platforms() -> list[str]:
         "dingtalk": "DINGTALK_CLIENT_ID",
         "feishu": "FEISHU_APP_ID",
         "wecom": "WECOM_BOT_ID",
+        "weixin": "WEIXIN_ACCOUNT_ID",
     }
     return [name for name, env in checks.items() if os.getenv(env)]
 

hermes_cli/gateway.py

@@ -14,6 +14,7 @@ from pathlib import Path
 
 PROJECT_ROOT = Path(__file__).parent.parent.resolve()
 
+from gateway.status import terminate_pid
 from hermes_cli.config import get_env_value, get_hermes_home, save_env_value, is_managed, managed_error
 # display_hermes_home is imported lazily at call sites to avoid ImportError
 # when hermes_constants is cached from a pre-update version during `hermes update`.
@@ -39,7 +40,7 @@ def _get_service_pids() -> set:
     pids: set = set()
 
     # --- systemd (Linux): user and system scopes ---
-    if is_linux():
+    if supports_systemd_services():
         for scope_args in [["systemctl", "--user"], ["systemctl"]]:
             try:
                 result = subprocess.run(
@@ -162,7 +163,7 @@ def kill_gateway_processes(force: bool = False, exclude_pids: set | None = None)
     """Kill any running gateway processes. Returns count killed.
 
     Args:
-        force: Use SIGKILL instead of SIGTERM.
+        force: Use the platform's force-kill mechanism instead of graceful terminate.
         exclude_pids: PIDs to skip (e.g. service-managed PIDs that were just
             restarted and should not be killed).
     """
@@ -171,10 +172,7 @@ def kill_gateway_processes(force: bool = False, exclude_pids: set | None = None)
     
     for pid in pids:
         try:
-            if force and not is_windows():
-                os.kill(pid, signal.SIGKILL)
-            else:
-                os.kill(pid, signal.SIGTERM)
+            terminate_pid(pid, force=force)
             killed += 1
         except ProcessLookupError:
             # Process already gone
@@ -182,6 +180,8 @@ def kill_gateway_processes(force: bool = False, exclude_pids: set | None = None)
         except PermissionError:
             print(f"⚠ Permission denied to kill PID {pid}")
     
+        except OSError as exc:
+            print(f"Failed to kill PID {pid}: {exc}")
     return killed
 
 
@@ -225,6 +225,14 @@ def stop_profile_gateway() -> bool:
 def is_linux() -> bool:
     return sys.platform.startswith('linux')
 
+
+from hermes_constants import is_termux
+
+
+def supports_systemd_services() -> bool:
+    return is_linux() and not is_termux()
+
+
 def is_macos() -> bool:
     return sys.platform == 'darwin'
 
@@ -243,18 +251,18 @@ SERVICE_DESCRIPTION = "Hermes Agent Gateway - Messaging Platform Integration"
 def _profile_suffix() -> str:
     """Derive a service-name suffix from the current HERMES_HOME.
 
-    Returns ``""`` for the default ``~/.hermes``, the profile name for
-    ``~/.hermes/profiles/<name>``, or a short hash for any other custom
-    HERMES_HOME path.
+    Returns ``""`` for the default root, the profile name for
+    ``<root>/profiles/<name>``, or a short hash for any other path.
+    Works correctly in Docker (HERMES_HOME=/opt/data) and standard deployments.
     """
     import hashlib
     import re
-    from pathlib import Path as _Path
+    from hermes_constants import get_default_hermes_root
     home = get_hermes_home().resolve()
-    default = (_Path.home() / ".hermes").resolve()
+    default = get_default_hermes_root().resolve()
     if home == default:
         return ""
-    # Detect ~/.hermes/profiles/<name> pattern → use the profile name
+    # Detect <root>/profiles/<name> pattern → use the profile name
     profiles_root = (default / "profiles").resolve()
     try:
         rel = home.relative_to(profiles_root)
@@ -279,9 +287,9 @@ def _profile_arg(hermes_home: str | None = None) -> str:
             service definition for a different user (e.g. system service).
     """
     import re
-    from pathlib import Path as _Path
+    from hermes_constants import get_default_hermes_root
     home = Path(hermes_home or str(get_hermes_home())).resolve()
-    default = (_Path.home() / ".hermes").resolve()
+    default = get_default_hermes_root().resolve()
     if home == default:
         return ""
     profiles_root = (default / "profiles").resolve()
@@ -475,13 +483,15 @@ def install_linux_gateway_from_setup(force: bool = False) -> tuple[str | None, b
 
 
 def get_systemd_linger_status() -> tuple[bool | None, str]:
-    """Return whether systemd user lingering is enabled for the current user.
+    """Return systemd linger status for the current user.
 
     Returns:
         (True, "") when linger is enabled.
         (False, "") when linger is disabled.
         (None, detail) when the status could not be determined.
     """
+    if is_termux():
+        return None, "not supported in Termux"
     if not is_linux():
         return None, "not supported on this platform"
 
@@ -595,6 +605,24 @@ def _build_user_local_paths(home: Path, path_entries: list[str]) -> list[str]:
     return [p for p in candidates if p not in path_entries and Path(p).exists()]
 
 
+def _remap_path_for_user(path: str, target_home_dir: str) -> str:
+    """Remap *path* from the current user's home to *target_home_dir*.
+
+    If *path* lives under ``Path.home()`` the corresponding prefix is swapped
+    to *target_home_dir*; otherwise the path is returned unchanged.
+
+      /root/.hermes/hermes-agent  -> /home/alice/.hermes/hermes-agent
+      /opt/hermes                 -> /opt/hermes  (kept as-is)
+    """
+    current_home = Path.home().resolve()
+    resolved = Path(path).resolve()
+    try:
+        relative = resolved.relative_to(current_home)
+        return str(Path(target_home_dir) / relative)
+    except ValueError:
+        return str(resolved)
+
+
 def _hermes_home_for_target_user(target_home_dir: str) -> str:
     """Remap the current HERMES_HOME to the equivalent under a target user's home.
 
@@ -642,6 +670,15 @@ def generate_systemd_unit(system: bool = False, run_as_user: str | None = None)
         username, group_name, home_dir = _system_service_identity(run_as_user)
         hermes_home = _hermes_home_for_target_user(home_dir)
         profile_arg = _profile_arg(hermes_home)
+        # Remap all paths that may resolve under the calling user's home
+        # (e.g. /root/) to the target user's home so the service can
+        # actually access them.
+        python_path = _remap_path_for_user(python_path, home_dir)
+        working_dir = _remap_path_for_user(working_dir, home_dir)
+        venv_dir = _remap_path_for_user(venv_dir, home_dir)
+        venv_bin = _remap_path_for_user(venv_bin, home_dir)
+        node_bin = _remap_path_for_user(node_bin, home_dir)
+        path_entries = [_remap_path_for_user(p, home_dir) for p in path_entries]
         path_entries.extend(_build_user_local_paths(Path(home_dir), path_entries))
         path_entries.extend(common_bin_paths)
         sane_path = ":".join(path_entries)
@@ -753,7 +790,7 @@ def _print_linger_enable_warning(username: str, detail: str | None = None) -> No
 
 def _ensure_linger_enabled() -> None:
     """Enable linger when possible so the user gateway survives logout."""
-    if not is_linux():
+    if is_termux() or not is_linux():
         return
 
     import getpass
@@ -1159,7 +1196,19 @@ def launchd_start():
 
 def launchd_stop():
     label = get_launchd_label()
-    subprocess.run(["launchctl", "kill", "SIGTERM", f"{_launchd_domain()}/{label}"], check=True, timeout=30)
+    target = f"{_launchd_domain()}/{label}"
+    # bootout unloads the service definition so KeepAlive doesn't respawn
+    # the process.  A plain `kill SIGTERM` only signals the process — launchd
+    # immediately restarts it because KeepAlive.SuccessfulExit = false.
+    # `hermes gateway start` re-bootstraps when it detects the job is unloaded.
+    try:
+        subprocess.run(["launchctl", "bootout", target], check=True, timeout=90)
+    except subprocess.CalledProcessError as e:
+        if e.returncode in (3, 113):
+            pass  # Already unloaded — nothing to stop.
+        else:
+            raise
+    _wait_for_gateway_exit(timeout=10.0, force_after=5.0)
     print("✓ Service stopped")
 
 def _wait_for_gateway_exit(timeout: float = 10.0, force_after: float = 5.0):
@@ -1171,7 +1220,7 @@ def _wait_for_gateway_exit(timeout: float = 10.0, force_after: float = 5.0):
 
     Args:
         timeout: Total seconds to wait before giving up.
-        force_after: Seconds of graceful waiting before sending SIGKILL.
+        force_after: Seconds of graceful waiting before escalating to force-kill.
     """
     import time
     from gateway.status import get_running_pid
@@ -1188,15 +1237,15 @@ def _wait_for_gateway_exit(timeout: float = 10.0, force_after: float = 5.0):
         if not force_sent and time.monotonic() >= force_deadline:
             # Grace period expired — force-kill the specific PID.
             try:
-                os.kill(pid, signal.SIGKILL)
+                terminate_pid(pid, force=True)
                 print(f"⚠ Gateway PID {pid} did not exit gracefully; sent SIGKILL")
-            except (ProcessLookupError, PermissionError):
+            except (ProcessLookupError, PermissionError, OSError):
                 return  # Already gone or we can't touch it.
             force_sent = True
 
         time.sleep(0.3)
 
-    # Timed out even after SIGKILL.
+    # Timed out even after force-kill.
     remaining_pid = get_running_pid()
     if remaining_pid is not None:
         print(f"⚠ Gateway PID {remaining_pid} still running after {timeout}s — restart may fail")
@@ -1575,6 +1624,12 @@ _PLATFORMS = [
              "help": "Chat ID for scheduled results and notifications."},
         ],
     },
+    {
+        "key": "weixin",
+        "label": "Weixin / WeChat",
+        "emoji": "💬",
+        "token_var": "WEIXIN_ACCOUNT_ID",
+    },
     {
         "key": "bluebubbles",
         "label": "BlueBubbles (iMessage)",
@@ -1647,6 +1702,13 @@ def _platform_status(platform: dict) -> str:
         if val or password or homeserver:
             return "partially configured"
         return "not configured"
+    if platform.get("key") == "weixin":
+        token = get_env_value("WEIXIN_TOKEN")
+        if val and token:
+            return "configured"
+        if val or token:
+            return "partially configured"
+        return "not configured"
     if val:
         return "configured"
     return "not configured"
@@ -1750,7 +1812,7 @@ def _setup_standard_platform(platform: dict):
                     print_warning("  Open access enabled — anyone can use your bot!")
                 elif access_idx == 1:
                     print_success("  DM pairing mode — users will receive a code to request access.")
-                    print_info("  Approve with: hermes pairing approve {platform} {code}")
+                    print_info("  Approve with: hermes pairing approve <platform> <code>")
                 else:
                     print_info("  Skipped — configure later with 'hermes gateway setup'")
             continue
@@ -1788,7 +1850,7 @@ def _setup_whatsapp():
 
 def _is_service_installed() -> bool:
     """Check if the gateway is installed as a system service."""
-    if is_linux():
+    if supports_systemd_services():
         return get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()
     elif is_macos():
         return get_launchd_plist_path().exists()
@@ -1797,7 +1859,7 @@ def _is_service_installed() -> bool:
 
 def _is_service_running() -> bool:
     """Check if the gateway service is currently running."""
-    if is_linux():
+    if supports_systemd_services():
         user_unit_exists = get_systemd_unit_path(system=False).exists()
         system_unit_exists = get_systemd_unit_path(system=True).exists()
 
@@ -1837,6 +1899,133 @@ def _is_service_running() -> bool:
     return len(find_gateway_pids()) > 0
 
 
+def _setup_weixin():
+    """Interactive setup for Weixin / WeChat personal accounts."""
+    print()
+    print(color("  ─── 💬 Weixin / WeChat Setup ───", Colors.CYAN))
+    print()
+    print_info("  1. Hermes will open Tencent iLink QR login in this terminal.")
+    print_info("  2. Use WeChat to scan and confirm the QR code.")
+    print_info("  3. Hermes will store the returned account_id/token in ~/.hermes/.env.")
+    print_info("  4. This adapter supports native text, image, video, and document delivery.")
+
+    existing_account = get_env_value("WEIXIN_ACCOUNT_ID")
+    existing_token = get_env_value("WEIXIN_TOKEN")
+    if existing_account and existing_token:
+        print()
+        print_success("Weixin is already configured.")
+        if not prompt_yes_no("  Reconfigure Weixin?", False):
+            return
+
+    try:
+        from gateway.platforms.weixin import check_weixin_requirements, qr_login
+    except Exception as exc:
+        print_error(f"  Weixin adapter import failed: {exc}")
+        print_info("  Install gateway dependencies first, then retry.")
+        return
+
+    if not check_weixin_requirements():
+        print_error("  Missing dependencies: Weixin needs aiohttp and cryptography.")
+        print_info("  Install them, then rerun `hermes gateway setup`.")
+        return
+
+    print()
+    if not prompt_yes_no("  Start QR login now?", True):
+        print_info("  Cancelled.")
+        return
+
+    import asyncio
+    try:
+        credentials = asyncio.run(qr_login(str(get_hermes_home())))
+    except KeyboardInterrupt:
+        print()
+        print_warning("  Weixin setup cancelled.")
+        return
+    except Exception as exc:
+        print_error(f"  QR login failed: {exc}")
+        return
+
+    if not credentials:
+        print_warning("  QR login did not complete.")
+        return
+
+    account_id = credentials.get("account_id", "")
+    token = credentials.get("token", "")
+    base_url = credentials.get("base_url", "")
+    user_id = credentials.get("user_id", "")
+
+    save_env_value("WEIXIN_ACCOUNT_ID", account_id)
+    save_env_value("WEIXIN_TOKEN", token)
+    if base_url:
+        save_env_value("WEIXIN_BASE_URL", base_url)
+    save_env_value("WEIXIN_CDN_BASE_URL", get_env_value("WEIXIN_CDN_BASE_URL") or "https://novac2c.cdn.weixin.qq.com/c2c")
+
+    print()
+    access_choices = [
+        "Use DM pairing approval (recommended)",
+        "Allow all direct messages",
+        "Only allow listed user IDs",
+        "Disable direct messages",
+    ]
+    access_idx = prompt_choice("  How should direct messages be authorized?", access_choices, 0)
+    if access_idx == 0:
+        save_env_value("WEIXIN_DM_POLICY", "pairing")
+        save_env_value("WEIXIN_ALLOW_ALL_USERS", "false")
+        save_env_value("WEIXIN_ALLOWED_USERS", "")
+        print_success("  DM pairing enabled.")
+        print_info("  Unknown DM users can request access and you approve them with `hermes pairing approve`.")
+    elif access_idx == 1:
+        save_env_value("WEIXIN_DM_POLICY", "open")
+        save_env_value("WEIXIN_ALLOW_ALL_USERS", "true")
+        save_env_value("WEIXIN_ALLOWED_USERS", "")
+        print_warning("  Open DM access enabled for Weixin.")
+    elif access_idx == 2:
+        default_allow = user_id or ""
+        allowlist = prompt("  Allowed Weixin user IDs (comma-separated)", default_allow, password=False).replace(" ", "")
+        save_env_value("WEIXIN_DM_POLICY", "allowlist")
+        save_env_value("WEIXIN_ALLOW_ALL_USERS", "false")
+        save_env_value("WEIXIN_ALLOWED_USERS", allowlist)
+        print_success("  Weixin allowlist saved.")
+    else:
+        save_env_value("WEIXIN_DM_POLICY", "disabled")
+        save_env_value("WEIXIN_ALLOW_ALL_USERS", "false")
+        save_env_value("WEIXIN_ALLOWED_USERS", "")
+        print_warning("  Direct messages disabled.")
+
+    print()
+    group_choices = [
+        "Disable group chats (recommended)",
+        "Allow all group chats",
+        "Only allow listed group chat IDs",
+    ]
+    group_idx = prompt_choice("  How should group chats be handled?", group_choices, 0)
+    if group_idx == 0:
+        save_env_value("WEIXIN_GROUP_POLICY", "disabled")
+        save_env_value("WEIXIN_GROUP_ALLOWED_USERS", "")
+        print_info("  Group chats disabled.")
+    elif group_idx == 1:
+        save_env_value("WEIXIN_GROUP_POLICY", "open")
+        save_env_value("WEIXIN_GROUP_ALLOWED_USERS", "")
+        print_warning("  All group chats enabled.")
+    else:
+        allow_groups = prompt("  Allowed group chat IDs (comma-separated)", "", password=False).replace(" ", "")
+        save_env_value("WEIXIN_GROUP_POLICY", "allowlist")
+        save_env_value("WEIXIN_GROUP_ALLOWED_USERS", allow_groups)
+        print_success("  Group allowlist saved.")
+
+    if user_id:
+        print()
+        if prompt_yes_no(f"  Use your Weixin user ID ({user_id}) as the home channel?", True):
+            save_env_value("WEIXIN_HOME_CHANNEL", user_id)
+            print_success(f"  Home channel set to {user_id}")
+
+    print()
+    print_success("Weixin configured!")
+    print_info(f"  Account ID: {account_id}")
+    if user_id:
+        print_info(f"  User ID: {user_id}")
+
+
 def _setup_signal():
     """Interactive setup for Signal messenger."""
     import shutil
@@ -1970,7 +2159,7 @@ def gateway_setup():
     service_installed = _is_service_installed()
     service_running = _is_service_running()
 
-    if is_linux() and has_conflicting_systemd_units():
+    if supports_systemd_services() and has_conflicting_systemd_units():
         print_systemd_scope_conflict_warning()
         print()
 
@@ -1980,7 +2169,7 @@ def gateway_setup():
         print_warning("Gateway service is installed but not running.")
         if prompt_yes_no("  Start it now?", True):
             try:
-                if is_linux():
+                if supports_systemd_services():
                     systemd_start()
                 elif is_macos():
                     launchd_start()
@@ -2012,6 +2201,8 @@ def gateway_setup():
             _setup_whatsapp()
         elif platform["key"] == "signal":
             _setup_signal()
+        elif platform["key"] == "weixin":
+            _setup_weixin()
         else:
             _setup_standard_platform(platform)
 
@@ -2031,7 +2222,7 @@ def gateway_setup():
         if service_running:
             if prompt_yes_no("  Restart the gateway to pick up changes?", True):
                 try:
-                    if is_linux():
+                    if supports_systemd_services():
                         systemd_restart()
                     elif is_macos():
                         launchd_restart()
@@ -2043,7 +2234,7 @@ def gateway_setup():
         elif service_installed:
             if prompt_yes_no("  Start the gateway service?", True):
                 try:
-                    if is_linux():
+                    if supports_systemd_services():
                         systemd_start()
                     elif is_macos():
                         launchd_start()
@@ -2051,13 +2242,13 @@ def gateway_setup():
                     print_error(f"  Start failed: {e}")
         else:
             print()
-            if is_linux() or is_macos():
-                platform_name = "systemd" if is_linux() else "launchd"
+            if supports_systemd_services() or is_macos():
+                platform_name = "systemd" if supports_systemd_services() else "launchd"
                 if prompt_yes_no(f"  Install the gateway as a {platform_name} service? (runs in background, starts on boot)", True):
                     try:
                         installed_scope = None
                         did_install = False
-                        if is_linux():
+                        if supports_systemd_services():
                             installed_scope, did_install = install_linux_gateway_from_setup(force=False)
                         else:
                             launchd_install(force=False)
@@ -2065,7 +2256,7 @@ def gateway_setup():
                         print()
                         if did_install and prompt_yes_no("  Start the service now?", True):
                             try:
-                                if is_linux():
+                                if supports_systemd_services():
                                     systemd_start(system=installed_scope == "system")
                                 else:
                                     launchd_start()
@@ -2076,12 +2267,18 @@ def gateway_setup():
                         print_info("  You can try manually: hermes gateway install")
                 else:
                     print_info("  You can install later: hermes gateway install")
-                    if is_linux():
+                    if supports_systemd_services():
                         print_info("  Or as a boot-time service: sudo hermes gateway install --system")
                     print_info("  Or run in foreground:  hermes gateway")
             else:
-                print_info("  Service install not supported on this platform.")
-                print_info("  Run in foreground: hermes gateway")
+                if is_termux():
+                    from hermes_constants import display_hermes_home as _dhh
+                    print_info("  Termux does not use systemd/launchd services.")
+                    print_info("  Run in foreground: hermes gateway")
+                    print_info(f"  Or start it manually in the background (best effort): nohup hermes gateway >{_dhh()}/logs/gateway.log 2>&1 &")
+                else:
+                    print_info("  Service install not supported on this platform.")
+                    print_info("  Run in foreground: hermes gateway")
     else:
         print()
         print_info("No platforms configured. Run 'hermes gateway setup' when ready.")
@@ -2117,7 +2314,11 @@ def gateway_command(args):
         force = getattr(args, 'force', False)
         system = getattr(args, 'system', False)
         run_as_user = getattr(args, 'run_as_user', None)
-        if is_linux():
+        if is_termux():
+            print("Gateway service installation is not supported on Termux.")
+            print("Run manually: hermes gateway")
+            sys.exit(1)
+        if supports_systemd_services():
             systemd_install(force=force, system=system, run_as_user=run_as_user)
         elif is_macos():
             launchd_install(force)
@@ -2131,7 +2332,11 @@ def gateway_command(args):
             managed_error("uninstall gateway service (managed by NixOS)")
             return
         system = getattr(args, 'system', False)
-        if is_linux():
+        if is_termux():
+            print("Gateway service uninstall is not supported on Termux because there is no managed service to remove.")
+            print("Stop manual runs with: hermes gateway stop")
+            sys.exit(1)
+        if supports_systemd_services():
             systemd_uninstall(system=system)
         elif is_macos():
             launchd_uninstall()
@@ -2141,7 +2346,11 @@ def gateway_command(args):
     
     elif subcmd == "start":
         system = getattr(args, 'system', False)
-        if is_linux():
+        if is_termux():
+            print("Gateway service start is not supported on Termux because there is no system service manager.")
+            print("Run manually: hermes gateway")
+            sys.exit(1)
+        if supports_systemd_services():
             systemd_start(system=system)
         elif is_macos():
             launchd_start()
@@ -2156,7 +2365,7 @@ def gateway_command(args):
         if stop_all:
             # --all: kill every gateway process on the machine
             service_available = False
-            if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+            if supports_systemd_services() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
                 try:
                     systemd_stop(system=system)
                     service_available = True
@@ -2177,7 +2386,7 @@ def gateway_command(args):
         else:
             # Default: stop only the current profile's gateway
             service_available = False
-            if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+            if supports_systemd_services() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
                 try:
                     systemd_stop(system=system)
                     service_available = True
@@ -2205,7 +2414,7 @@ def gateway_command(args):
         system = getattr(args, 'system', False)
         service_configured = False
         
-        if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+        if supports_systemd_services() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
             service_configured = True
             try:
                 systemd_restart(system=system)
@@ -2222,7 +2431,7 @@ def gateway_command(args):
         
         if not service_available:
             # systemd/launchd restart failed — check if linger is the issue
-            if is_linux():
+            if supports_systemd_services():
                 linger_ok, _detail = get_systemd_linger_status()
                 if linger_ok is not True:
                     import getpass
@@ -2259,7 +2468,7 @@ def gateway_command(args):
         system = getattr(args, 'system', False)
         
         # Check for service first
-        if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+        if supports_systemd_services() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
             systemd_status(deep, system=system)
         elif is_macos() and get_launchd_plist_path().exists():
             launchd_status(deep)
@@ -2276,9 +2485,13 @@ def gateway_command(args):
                     for line in runtime_lines:
                         print(f"  {line}")
                 print()
-                print("To install as a service:")
-                print("  hermes gateway install")
-                print("  sudo hermes gateway install --system")
+                if is_termux():
+                    print("Termux note:")
+                    print("  Android may stop background jobs when Termux is suspended")
+                else:
+                    print("To install as a service:")
+                    print("  hermes gateway install")
+                    print("  sudo hermes gateway install --system")
             else:
                 print("✗ Gateway is not running")
                 runtime_lines = _runtime_health_lines()
@@ -2290,5 +2503,8 @@ def gateway_command(args):
                 print()
                 print("To start:")
                 print("  hermes gateway          # Run in foreground")
-                print("  hermes gateway install  # Install as user service")
-                print("  sudo hermes gateway install --system  # Install as boot-time system service")
+                if is_termux():
+                    print("  nohup hermes gateway > ~/.hermes/logs/gateway.log 2>&1 &  # Best-effort background start")
+                else:
+                    print("  hermes gateway install  # Install as user service")
+                    print("  sudo hermes gateway install --system  # Install as boot-time system service")

hermes_cli/main.py

@@ -97,10 +97,11 @@ def _apply_profile_override() -> None:
             consume = 1
             break
 
-    # 2. If no flag, check ~/.hermes/active_profile
+    # 2. If no flag, check active_profile in the hermes root
     if profile_name is None:
         try:
-            active_path = Path.home() / ".hermes" / "active_profile"
+            from hermes_constants import get_default_hermes_root
+            active_path = get_default_hermes_root() / "active_profile"
             if active_path.exists():
                 name = active_path.read_text().strip()
                 if name and name != "default":
@@ -646,6 +647,7 @@ def cmd_chat(args):
         "verbose": args.verbose,
         "quiet": getattr(args, "quiet", False),
         "query": args.query,
+        "image": getattr(args, "image", None),
         "resume": getattr(args, "resume", None),
         "worktree": getattr(args, "worktree", False),
         "checkpoints": getattr(args, "checkpoints", False),
@@ -857,7 +859,6 @@ def cmd_whatsapp(args):
 
 def cmd_setup(args):
     """Interactive setup wizard."""
-    _require_tty("setup")
     from hermes_cli.setup import run_setup_wizard
     run_setup_wizard(args)
 
@@ -967,10 +968,11 @@ def select_provider_and_model(args=None):
         ("alibaba", "Alibaba Cloud / DashScope Coding (Qwen + multi-provider)"),
     ]
 
-    # Add user-defined custom providers from config.yaml
-    custom_providers_cfg = config.get("custom_providers") or []
-    _custom_provider_map = {}  # key → {name, base_url, api_key}
-    if isinstance(custom_providers_cfg, list):
+    def _named_custom_provider_map(cfg) -> dict[str, dict[str, str]]:
+        custom_providers_cfg = cfg.get("custom_providers") or []
+        custom_provider_map = {}
+        if not isinstance(custom_providers_cfg, list):
+            return custom_provider_map
         for entry in custom_providers_cfg:
             if not isinstance(entry, dict):
                 continue
@@ -979,16 +981,23 @@ def select_provider_and_model(args=None):
             if not name or not base_url:
                 continue
             key = "custom:" + name.lower().replace(" ", "-")
-            short_url = base_url.replace("https://", "").replace("http://", "").rstrip("/")
-            saved_model = entry.get("model", "")
-            model_hint = f" — {saved_model}" if saved_model else ""
-            top_providers.append((key, f"{name} ({short_url}){model_hint}"))
-            _custom_provider_map[key] = {
+            custom_provider_map[key] = {
                 "name": name,
                 "base_url": base_url,
                 "api_key": entry.get("api_key", ""),
-                "model": saved_model,
+                "model": entry.get("model", ""),
             }
+        return custom_provider_map
+
+    # Add user-defined custom providers from config.yaml
+    _custom_provider_map = _named_custom_provider_map(config)  # key → {name, base_url, api_key}
+    for key, provider_info in _custom_provider_map.items():
+        name = provider_info["name"]
+        base_url = provider_info["base_url"]
+        short_url = base_url.replace("https://", "").replace("http://", "").rstrip("/")
+        saved_model = provider_info.get("model", "")
+        model_hint = f" — {saved_model}" if saved_model else ""
+        top_providers.append((key, f"{name} ({short_url}){model_hint}"))
 
     top_keys = {k for k, _ in top_providers}
     extended_keys = {k for k, _ in extended_providers}
@@ -1053,8 +1062,15 @@ def select_provider_and_model(args=None):
         _model_flow_copilot(config, current_model)
     elif selected_provider == "custom":
         _model_flow_custom(config)
-    elif selected_provider.startswith("custom:") and selected_provider in _custom_provider_map:
-        _model_flow_named_custom(config, _custom_provider_map[selected_provider])
+    elif selected_provider.startswith("custom:"):
+        provider_info = _named_custom_provider_map(load_config()).get(selected_provider)
+        if provider_info is None:
+            print(
+                "Warning: the selected saved custom provider is no longer available. "
+                "It may have been removed from config.yaml. No change."
+            )
+            return
+        _model_flow_named_custom(config, provider_info)
     elif selected_provider == "remove-custom":
         _remove_custom_provider(config)
     elif selected_provider == "anthropic":
@@ -1127,10 +1143,10 @@ def _model_flow_openrouter(config, current_model=""):
         print()
 
     from hermes_cli.models import model_ids, get_pricing_for_provider
-    openrouter_models = model_ids()
+    openrouter_models = model_ids(force_refresh=True)
 
     # Fetch live pricing (non-blocking — returns empty dict on failure)
-    pricing = get_pricing_for_provider("openrouter")
+    pricing = get_pricing_for_provider("openrouter", force_refresh=True)
 
     selected = _prompt_model_selection(openrouter_models, current_model=current_model, pricing=pricing)
     if selected:
@@ -1657,8 +1673,10 @@ def _remove_custom_provider(config):
             title="Select provider to remove:",
         )
         idx = menu.show()
+        from hermes_cli.curses_ui import flush_stdin
+        flush_stdin()
         print()
-    except (ImportError, NotImplementedError):
+    except (ImportError, NotImplementedError, OSError, subprocess.SubprocessError):
         for i, c in enumerate(choices, 1):
             print(f"  {i}. {c}")
         print()
@@ -1682,8 +1700,9 @@ def _remove_custom_provider(config):
 def _model_flow_named_custom(config, provider_info):
     """Handle a named custom provider from config.yaml custom_providers list.
 
-    If the entry has a saved model name, activates it immediately.
-    Otherwise probes the endpoint's /models API to let the user pick one.
+    Always probes the endpoint's /models API to let the user pick a model.
+    If a model was previously saved, it is pre-selected in the menu.
+    Falls back to the saved model if probing fails.
     """
     from hermes_cli.auth import _save_model_choice, deactivate_provider
     from hermes_cli.config import load_config, save_config
@@ -1694,54 +1713,46 @@ def _model_flow_named_custom(config, provider_info):
     api_key = provider_info.get("api_key", "")
     saved_model = provider_info.get("model", "")
 
-    # If a model is saved, just activate immediately — no probing needed
-    if saved_model:
-        _save_model_choice(saved_model)
-
-        cfg = load_config()
-        model = cfg.get("model")
-        if not isinstance(model, dict):
-            model = {"default": model} if model else {}
-            cfg["model"] = model
-        model["provider"] = "custom"
-        model["base_url"] = base_url
-        if api_key:
-            model["api_key"] = api_key
-        save_config(cfg)
-        deactivate_provider()
-
-        print(f"✅ Switched to: {saved_model}")
-        print(f"   Provider: {name} ({base_url})")
-        return
-
-    # No saved model — probe endpoint and let user pick
     print(f"  Provider: {name}")
     print(f"  URL:      {base_url}")
+    if saved_model:
+        print(f"  Current:  {saved_model}")
     print()
-    print("No model saved for this provider. Fetching available models...")
+
+    print("Fetching available models...")
     models = fetch_api_models(api_key, base_url, timeout=8.0)
 
     if models:
+        default_idx = 0
+        if saved_model and saved_model in models:
+            default_idx = models.index(saved_model)
+
         print(f"Found {len(models)} model(s):\n")
         try:
             from simple_term_menu import TerminalMenu
-            menu_items = [f"  {m}" for m in models] + ["  Cancel"]
+            menu_items = [
+                f"  {m} (current)" if m == saved_model else f"  {m}"
+                for m in models
+            ] + ["  Cancel"]
             menu = TerminalMenu(
-                menu_items, cursor_index=0,
+                menu_items, cursor_index=default_idx,
                 menu_cursor="-> ", menu_cursor_style=("fg_green", "bold"),
                 menu_highlight_style=("fg_green",),
                 cycle_cursor=True, clear_screen=False,
                 title=f"Select model from {name}:",
             )
             idx = menu.show()
+            from hermes_cli.curses_ui import flush_stdin
+            flush_stdin()
             print()
             if idx is None or idx >= len(models):
                 print("Cancelled.")
                 return
             model_name = models[idx]
-        except (ImportError, NotImplementedError):
+        except (ImportError, NotImplementedError, OSError, subprocess.SubprocessError):
             for i, m in enumerate(models, 1):
-                print(f"  {i}. {m}")
+                suffix = " (current)" if m == saved_model else ""
+                print(f"  {i}. {m}{suffix}")
             print(f"  {len(models) + 1}. Cancel")
             print()
             try:
@@ -1757,6 +1768,13 @@ def _model_flow_named_custom(config, provider_info):
             except (ValueError, KeyboardInterrupt, EOFError):
                 print("\nCancelled.")
                 return
+    elif saved_model:
+        print("Could not fetch models from endpoint.")
+        try:
+            model_name = input(f"Model name [{saved_model}]: ").strip() or saved_model
+        except (KeyboardInterrupt, EOFError):
+            print("\nCancelled.")
+            return
     else:
         print("Could not fetch models from endpoint. Enter model name manually.")
         try:
@@ -1852,6 +1870,8 @@ def _prompt_reasoning_effort_selection(efforts, current_effort=""):
             title="Select reasoning effort:",
         )
         idx = menu.show()
+        from hermes_cli.curses_ui import flush_stdin
+        flush_stdin()
         if idx is None:
             return None
         print()
@@ -1860,7 +1880,7 @@ def _prompt_reasoning_effort_selection(efforts, current_effort=""):
         if idx == len(ordered):
             return "none"
         return None
-    except (ImportError, NotImplementedError):
+    except (ImportError, NotImplementedError, OSError, subprocess.SubprocessError):
         pass
 
     print("Select reasoning effort:")
@@ -3021,33 +3041,19 @@ def _restore_stashed_changes(
         print("\nYour stashed changes are preserved — nothing is lost.")
         print(f"  Stash ref: {stash_ref}")
 
-        # Ask before resetting (if interactive)
-        do_reset = True
-        if prompt_user:
-            print("\nReset working tree to clean state so Hermes can run?")
-            print("  (You can re-apply your changes later with: git stash apply)")
-            print("[Y/n] ", end="", flush=True)
-            response = input().strip().lower()
-            if response not in ("", "y", "yes"):
-                do_reset = False
-
-        if do_reset:
-            subprocess.run(
-                git_cmd + ["reset", "--hard", "HEAD"],
-                cwd=cwd,
-                capture_output=True,
-            )
-            print("Working tree reset to clean state.")
-        else:
-            print("Working tree left as-is (may have conflict markers).")
-            print("Resolve conflicts manually, then run: git stash drop")
-
-        print(f"Restore your changes with: git stash apply {stash_ref}")
-        # In non-interactive mode (gateway /update), don't abort — the code
-        # update itself succeeded, only the stash restore had conflicts.
-        # Aborting would report the entire update as failed.
-        if prompt_user:
-            sys.exit(1)
+        # Always reset to clean state — leaving conflict markers in source
+        # files makes hermes completely unrunnable (SyntaxError on import).
+        # The user's changes are safe in the stash for manual recovery.
+        subprocess.run(
+            git_cmd + ["reset", "--hard", "HEAD"],
+            cwd=cwd,
+            capture_output=True,
+        )
+        print("Working tree reset to clean state.")
+        print(f"Restore your changes later with: git stash apply {stash_ref}")
+        # Don't sys.exit — the code update itself succeeded, only the stash
+        # restore had conflicts.  Let cmd_update continue with pip install,
+        # skill sync, and gateway restart.
         return False
 
     stash_selector = _resolve_stash_selector(git_cmd, cwd, stash_ref)
@@ -3308,10 +3314,11 @@ def _invalidate_update_cache():
     ``hermes update``, every profile is now current.
     """
     homes = []
-    # Default profile home
-    default_home = Path.home() / ".hermes"
+    # Default profile home (Docker-aware — uses /opt/data in Docker)
+    from hermes_constants import get_default_hermes_root
+    default_home = get_default_hermes_root()
     homes.append(default_home)
-    # Named profiles under ~/.hermes/profiles/
+    # Named profiles under <root>/profiles/
     profiles_root = default_home / "profiles"
     if profiles_root.is_dir():
         for entry in profiles_root.iterdir():
@@ -3763,7 +3770,7 @@ def cmd_update(args):
         # running gateway needs restarting to pick up the new code.
         try:
             from hermes_cli.gateway import (
-                is_macos, is_linux, _ensure_user_systemd_env,
+                is_macos, supports_systemd_services, _ensure_user_systemd_env,
                 find_gateway_pids,
                 _get_service_pids,
             )
@@ -3774,7 +3781,7 @@ def cmd_update(args):
 
             # --- Systemd services (Linux) ---
             # Discover all hermes-gateway* units (default + profiles)
-            if is_linux():
+            if supports_systemd_services():
                 try:
                     _ensure_user_systemd_env()
                 except Exception:
@@ -4048,7 +4055,10 @@ def cmd_profile(args):
             print(f"  {name} chat               Start chatting")
             print(f"  {name} gateway start      Start the messaging gateway")
             if clone or clone_all:
-                profile_dir_display = f"~/.hermes/profiles/{name}"
+                try:
+                    profile_dir_display = "~/" + str(profile_dir.relative_to(Path.home()))
+                except ValueError:
+                    profile_dir_display = str(profile_dir)
                 print(f"\n  Edit {profile_dir_display}/.env for different API keys")
                 print(f"  Edit {profile_dir_display}/SOUL.md for different personality")
             print()
@@ -4291,6 +4301,10 @@ For more help on a command:
         "-q", "--query",
         help="Single query (non-interactive mode)"
     )
+    chat_parser.add_argument(
+        "--image",
+        help="Optional local image path to attach to a single query"
+    )
     chat_parser.add_argument(
         "-m", "--model",
         help="Model to use (e.g., anthropic/claude-sonnet-4)"
@@ -4481,12 +4495,12 @@ For more help on a command:
         "setup",
         help="Interactive setup wizard",
         description="Configure Hermes Agent with an interactive wizard. "
-                    "Run a specific section: hermes setup model|terminal|gateway|tools|agent"
+                    "Run a specific section: hermes setup model|tts|terminal|gateway|tools|agent"
     )
     setup_parser.add_argument(
         "section",
         nargs="?",
-        choices=["model", "terminal", "gateway", "tools", "agent"],
+        choices=["model", "tts", "terminal", "gateway", "tools", "agent"],
         default=None,
         help="Run a specific setup section instead of the full wizard"
     )

hermes_cli/model_normalize.py

@@ -76,17 +76,22 @@ _STRIP_VENDOR_ONLY_PROVIDERS: frozenset[str] = frozenset({
     "copilot-acp",
 })
 
-# Providers whose own naming is authoritative -- pass through unchanged.
-_PASSTHROUGH_PROVIDERS: frozenset[str] = frozenset({
+# Providers whose native naming is authoritative -- pass through unchanged.
+_AUTHORITATIVE_NATIVE_PROVIDERS: frozenset[str] = frozenset({
     "gemini",
+    "huggingface",
+    "openai-codex",
+})
+
+# Direct providers that accept bare native names but should repair a matching
+# provider/ prefix when users copy the aggregator form into config.yaml.
+_MATCHING_PREFIX_STRIP_PROVIDERS: frozenset[str] = frozenset({
     "zai",
     "kimi-coding",
     "minimax",
     "minimax-cn",
     "alibaba",
     "qwen-oauth",
-    "huggingface",
-    "openai-codex",
     "custom",
 })
 
@@ -168,6 +173,40 @@ def _dots_to_hyphens(model_name: str) -> str:
     return model_name.replace(".", "-")
 
 
+def _normalize_provider_alias(provider_name: str) -> str:
+    """Resolve provider aliases to Hermes' canonical ids."""
+    raw = (provider_name or "").strip().lower()
+    if not raw:
+        return raw
+    try:
+        from hermes_cli.models import normalize_provider
+
+        return normalize_provider(raw)
+    except Exception:
+        return raw
+
+
+def _strip_matching_provider_prefix(model_name: str, target_provider: str) -> str:
+    """Strip ``provider/`` only when the prefix matches the target provider.
+
+    This prevents arbitrary slash-bearing model IDs from being mangled on
+    native providers while still repairing manual config values like
+    ``zai/glm-5.1`` for the ``zai`` provider.
+    """
+    if "/" not in model_name:
+        return model_name
+
+    prefix, remainder = model_name.split("/", 1)
+    if not prefix.strip() or not remainder.strip():
+        return model_name
+
+    normalized_prefix = _normalize_provider_alias(prefix)
+    normalized_target = _normalize_provider_alias(target_provider)
+    if normalized_prefix and normalized_prefix == normalized_target:
+        return remainder.strip()
+    return model_name
+
+
 def detect_vendor(model_name: str) -> Optional[str]:
     """Detect the vendor slug from a bare model name.
 
@@ -305,24 +344,37 @@ def normalize_model_for_provider(model_input: str, target_provider: str) -> str:
     if not name:
         return name
 
-    provider = (target_provider or "").strip().lower()
+    provider = _normalize_provider_alias(target_provider)
 
     # --- Aggregators: need vendor/model format ---
     if provider in _AGGREGATOR_PROVIDERS:
         return _prepend_vendor(name)
 
-    # --- Anthropic / OpenCode: strip vendor, dots -> hyphens ---
+    # --- Anthropic / OpenCode: strip matching provider prefix, dots -> hyphens ---
     if provider in _DOT_TO_HYPHEN_PROVIDERS:
-        bare = _strip_vendor_prefix(name)
+        bare = _strip_matching_provider_prefix(name, provider)
+        if "/" in bare:
+            return bare
         return _dots_to_hyphens(bare)
 
-    # --- Copilot: strip vendor, keep dots ---
+    # --- Copilot: strip matching provider prefix, keep dots ---
     if provider in _STRIP_VENDOR_ONLY_PROVIDERS:
-        return _strip_vendor_prefix(name)
+        return _strip_matching_provider_prefix(name, provider)
 
     # --- DeepSeek: map to one of two canonical names ---
     if provider == "deepseek":
-        return _normalize_for_deepseek(name)
+        bare = _strip_matching_provider_prefix(name, provider)
+        if "/" in bare:
+            return bare
+        return _normalize_for_deepseek(bare)
+
+    # --- Direct providers: repair matching provider prefixes only ---
+    if provider in _MATCHING_PREFIX_STRIP_PROVIDERS:
+        return _strip_matching_provider_prefix(name, provider)
+
+    # --- Authoritative native providers: preserve user-facing slugs as-is ---
+    if provider in _AUTHORITATIVE_NATIVE_PROVIDERS:
+        return name
 
     # --- Custom & all others: pass through as-is ---
     return name

hermes_cli/model_switch.py

@@ -25,6 +25,7 @@ from dataclasses import dataclass
 from typing import List, NamedTuple, Optional
 
 from hermes_cli.providers import (
+    custom_provider_slug,
     determine_api_mode,
     get_label,
     is_aggregator,
@@ -336,6 +337,7 @@ def resolve_alias(
 def get_authenticated_provider_slugs(
     current_provider: str = "",
     user_providers: dict = None,
+    custom_providers: list | None = None,
 ) -> list[str]:
     """Return slugs of providers that have credentials.
 
@@ -346,6 +348,7 @@ def get_authenticated_provider_slugs(
         providers = list_authenticated_providers(
             current_provider=current_provider,
             user_providers=user_providers,
+            custom_providers=custom_providers,
             max_models=0,
         )
         return [p["slug"] for p in providers]
@@ -383,6 +386,7 @@ def switch_model(
     is_global: bool = False,
     explicit_provider: str = "",
     user_providers: dict = None,
+    custom_providers: list | None = None,
 ) -> ModelSwitchResult:
     """Core model-switching pipeline shared between CLI and gateway.
 
@@ -416,6 +420,7 @@ def switch_model(
         is_global: Whether to persist the switch.
         explicit_provider: From --provider flag (empty = no explicit provider).
         user_providers: The ``providers:`` dict from config.yaml (for user endpoints).
+        custom_providers: The ``custom_providers:`` list from config.yaml.
 
     Returns:
         ModelSwitchResult with all information the caller needs.
@@ -436,7 +441,11 @@ def switch_model(
     # =================================================================
     if explicit_provider:
         # Resolve the provider
-        pdef = resolve_provider_full(explicit_provider, user_providers)
+        pdef = resolve_provider_full(
+            explicit_provider,
+            user_providers,
+            custom_providers,
+        )
         if pdef is None:
             _switch_err = (
                 f"Unknown provider '{explicit_provider}'. "
@@ -516,6 +525,7 @@ def switch_model(
                 authed = get_authenticated_provider_slugs(
                     current_provider=current_provider,
                     user_providers=user_providers,
+                    custom_providers=custom_providers,
                 )
                 fallback_result = _resolve_alias_fallback(raw_input, authed)
                 if fallback_result is not None:
@@ -590,6 +600,14 @@ def switch_model(
 
     provider_changed = target_provider != current_provider
     provider_label = get_label(target_provider)
+    if target_provider.startswith("custom:"):
+        custom_pdef = resolve_provider_full(
+            target_provider,
+            user_providers,
+            custom_providers,
+        )
+        if custom_pdef is not None:
+            provider_label = custom_pdef.name
 
     # --- Resolve credentials ---
     api_key = current_api_key
@@ -708,6 +726,7 @@ def switch_model(
 def list_authenticated_providers(
     current_provider: str = "",
     user_providers: dict = None,
+    custom_providers: list | None = None,
     max_models: int = 8,
 ) -> List[dict]:
     """Detect which providers have credentials and list their curated models.
@@ -790,42 +809,69 @@ def list_authenticated_providers(
         })
         seen_slugs.add(slug)
 
-    # --- 2. Check Hermes-only providers (nous, openai-codex, copilot) ---
+    # --- 2. Check Hermes-only providers (nous, openai-codex, copilot, opencode-go) ---
     from hermes_cli.providers import HERMES_OVERLAYS
+    from hermes_cli.auth import PROVIDER_REGISTRY as _auth_registry
+
+    # Build reverse mapping: models.dev ID → Hermes provider ID.
+    # HERMES_OVERLAYS keys may be models.dev IDs (e.g. "github-copilot")
+    # while _PROVIDER_MODELS and config.yaml use Hermes IDs ("copilot").
+    _mdev_to_hermes = {v: k for k, v in PROVIDER_TO_MODELS_DEV.items()}
+
     for pid, overlay in HERMES_OVERLAYS.items():
         if pid in seen_slugs:
             continue
+
+        # Resolve Hermes slug — e.g. "github-copilot" → "copilot"
+        hermes_slug = _mdev_to_hermes.get(pid, pid)
+        if hermes_slug in seen_slugs:
+            continue
+
         # Check if credentials exist
         has_creds = False
         if overlay.extra_env_vars:
             has_creds = any(os.environ.get(ev) for ev in overlay.extra_env_vars)
-        if overlay.auth_type in ("oauth_device_code", "oauth_external", "external_process"):
+        # Also check api_key_env_vars from PROVIDER_REGISTRY for api_key auth_type
+        if not has_creds and overlay.auth_type == "api_key":
+            for _key in (pid, hermes_slug):
+                pcfg = _auth_registry.get(_key)
+                if pcfg and pcfg.api_key_env_vars:
+                    if any(os.environ.get(ev) for ev in pcfg.api_key_env_vars):
+                        has_creds = True
+                        break
+        if not has_creds and overlay.auth_type in ("oauth_device_code", "oauth_external", "external_process"):
             # These use auth stores, not env vars — check for auth.json entries
             try:
                 from hermes_cli.auth import _load_auth_store
                 store = _load_auth_store()
-                if store and (pid in store.get("providers", {}) or pid in store.get("credential_pool", {})):
+                providers_store = store.get("providers", {})
+                pool_store = store.get("credential_pool", {})
+                if store and (
+                    pid in providers_store or hermes_slug in providers_store
+                    or pid in pool_store or hermes_slug in pool_store
+                ):
                     has_creds = True
             except Exception as exc:
                 logger.debug("Auth store check failed for %s: %s", pid, exc)
         if not has_creds:
             continue
 
-        # Use curated list
-        model_ids = curated.get(pid, [])
+        # Use curated list — look up by Hermes slug, fall back to overlay key
+        model_ids = curated.get(hermes_slug, []) or curated.get(pid, [])
         total = len(model_ids)
         top = model_ids[:max_models]
 
         results.append({
-            "slug": pid,
-            "name": get_label(pid),
-            "is_current": pid == current_provider,
+            "slug": hermes_slug,
+            "name": get_label(hermes_slug),
+            "is_current": hermes_slug == current_provider or pid == current_provider,
             "is_user_defined": False,
             "models": top,
             "total_models": total,
             "source": "hermes",
         })
         seen_slugs.add(pid)
+        seen_slugs.add(hermes_slug)
 
     # --- 3. User-defined endpoints from config ---
     if user_providers and isinstance(user_providers, dict):
@@ -853,6 +899,43 @@ def list_authenticated_providers(
                 "api_url": api_url,
             })
 
+    # --- 4. Saved custom providers from config ---
+    if custom_providers and isinstance(custom_providers, list):
+        for entry in custom_providers:
+            if not isinstance(entry, dict):
+                continue
+
+            display_name = (entry.get("name") or "").strip()
+            api_url = (
+                entry.get("base_url", "")
+                or entry.get("url", "")
+                or entry.get("api", "")
+                or ""
+            ).strip()
+            if not display_name or not api_url:
+                continue
+
+            slug = custom_provider_slug(display_name)
+            if slug in seen_slugs:
+                continue
+
+            models_list = []
+            default_model = (entry.get("model") or "").strip()
+            if default_model:
+                models_list.append(default_model)
+
+            results.append({
+                "slug": slug,
+                "name": display_name,
+                "is_current": slug == current_provider,
+                "is_user_defined": True,
+                "models": models_list,
+                "total_models": len(models_list),
+                "source": "user-config",
+                "api_url": api_url,
+            })
+            seen_slugs.add(slug)
+
     # Sort: current provider first, then by model count descending
     results.sort(key=lambda r: (not r["is_current"], -r["total_models"]))
 

hermes_cli/models.py

@@ -20,18 +20,20 @@ COPILOT_EDITOR_VERSION = "vscode/1.104.1"
 COPILOT_REASONING_EFFORTS_GPT5 = ["minimal", "low", "medium", "high"]
 COPILOT_REASONING_EFFORTS_O_SERIES = ["low", "medium", "high"]
 
+
+# Fallback OpenRouter snapshot used when the live catalog is unavailable.
 # (model_id, display description shown in menus)
 OPENROUTER_MODELS: list[tuple[str, str]] = [
     ("anthropic/claude-opus-4.6",       "recommended"),
     ("anthropic/claude-sonnet-4.6",     ""),
-    ("qwen/qwen3.6-plus:free", "free"),
+    ("qwen/qwen3.6-plus",               ""),
     ("anthropic/claude-sonnet-4.5",     ""),
     ("anthropic/claude-haiku-4.5",      ""),
     ("openai/gpt-5.4",                  ""),
     ("openai/gpt-5.4-mini",             ""),
     ("xiaomi/mimo-v2-pro",               ""),
     ("openai/gpt-5.3-codex",            ""),
-    ("google/gemini-3-pro-preview",     ""),
+    ("google/gemini-3-pro-image-preview", ""),
     ("google/gemini-3-flash-preview",   ""),
     ("google/gemini-3.1-pro-preview",     ""),
     ("google/gemini-3.1-flash-lite-preview",   ""),
@@ -43,7 +45,7 @@ OPENROUTER_MODELS: list[tuple[str, str]] = [
     ("z-ai/glm-5.1",                    ""),
     ("z-ai/glm-5-turbo",                ""),
     ("moonshotai/kimi-k2.5",            ""),
-    ("x-ai/grok-4.20-beta",             ""),
+    ("x-ai/grok-4.20",                  ""),
     ("nvidia/nemotron-3-super-120b-a12b",      ""),
     ("nvidia/nemotron-3-super-120b-a12b:free", "free"),
     ("arcee-ai/trinity-large-preview:free", "free"),
@@ -52,6 +54,8 @@ OPENROUTER_MODELS: list[tuple[str, str]] = [
     ("openai/gpt-5.4-nano",             ""),
 ]
 
+_openrouter_catalog_cache: list[tuple[str, str]] | None = None
+
 _PROVIDER_MODELS: dict[str, list[str]] = {
     "nous": [
         "anthropic/claude-opus-4.6",
@@ -125,6 +129,19 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "glm-4.5",
         "glm-4.5-flash",
     ],
+    "xai": [
+        "grok-4.20-0309-reasoning",
+        "grok-4.20-0309-non-reasoning",
+        "grok-4.20-multi-agent-0309",
+        "grok-4-1-fast-reasoning",
+        "grok-4-1-fast-non-reasoning",
+        "grok-4-fast-reasoning",
+        "grok-4-fast-non-reasoning",
+        "grok-4-0709",
+        "grok-code-fast-1",
+        "grok-3",
+        "grok-3-mini",
+    ],
     "kimi-coding": [
         "kimi-for-coding",
         "kimi-k2.5",
@@ -520,9 +537,82 @@ _PROVIDER_ALIASES = {
 }
 
 
-def model_ids() -> list[str]:
+def _openrouter_model_is_free(pricing: Any) -> bool:
+    """Return True when both prompt and completion pricing are zero."""
+    if not isinstance(pricing, dict):
+        return False
+    try:
+        return float(pricing.get("prompt", "0")) == 0 and float(pricing.get("completion", "0")) == 0
+    except (TypeError, ValueError):
+        return False
+
+
+def fetch_openrouter_models(
+    timeout: float = 8.0,
+    *,
+    force_refresh: bool = False,
+) -> list[tuple[str, str]]:
+    """Return the curated OpenRouter picker list, refreshed from the live catalog when possible."""
+    global _openrouter_catalog_cache
+
+    if _openrouter_catalog_cache is not None and not force_refresh:
+        return list(_openrouter_catalog_cache)
+
+    fallback = list(OPENROUTER_MODELS)
+    preferred_ids = [mid for mid, _ in fallback]
+
+    try:
+        req = urllib.request.Request(
+            "https://openrouter.ai/api/v1/models",
+            headers={"Accept": "application/json"},
+        )
+        with urllib.request.urlopen(req, timeout=timeout) as resp:
+            payload = json.loads(resp.read().decode())
+    except Exception:
+        return list(_openrouter_catalog_cache or fallback)
+
+    live_items = payload.get("data", [])
+    if not isinstance(live_items, list):
+        return list(_openrouter_catalog_cache or fallback)
+
+    live_by_id: dict[str, dict[str, Any]] = {}
+    for item in live_items:
+        if not isinstance(item, dict):
+            continue
+        mid = str(item.get("id") or "").strip()
+        if not mid:
+            continue
+        live_by_id[mid] = item
+
+    curated: list[tuple[str, str]] = []
+    for preferred_id in preferred_ids:
+        live_item = live_by_id.get(preferred_id)
+        if live_item is None:
+            continue
+        desc = "free" if _openrouter_model_is_free(live_item.get("pricing")) else ""
+        curated.append((preferred_id, desc))
+
+    if not curated:
+        return list(_openrouter_catalog_cache or fallback)
+
+    first_id, _ = curated[0]
+    curated[0] = (first_id, "recommended")
+    _openrouter_catalog_cache = curated
+    return list(curated)
+
+
+def model_ids(*, force_refresh: bool = False) -> list[str]:
     """Return just the OpenRouter model-id strings."""
-    return [mid for mid, _ in OPENROUTER_MODELS]
+    return [mid for mid, _ in fetch_openrouter_models(force_refresh=force_refresh)]
+
+
+def menu_labels(*, force_refresh: bool = False) -> list[str]:
+    """Return display labels like 'anthropic/claude-opus-4.6 (recommended)'."""
+    labels = []
+    for mid, desc in fetch_openrouter_models(force_refresh=force_refresh):
+        labels.append(f"{mid} ({desc})" if desc else mid)
+    return labels
+
 
 
 # ---------------------------------------------------------------------------
@@ -684,13 +774,14 @@ def _resolve_nous_pricing_credentials() -> tuple[str, str]:
     return ("", "")
 
 
-def get_pricing_for_provider(provider: str) -> dict[str, dict[str, str]]:
+def get_pricing_for_provider(provider: str, *, force_refresh: bool = False) -> dict[str, dict[str, str]]:
     """Return live pricing for providers that support it (openrouter, nous)."""
     normalized = normalize_provider(provider)
     if normalized == "openrouter":
         return fetch_models_with_pricing(
             api_key=_resolve_openrouter_api_key(),
             base_url="https://openrouter.ai/api",
+            force_refresh=force_refresh,
         )
     if normalized == "nous":
         api_key, base_url = _resolve_nous_pricing_credentials()
@@ -703,6 +794,7 @@ def get_pricing_for_provider(provider: str) -> dict[str, dict[str, str]]:
             return fetch_models_with_pricing(
                 api_key=api_key,
                 base_url=stripped,
+                force_refresh=force_refresh,
             )
     return {}
 
@@ -811,7 +903,11 @@ def _get_custom_base_url() -> str:
     return ""
 
 
-def curated_models_for_provider(provider: Optional[str]) -> list[tuple[str, str]]:
+def curated_models_for_provider(
+    provider: Optional[str],
+    *,
+    force_refresh: bool = False,
+) -> list[tuple[str, str]]:
     """Return ``(model_id, description)`` tuples for a provider's model list.
 
     Tries to fetch the live model list from the provider's API first,
@@ -820,7 +916,7 @@ def curated_models_for_provider(provider: Optional[str]) -> list[tuple[str, str]
     """
     normalized = normalize_provider(provider)
     if normalized == "openrouter":
-        return list(OPENROUTER_MODELS)
+        return fetch_openrouter_models(force_refresh=force_refresh)
 
     # Try live API first (Codex, Nous, etc. all support /models)
     live = provider_model_ids(normalized)
@@ -939,12 +1035,12 @@ def _find_openrouter_slug(model_name: str) -> Optional[str]:
         return None
 
     # Exact match (already has provider/ prefix)
-    for mid, _ in OPENROUTER_MODELS:
+    for mid in model_ids():
         if name_lower == mid.lower():
             return mid
 
     # Try matching just the model part (after the /)
-    for mid, _ in OPENROUTER_MODELS:
+    for mid in model_ids():
         if "/" in mid:
             _, model_part = mid.split("/", 1)
             if name_lower == model_part.lower():
@@ -974,6 +1070,79 @@ def provider_label(provider: Optional[str]) -> str:
     return _PROVIDER_LABELS.get(normalized, original or "OpenRouter")
 
 
+# Models that support OpenAI Priority Processing (service_tier="priority").
+# See https://openai.com/api-priority-processing/ for the canonical list.
+# Only the bare model slug is stored (no vendor prefix).
+_PRIORITY_PROCESSING_MODELS: frozenset[str] = frozenset({
+    "gpt-5.4",
+    "gpt-5.4-mini",
+    "gpt-5.2",
+    "gpt-5.1",
+    "gpt-5",
+    "gpt-5-mini",
+    "gpt-4.1",
+    "gpt-4.1-mini",
+    "gpt-4.1-nano",
+    "gpt-4o",
+    "gpt-4o-mini",
+    "o3",
+    "o4-mini",
+})
+
+# Models that support Anthropic Fast Mode (speed="fast").
+# See https://platform.claude.com/docs/en/build-with-claude/fast-mode
+# Currently only Claude Opus 4.6.  Both hyphen and dot variants are stored
+# to handle native Anthropic (claude-opus-4-6) and OpenRouter (claude-opus-4.6).
+_ANTHROPIC_FAST_MODE_MODELS: frozenset[str] = frozenset({
+    "claude-opus-4-6",
+    "claude-opus-4.6",
+})
+
+
+def _strip_vendor_prefix(model_id: str) -> str:
+    """Strip vendor/ prefix from a model ID (e.g. 'anthropic/claude-opus-4-6' -> 'claude-opus-4-6')."""
+    raw = str(model_id or "").strip().lower()
+    if "/" in raw:
+        raw = raw.split("/", 1)[1]
+    return raw
+
+
+def model_supports_fast_mode(model_id: Optional[str]) -> bool:
+    """Return whether Hermes should expose the /fast toggle for this model."""
+    raw = _strip_vendor_prefix(str(model_id or ""))
+    if raw in _PRIORITY_PROCESSING_MODELS:
+        return True
+    # Anthropic fast mode — strip date suffixes (e.g. claude-opus-4-6-20260401)
+    # and OpenRouter variant tags (:fast, :beta) for matching.
+    base = raw.split(":")[0]
+    return base in _ANTHROPIC_FAST_MODE_MODELS
+
+
+def _is_anthropic_fast_model(model_id: Optional[str]) -> bool:
+    """Return True if the model supports Anthropic's fast mode (speed='fast')."""
+    raw = _strip_vendor_prefix(str(model_id or ""))
+    base = raw.split(":")[0]
+    return base in _ANTHROPIC_FAST_MODE_MODELS
+
+
+def resolve_fast_mode_overrides(model_id: Optional[str]) -> dict[str, Any] | None:
+    """Return request_overrides for fast/priority mode, or None if unsupported.
+
+    Returns provider-appropriate overrides:
+    - OpenAI models: ``{"service_tier": "priority"}`` (Priority Processing)
+    - Anthropic models: ``{"speed": "fast"}`` (Anthropic Fast Mode beta)
+
+    The overrides are injected into the API request kwargs by
+    ``_build_api_kwargs`` in run_agent.py — each API path handles its own
+    keys (service_tier for OpenAI/Codex, speed for Anthropic Messages).
+    """
+    if not model_supports_fast_mode(model_id):
+        return None
+    if _is_anthropic_fast_model(model_id):
+        return {"speed": "fast"}
+    return {"service_tier": "priority"}
+
+
 def _resolve_copilot_catalog_api_key() -> str:
     """Best-effort GitHub token for fetching the Copilot model catalog."""
     try:
@@ -985,7 +1154,7 @@ def _resolve_copilot_catalog_api_key() -> str:
         return ""
 
 
-def provider_model_ids(provider: Optional[str]) -> list[str]:
+def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False) -> list[str]:
     """Return the best known model catalog for a provider.
 
     Tries live API endpoints for providers that support them (Codex, Nous),
@@ -993,7 +1162,7 @@ def provider_model_ids(provider: Optional[str]) -> list[str]:
     """
     normalized = normalize_provider(provider)
     if normalized == "openrouter":
-        return model_ids()
+        return model_ids(force_refresh=force_refresh)
     if normalized == "openai-codex":
         from hermes_cli.codex_models import get_codex_model_ids
 

hermes_cli/profiles.py

@@ -42,6 +42,11 @@ _PROFILE_DIRS = [
     "plans",
     "workspace",
     "cron",
+    # Per-profile HOME for subprocesses: isolates system tool configs (git,
+    # ssh, gh, npm …) so credentials don't bleed between profiles.  In Docker
+    # this also ensures tool configs land inside the persistent volume.
+    # See hermes_constants.get_subprocess_home() and issue #4426.
+    "home",
 ]
 
 # Files copied during --clone (if they exist in the source)
@@ -115,16 +120,26 @@ _HERMES_SUBCOMMANDS = frozenset({
 def _get_profiles_root() -> Path:
     """Return the directory where named profiles are stored.
 
-    Always ``~/.hermes/profiles/`` — anchored to the user's home,
-    NOT to the current HERMES_HOME (which may itself be a profile).
-    This ensures ``coder profile list`` can see all profiles.
+    Anchored to the hermes root, NOT to the current HERMES_HOME
+    (which may itself be a profile).  This ensures ``coder profile list``
+    can see all profiles.
+
+    In Docker/custom deployments where HERMES_HOME points outside
+    ``~/.hermes``, profiles live under ``HERMES_HOME/profiles/`` so
+    they persist on the mounted volume.
     """
-    return Path.home() / ".hermes" / "profiles"
+    return _get_default_hermes_home() / "profiles"
 
 
 def _get_default_hermes_home() -> Path:
-    """Return the default (pre-profile) HERMES_HOME path."""
-    return Path.home() / ".hermes"
+    """Return the default (pre-profile) HERMES_HOME path.
+
+    In standard deployments this is ``~/.hermes``.
+    In Docker/custom deployments where HERMES_HOME is outside ``~/.hermes``
+    (e.g. ``/opt/data``), returns HERMES_HOME directly.
+    """
+    from hermes_constants import get_default_hermes_root
+    return get_default_hermes_root()
 
 
 def _get_active_profile_path() -> Path:

hermes_cli/providers.py

@@ -127,6 +127,11 @@ HERMES_OVERLAYS: Dict[str, HermesOverlay] = {
         is_aggregator=True,
         base_url_env_var="HF_BASE_URL",
     ),
+    "xai": HermesOverlay(
+        transport="openai_chat",
+        base_url_override="https://api.x.ai/v1",
+        base_url_env_var="XAI_BASE_URL",
+    ),
 }
 
 
@@ -163,6 +168,10 @@ ALIASES: Dict[str, str] = {
     "z.ai": "zai",
     "zhipu": "zai",
 
+    # xai
+    "x-ai": "xai",
+    "x.ai": "xai",
+
     # kimi-for-coding (models.dev ID)
     "kimi": "kimi-for-coding",
     "kimi-coding": "kimi-for-coding",
@@ -341,6 +350,7 @@ def get_label(provider_id: str) -> str:
 
 
 
+
 def is_aggregator(provider: str) -> bool:
     """Return True when the provider is a multi-model aggregator."""
     pdef = get_provider(provider)
@@ -411,9 +421,64 @@ def resolve_user_provider(name: str, user_config: Dict[str, Any]) -> Optional[Pr
     )
 
 
+def custom_provider_slug(display_name: str) -> str:
+    """Build a canonical slug for a custom_providers entry.
+
+    Matches the convention used by runtime_provider and credential_pool
+    (``custom:<normalized-name>``).  Centralised here so all call-sites
+    produce identical slugs.
+    """
+    return "custom:" + display_name.strip().lower().replace(" ", "-")
+
+
+def resolve_custom_provider(
+    name: str,
+    custom_providers: Optional[List[Dict[str, Any]]],
+) -> Optional[ProviderDef]:
+    """Resolve a provider from the user's config.yaml ``custom_providers`` list."""
+    if not custom_providers or not isinstance(custom_providers, list):
+        return None
+
+    requested = (name or "").strip().lower()
+    if not requested:
+        return None
+
+    for entry in custom_providers:
+        if not isinstance(entry, dict):
+            continue
+
+        display_name = (entry.get("name") or "").strip()
+        api_url = (
+            entry.get("base_url", "")
+            or entry.get("url", "")
+            or entry.get("api", "")
+            or ""
+        ).strip()
+        if not display_name or not api_url:
+            continue
+
+        slug = custom_provider_slug(display_name)
+        if requested not in {display_name.lower(), slug}:
+            continue
+
+        return ProviderDef(
+            id=slug,
+            name=display_name,
+            transport="openai_chat",
+            api_key_env_vars=(),
+            base_url=api_url,
+            is_aggregator=False,
+            auth_type="api_key",
+            source="user-config",
+        )
+
+    return None
+
+
 def resolve_provider_full(
     name: str,
     user_providers: Optional[Dict[str, Any]] = None,
+    custom_providers: Optional[List[Dict[str, Any]]] = None,
 ) -> Optional[ProviderDef]:
     """Full resolution chain: built-in → models.dev → user config.
 
@@ -422,6 +487,7 @@ def resolve_provider_full(
     Args:
         name: Provider name or alias.
         user_providers: The ``providers:`` dict from config.yaml (optional).
+        custom_providers: The ``custom_providers:`` list from config.yaml (optional).
 
     Returns:
         ProviderDef if found, else None.
@@ -444,6 +510,11 @@ def resolve_provider_full(
         if user_pdef is not None:
             return user_pdef
 
+    # 2b. Saved custom providers from config
+    custom_pdef = resolve_custom_provider(name, custom_providers)
+    if custom_pdef is not None:
+        return custom_pdef
+
     # 3. Try models.dev directly (for providers not in our ALIASES)
     try:
         from agent.models_dev import get_provider_info as _mdev_provider

hermes_cli/runtime_provider.py

@@ -16,6 +16,7 @@ from hermes_cli.auth import (
     DEFAULT_CODEX_BASE_URL,
     DEFAULT_QWEN_BASE_URL,
     PROVIDER_REGISTRY,
+    _agent_key_is_usable,
     format_auth_error,
     resolve_provider,
     resolve_nous_runtime_credentials,
@@ -644,6 +645,21 @@ def resolve_runtime_provider(
                 getattr(entry, "runtime_api_key", None)
                 or getattr(entry, "access_token", "")
             )
+        # For Nous, the pool entry's runtime_api_key is the agent_key — a
+        # short-lived inference credential (~30 min TTL).  The pool doesn't
+        # refresh it during selection (that would trigger network calls in
+        # non-runtime contexts like `hermes auth list`).  If the key is
+        # expired, clear pool_api_key so we fall through to
+        # resolve_nous_runtime_credentials() which handles refresh + mint.
+        if provider == "nous" and entry is not None and pool_api_key:
+            min_ttl = max(60, int(os.getenv("HERMES_NOUS_MIN_KEY_TTL_SECONDS", "1800")))
+            nous_state = {
+                "agent_key": getattr(entry, "agent_key", None),
+                "agent_key_expires_at": getattr(entry, "agent_key_expires_at", None),
+            }
+            if not _agent_key_is_usable(nous_state, min_ttl):
+                logger.debug("Nous pool entry agent_key expired/missing, falling through to runtime resolution")
+                pool_api_key = ""
         if entry is not None and pool_api_key:
             return _resolve_runtime_from_pool_entry(
                 provider=provider,

hermes_cli/setup.py

@@ -16,6 +16,7 @@ import logging
 import os
 import shutil
 import sys
+import copy
 from pathlib import Path
 from typing import Optional, Dict, Any
 
@@ -175,6 +176,7 @@ def _setup_copilot_reasoning_selection(
 
 # Import config helpers
 from hermes_cli.config import (
+    DEFAULT_CONFIG,
     get_hermes_home,
     get_config_path,
     get_env_path,
@@ -336,6 +338,8 @@ def _curses_prompt_choice(question: str, choices: list, default: int = 0) -> int
                     return
 
         curses.wrapper(_curses_menu)
+        from hermes_cli.curses_ui import flush_stdin
+        flush_stdin()
         return result_holder[0]
     except Exception:
         return -1
@@ -780,8 +784,10 @@ def setup_model_provider(config: dict, *, quick: bool = False):
     # changes with stale values (#4172).
     _refreshed = load_config()
     config["model"] = _refreshed.get("model", config.get("model"))
-    if _refreshed.get("custom_providers"):
+    if "custom_providers" in _refreshed:
         config["custom_providers"] = _refreshed["custom_providers"]
+    else:
+        config.pop("custom_providers", None)
 
     # Derive the selected provider for downstream steps (vision setup).
     selected_provider = None
@@ -865,8 +871,6 @@ def setup_model_provider(config: dict, *, quick: bool = False):
                 strategy_value = ["fill_first", "round_robin", "random"][strategy_idx]
                 _set_credential_pool_strategy(config, selected_provider, strategy_value)
                 print_success(f"Saved {selected_provider} rotation strategy: {strategy_value}")
-            else:
-                _set_credential_pool_strategy(config, selected_provider, "fill_first")
         except Exception as exc:
             logger.debug("Could not configure same-provider fallback in setup: %s", exc)
 
@@ -2026,6 +2030,12 @@ def _setup_whatsapp():
         print_info("or personal self-chat) and pair via QR code.")
 
 
+def _setup_weixin():
+    """Configure Weixin (personal WeChat) via iLink Bot API QR login."""
+    from hermes_cli.gateway import _setup_weixin as _gateway_setup_weixin
+    _gateway_setup_weixin()
+
+
 def _setup_bluebubbles():
     """Configure BlueBubbles iMessage gateway."""
     print_header("BlueBubbles (iMessage)")
@@ -2145,6 +2155,7 @@ _GATEWAY_PLATFORMS = [
     ("Matrix", "MATRIX_ACCESS_TOKEN", _setup_matrix),
     ("Mattermost", "MATTERMOST_TOKEN", _setup_mattermost),
     ("WhatsApp", "WHATSAPP_ENABLED", _setup_whatsapp),
+    ("Weixin (WeChat)", "WEIXIN_ACCOUNT_ID", _setup_weixin),
     ("BlueBubbles (iMessage)", "BLUEBUBBLES_SERVER_URL", _setup_bluebubbles),
     ("Webhooks (GitHub, GitLab, etc.)", "WEBHOOK_ENABLED", _setup_webhooks),
 ]
@@ -2703,6 +2714,7 @@ def run_setup_wizard(args):
     Supports full, quick, and section-specific setup:
       hermes setup           — full or quick (auto-detected)
       hermes setup model     — just model/provider
+      hermes setup tts       — just text-to-speech
       hermes setup terminal  — just terminal backend
       hermes setup gateway   — just messaging platforms
       hermes setup tools     — just tool configuration
@@ -2714,6 +2726,11 @@ def run_setup_wizard(args):
         return
     ensure_hermes_home()
 
+    reset_requested = bool(getattr(args, "reset", False))
+    if reset_requested:
+        save_config(copy.deepcopy(DEFAULT_CONFIG))
+        print_success("Configuration reset to defaults.")
+
     config = load_config()
     hermes_home = get_hermes_home()
 
@@ -2814,18 +2831,13 @@ def run_setup_wizard(args):
         menu_choices = [
             "Quick Setup - configure missing items only",
             "Full Setup - reconfigure everything",
-            "---",
             "Model & Provider",
             "Terminal Backend",
             "Messaging Platforms (Gateway)",
             "Tools",
             "Agent Settings",
-            "---",
             "Exit",
         ]
-
-        # Separator indices (not selectable, but prompt_choice doesn't filter them,
-        # so we handle them below)
         choice = prompt_choice("What would you like to do?", menu_choices, 0)
 
         if choice == 0:
@@ -2835,18 +2847,14 @@ def run_setup_wizard(args):
         elif choice == 1:
             # Full setup — fall through to run all sections
             pass
-        elif choice in (2, 8):
-            # Separator — treat as exit
+        elif choice == 7:
             print_info("Exiting. Run 'hermes setup' again when ready.")
             return
-        elif choice == 9:
-            print_info("Exiting. Run 'hermes setup' again when ready.")
-            return
-        elif 3 <= choice <= 7:
+        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 - 3) would dispatch the wrong section.
-            section_key = RETURNING_USER_MENU_SECTION_KEYS[choice - 3]
+            # 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

hermes_cli/skills_config.py

@@ -31,6 +31,7 @@ PLATFORMS = {
     "dingtalk": "💬 DingTalk",
     "feishu": "🪽 Feishu",
     "wecom": "💬 WeCom",
+    "weixin": "💬 Weixin",
     "webhook": "🔗 Webhook",
 }
 

hermes_cli/skills_hub.py

@@ -151,7 +151,8 @@ def do_search(query: str, source: str = "all", limit: int = 10,
 
     auth = GitHubAuth()
     sources = create_source_router(auth)
-    results = unified_search(query, sources, source_filter=source, limit=limit)
+    with c.status("[bold]Searching registries..."):
+        results = unified_search(query, sources, source_filter=source, limit=limit)
 
     if not results:
         c.print("[dim]No skills found matching your query.[/]\n")
@@ -187,7 +188,7 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
     Official skills are always shown first, regardless of source filter.
     """
     from tools.skills_hub import (
-        GitHubAuth, create_source_router,
+        GitHubAuth, create_source_router, parallel_search_sources,
     )
 
     # Clamp page_size to safe range
@@ -198,27 +199,23 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
     auth = GitHubAuth()
     sources = create_source_router(auth)
 
-    # Collect results from all (or filtered) sources
-    # Use empty query to get everything; per-source limits prevent overload
+    # Collect results from all (or filtered) sources in parallel.
+    # Per-source limits are generous — parallelism + 30s timeout cap prevents hangs.
     _TRUST_RANK = {"builtin": 3, "trusted": 2, "community": 1}
-    _PER_SOURCE_LIMIT = {"official": 100, "skills-sh": 100, "well-known": 25, "github": 100, "clawhub": 50,
-                         "claude-marketplace": 50, "lobehub": 50}
+    _PER_SOURCE_LIMIT = {
+        "official": 200, "skills-sh": 200, "well-known": 50,
+        "github": 200, "clawhub": 500, "claude-marketplace": 100,
+        "lobehub": 500,
+    }
 
-    all_results: list = []
-    source_counts: dict = {}
-
-    for src in sources:
-        sid = src.source_id()
-        if source != "all" and sid != source and sid != "official":
-            # Always include official source for the "first" placement
-            continue
-        try:
-            limit = _PER_SOURCE_LIMIT.get(sid, 50)
-            results = src.search("", limit=limit)
-            source_counts[sid] = len(results)
-            all_results.extend(results)
-        except Exception:
-            continue
+    with c.status("[bold]Fetching skills from registries..."):
+        all_results, source_counts, timed_out = parallel_search_sources(
+            sources,
+            query="",
+            per_source_limits=_PER_SOURCE_LIMIT,
+            source_filter=source,
+            overall_timeout=30,
+        )
 
     if not all_results:
         c.print("[dim]No skills found in the Skills Hub.[/]\n")
@@ -252,8 +249,11 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
 
     # Build header
     source_label = f"— {source}" if source != "all" else "— all sources"
+    loaded_label = f"{total} skills loaded"
+    if timed_out:
+        loaded_label += f", {len(timed_out)} source(s) still loading"
     c.print(f"\n[bold]Skills Hub — Browse {source_label}[/]"
-            f"  [dim]({total} skills, page {page}/{total_pages})[/]")
+            f"  [dim]({loaded_label}, page {page}/{total_pages})[/]")
     if official_count > 0 and page == 1:
         c.print(f"[bright_cyan]★ {official_count} official optional skill(s) from Nous Research[/]")
     c.print()
@@ -300,8 +300,11 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
         parts = [f"{sid}: {ct}" for sid, ct in sorted(source_counts.items())]
         c.print(f"  [dim]Sources: {', '.join(parts)}[/]")
 
-    c.print("[dim]Use: hermes skills inspect <identifier> to preview, "
-            "hermes skills install <identifier> to install[/]\n")
+    if timed_out:
+        c.print(f"  [yellow]⚡ Slow sources skipped: {', '.join(timed_out)} "
+                f"— run again for cached results[/]")
+
+    c.print("[dim]Tip: 'hermes skills search <query>' searches deeper across all registries[/]\n")
 
 
 def do_install(identifier: str, category: str = "", force: bool = False,

hermes_cli/status.py

@@ -79,6 +79,9 @@ def _effective_provider_label() -> str:
     return provider_label(effective)
 
 
+from hermes_constants import is_termux as _is_termux
+
+
 def show_status(args):
     """Show status of all Hermes Agent components."""
     show_all = getattr(args, 'all', False)
@@ -302,6 +305,7 @@ def show_status(args):
         "DingTalk": ("DINGTALK_CLIENT_ID", None),
         "Feishu": ("FEISHU_APP_ID", "FEISHU_HOME_CHANNEL"),
         "WeCom": ("WECOM_BOT_ID", "WECOM_HOME_CHANNEL"),
+        "Weixin": ("WEIXIN_ACCOUNT_ID", "WEIXIN_HOME_CHANNEL"),
         "BlueBubbles": ("BLUEBUBBLES_SERVER_URL", "BLUEBUBBLES_HOME_CHANNEL"),
     }
     
@@ -325,7 +329,25 @@ def show_status(args):
     print()
     print(color("◆ Gateway Service", Colors.CYAN, Colors.BOLD))
     
-    if sys.platform.startswith('linux'):
+    if _is_termux():
+        try:
+            from hermes_cli.gateway import find_gateway_pids
+            gateway_pids = find_gateway_pids()
+        except Exception:
+            gateway_pids = []
+        is_running = bool(gateway_pids)
+        print(f"  Status:       {check_mark(is_running)} {'running' if is_running else 'stopped'}")
+        print("  Manager:      Termux / manual process")
+        if gateway_pids:
+            rendered = ", ".join(str(pid) for pid in gateway_pids[:3])
+            if len(gateway_pids) > 3:
+                rendered += ", ..."
+            print(f"  PID(s):       {rendered}")
+        else:
+            print("  Start with:   hermes gateway")
+            print("  Note:         Android may stop background jobs when Termux is suspended")
+
+    elif sys.platform.startswith('linux'):
         try:
             from hermes_cli.gateway import get_service_name
             _gw_svc = get_service_name()
@@ -339,7 +361,7 @@ def show_status(args):
                 timeout=5
             )
             is_active = result.stdout.strip() == "active"
-        except subprocess.TimeoutExpired:
+        except (FileNotFoundError, subprocess.TimeoutExpired):
             is_active = False
         print(f"  Status:       {check_mark(is_active)} {'running' if is_active else 'stopped'}")
         print("  Manager:      systemd (user)")

hermes_cli/tools_config.py

@@ -133,6 +133,7 @@ PLATFORMS = {
  "dingtalk": {"label": "💬 DingTalk", "default_toolset": "hermes-dingtalk"},
     "feishu": {"label": "🪽 Feishu", "default_toolset": "hermes-feishu"},
     "wecom": {"label": "💬 WeCom", "default_toolset": "hermes-wecom"},
+    "weixin": {"label": "💬 Weixin", "default_toolset": "hermes-weixin"},
     "api_server": {"label": "🌐 API Server", "default_toolset": "hermes-api-server"},
     "mattermost": {"label": "💬 Mattermost", "default_toolset": "hermes-mattermost"},
     "webhook": {"label": "🔗 Webhook", "default_toolset": "hermes-webhook"},
@@ -720,6 +721,8 @@ def _prompt_choice(question: str, choices: list, default: int = 0) -> int:
                     return
 
         curses.wrapper(_curses_menu)
+        from hermes_cli.curses_ui import flush_stdin
+        flush_stdin()
         return result_holder[0]
 
     except Exception:

hermes_cli/uninstall.py

@@ -6,6 +6,8 @@ Provides options for:
 - Keep data: Remove code but keep ~/.hermes/ (configs, sessions, logs)
 """
 
+import os
+import platform
 import shutil
 import subprocess
 from pathlib import Path
@@ -122,6 +124,10 @@ def uninstall_gateway_service():
     
     if platform.system() != "Linux":
         return False
+
+    prefix = os.getenv("PREFIX", "")
+    if os.getenv("TERMUX_VERSION") or "com.termux/files/usr" in prefix:
+        return False
     
     try:
         from hermes_cli.gateway import get_service_name

hermes_constants.py

@@ -17,6 +17,45 @@ def get_hermes_home() -> Path:
     return Path(os.getenv("HERMES_HOME", Path.home() / ".hermes"))
 
 
+def get_default_hermes_root() -> Path:
+    """Return the root Hermes directory for profile-level operations.
+
+    In standard deployments this is ``~/.hermes``.
+
+    In Docker or custom deployments where ``HERMES_HOME`` points outside
+    ``~/.hermes`` (e.g. ``/opt/data``), returns ``HERMES_HOME`` directly
+    — that IS the root.
+
+    In profile mode where ``HERMES_HOME`` is ``<root>/profiles/<name>``,
+    returns ``<root>`` so that ``profile list`` can see all profiles.
+    Works both for standard (``~/.hermes/profiles/coder``) and Docker
+    (``/opt/data/profiles/coder``) layouts.
+
+    Import-safe — no dependencies beyond stdlib.
+    """
+    native_home = Path.home() / ".hermes"
+    env_home = os.environ.get("HERMES_HOME", "")
+    if not env_home:
+        return native_home
+    env_path = Path(env_home)
+    try:
+        env_path.resolve().relative_to(native_home.resolve())
+        # HERMES_HOME is under ~/.hermes (normal or profile mode)
+        return native_home
+    except ValueError:
+        pass
+
+    # Docker / custom deployment.
+    # Check if this is a profile path: <root>/profiles/<name>
+    # If the immediate parent dir is named "profiles", the root is
+    # the grandparent — this covers Docker profiles correctly.
+    if env_path.parent.name == "profiles":
+        return env_path.parent.parent
+
+    # Not a profile path — HERMES_HOME itself is the root
+    return env_path
+
+
 def get_optional_skills_dir(default: Path | None = None) -> Path:
     """Return the optional-skills directory, honoring package-manager wrappers.
 
@@ -72,6 +111,32 @@ def display_hermes_home() -> str:
         return str(home)
 
 
+def get_subprocess_home() -> str | None:
+    """Return a per-profile HOME directory for subprocesses, or None.
+
+    When ``{HERMES_HOME}/home/`` exists on disk, subprocesses should use it
+    as ``HOME`` so system tools (git, ssh, gh, npm …) write their configs
+    inside the Hermes data directory instead of the OS-level ``/root`` or
+    ``~/``.  This provides:
+
+    * **Docker persistence** — tool configs land inside the persistent volume.
+    * **Profile isolation** — each profile gets its own git identity, SSH
+      keys, gh tokens, etc.
+
+    The Python process's own ``os.environ["HOME"]`` and ``Path.home()`` are
+    **never** modified — only subprocess environments should inject this value.
+    Activation is directory-based: if the ``home/`` subdirectory doesn't
+    exist, returns ``None`` and behavior is unchanged.
+    """
+    hermes_home = os.getenv("HERMES_HOME")
+    if not hermes_home:
+        return None
+    profile_home = os.path.join(hermes_home, "home")
+    if os.path.isdir(profile_home):
+        return profile_home
+    return None
+
+
 VALID_REASONING_EFFORTS = ("minimal", "low", "medium", "high", "xhigh")
 
 
@@ -93,6 +158,16 @@ def parse_reasoning_effort(effort: str) -> dict | None:
     return None
 
 
+def is_termux() -> bool:
+    """Return True when running inside a Termux (Android) environment.
+
+    Checks ``TERMUX_VERSION`` (set by Termux) or the Termux-specific
+    ``PREFIX`` path.  Import-safe — no heavy deps.
+    """
+    prefix = os.getenv("PREFIX", "")
+    return bool(os.getenv("TERMUX_VERSION") or "com.termux/files/usr" in prefix)
+
+
 OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1"
 OPENROUTER_MODELS_URL = f"{OPENROUTER_BASE_URL}/models"
 

pyproject.toml

@@ -16,7 +16,7 @@ dependencies = [
   "anthropic>=0.39.0,<1",
   "python-dotenv>=1.2.1,<2",
   "fire>=0.7.1,<1",
-  "httpx>=0.28.1,<1",
+  "httpx[socks]>=0.28.1,<1",
   "rich>=14.3.3,<15",
   "tenacity>=9.1.4,<10",
   "pyyaml>=6.0.2,<7",
@@ -63,6 +63,17 @@ homeassistant = ["aiohttp>=3.9.0,<4"]
 sms = ["aiohttp>=3.9.0,<4"]
 acp = ["agent-client-protocol>=0.9.0,<1.0"]
 mistral = ["mistralai>=2.3.0,<3"]
+termux = [
+  # Tested Android / Termux path: keeps the core CLI feature-rich while
+  # avoiding extras that currently depend on non-Android wheels (notably
+  # faster-whisper -> ctranslate2 via the voice extra).
+  "hermes-agent[cron]",
+  "hermes-agent[cli]",
+  "hermes-agent[pty]",
+  "hermes-agent[mcp]",
+  "hermes-agent[honcho]",
+  "hermes-agent[acp]",
+]
 dingtalk = ["dingtalk-stream>=0.1.0,<1"]
 feishu = ["lark-oapi>=1.5.3,<2"]
 rl = [
@@ -77,10 +88,10 @@ all = [
   "hermes-agent[modal]",
   "hermes-agent[daytona]",
   "hermes-agent[messaging]",
-  # matrix excluded: python-olm (required by matrix-nio[e2e]) is upstream-broken
-  # on modern macOS (archived libolm, C++ errors with Clang 21+). Including it
-  # here causes the entire [all] install to fail, dropping all other extras.
-  # Users who need Matrix can install manually: pip install 'hermes-agent[matrix]'
+  # matrix: python-olm (required by matrix-nio[e2e]) is upstream-broken on
+  # modern macOS (archived libolm, C++ errors with Clang 21+).  On Linux the
+  # [matrix] extra's own marker pulls in the [e2e] variant automatically.
+  "hermes-agent[matrix]; sys_platform == 'linux'",
   "hermes-agent[cron]",
   "hermes-agent[cli]",
   "hermes-agent[dev]",

run_agent.py

@@ -359,8 +359,9 @@ def _sanitize_surrogates(text: str) -> str:
 def _sanitize_messages_surrogates(messages: list) -> bool:
     """Sanitize surrogate characters from all string content in a messages list.
 
-    Walks message dicts in-place.  Returns True if any surrogates were found
-    and replaced, False otherwise.
+    Walks message dicts in-place. Returns True if any surrogates were found
+    and replaced, False otherwise. Covers content/text, name, and tool call
+    metadata/arguments so retries don't fail on a non-content field.
     """
     found = False
     for msg in messages:
@@ -377,6 +378,88 @@ def _sanitize_messages_surrogates(messages: list) -> bool:
                     if isinstance(text, str) and _SURROGATE_RE.search(text):
                         part["text"] = _SURROGATE_RE.sub('\ufffd', text)
                         found = True
+        name = msg.get("name")
+        if isinstance(name, str) and _SURROGATE_RE.search(name):
+            msg["name"] = _SURROGATE_RE.sub('\ufffd', name)
+            found = True
+        tool_calls = msg.get("tool_calls")
+        if isinstance(tool_calls, list):
+            for tc in tool_calls:
+                if not isinstance(tc, dict):
+                    continue
+                tc_id = tc.get("id")
+                if isinstance(tc_id, str) and _SURROGATE_RE.search(tc_id):
+                    tc["id"] = _SURROGATE_RE.sub('\ufffd', tc_id)
+                    found = True
+                fn = tc.get("function")
+                if isinstance(fn, dict):
+                    fn_name = fn.get("name")
+                    if isinstance(fn_name, str) and _SURROGATE_RE.search(fn_name):
+                        fn["name"] = _SURROGATE_RE.sub('\ufffd', fn_name)
+                        found = True
+                    fn_args = fn.get("arguments")
+                    if isinstance(fn_args, str) and _SURROGATE_RE.search(fn_args):
+                        fn["arguments"] = _SURROGATE_RE.sub('\ufffd', fn_args)
+                        found = True
+    return found
+
+
+def _strip_non_ascii(text: str) -> str:
+    """Remove non-ASCII characters, replacing with closest ASCII equivalent or removing.
+
+    Used as a last resort when the system encoding is ASCII and can't handle
+    any non-ASCII characters (e.g. LANG=C on Chromebooks).
+    """
+    return text.encode('ascii', errors='ignore').decode('ascii')
+
+
+def _sanitize_messages_non_ascii(messages: list) -> bool:
+    """Strip non-ASCII characters from all string content in a messages list.
+
+    This is a last-resort recovery for systems with ASCII-only encoding
+    (LANG=C, Chromebooks, minimal containers).  Returns True if any
+    non-ASCII content was found and sanitized.
+    """
+    found = False
+    for msg in messages:
+        if not isinstance(msg, dict):
+            continue
+        # Sanitize content (string)
+        content = msg.get("content")
+        if isinstance(content, str):
+            sanitized = _strip_non_ascii(content)
+            if sanitized != content:
+                msg["content"] = sanitized
+                found = True
+        elif isinstance(content, list):
+            for part in content:
+                if isinstance(part, dict):
+                    text = part.get("text")
+                    if isinstance(text, str):
+                        sanitized = _strip_non_ascii(text)
+                        if sanitized != text:
+                            part["text"] = sanitized
+                            found = True
+        # Sanitize name field (can contain non-ASCII in tool results)
+        name = msg.get("name")
+        if isinstance(name, str):
+            sanitized = _strip_non_ascii(name)
+            if sanitized != name:
+                msg["name"] = sanitized
+                found = True
+        # Sanitize tool_calls
+        tool_calls = msg.get("tool_calls")
+        if isinstance(tool_calls, list):
+            for tc in tool_calls:
+                if isinstance(tc, dict):
+                    fn = tc.get("function", {})
+                    if isinstance(fn, dict):
+                        fn_args = fn.get("arguments")
+                        if isinstance(fn_args, str):
+                            sanitized = _strip_non_ascii(fn_args)
+                            if sanitized != fn_args:
+                                fn["arguments"] = sanitized
+                                found = True
     return found
 
 
@@ -500,6 +583,8 @@ class AIAgent:
         status_callback: callable = None,
         max_tokens: int = None,
         reasoning_config: Dict[str, Any] = None,
+        service_tier: str = None,
+        request_overrides: Dict[str, Any] = None,
         prefill_messages: List[Dict[str, Any]] = None,
         platform: str = None,
         user_id: str = None,
@@ -604,6 +689,17 @@ class AIAgent:
         else:
             self.api_mode = "chat_completions"
 
+        try:
+            from hermes_cli.model_normalize import (
+                _AGGREGATOR_PROVIDERS,
+                normalize_model_for_provider,
+            )
+
+            if self.provider not in _AGGREGATOR_PROVIDERS:
+                self.model = normalize_model_for_provider(self.model, self.provider)
+        except Exception:
+            pass
+
         # Direct OpenAI sessions use the Responses API path.  GPT-5.x tool
         # calls with reasoning are rejected on /v1/chat/completions, and
         # Hermes is a tool-using client by default.
@@ -622,6 +718,7 @@ class AIAgent:
         self.tool_progress_callback = tool_progress_callback
         self.tool_start_callback = tool_start_callback
         self.tool_complete_callback = tool_complete_callback
+        self.suppress_status_output = False
         self.thinking_callback = thinking_callback
         self.reasoning_callback = reasoning_callback
         self.clarify_callback = clarify_callback
@@ -660,6 +757,8 @@ class AIAgent:
         # Model response configuration
         self.max_tokens = max_tokens  # None = use model default
         self.reasoning_config = reasoning_config  # None = use default (medium for OpenRouter)
+        self.service_tier = service_tier
+        self.request_overrides = dict(request_overrides or {})
         self.prefill_messages = prefill_messages or []  # Prefilled conversation turns
         
         # Anthropic prompt caching: auto-enabled for Claude models via OpenRouter.
@@ -788,7 +887,7 @@ class AIAgent:
                     client_kwargs["default_headers"] = copilot_default_headers()
                 elif "api.kimi.com" in effective_base.lower():
                     client_kwargs["default_headers"] = {
-                        "User-Agent": "KimiCLI/1.3",
+                        "User-Agent": "KimiCLI/1.30.0",
                     }
                 elif "portal.qwen.ai" in effective_base.lower():
                     client_kwargs["default_headers"] = _qwen_portal_headers()
@@ -848,6 +947,7 @@ class AIAgent:
                     client_kwargs["default_headers"] = headers
 
             self.api_key = client_kwargs.get("api_key", "")
+            self.base_url = client_kwargs.get("base_url", self.base_url)
             try:
                 self.client = self._create_openai_client(client_kwargs, reason="agent_init", shared=True)
                 if not self.quiet_mode:
@@ -1144,6 +1244,9 @@ class AIAgent:
             except (TypeError, ValueError):
                 _config_context_length = None
 
+        # Store for reuse in switch_model (so config override persists across model switches)
+        self._config_context_length = _config_context_length
+
         # Check custom_providers per-model context_length
         if _config_context_length is None:
             _custom_providers = _agent_cfg.get("custom_providers")
@@ -1381,6 +1484,7 @@ class AIAgent:
                 base_url=self.base_url,
                 api_key=self.api_key,
                 provider=self.provider,
+                config_context_length=getattr(self, "_config_context_length", None),
             )
             self.context_compressor.model = self.model
             self.context_compressor.base_url = self.base_url
@@ -1458,7 +1562,14 @@ class AIAgent:
         After the main response has been delivered and the remaining tool
         calls are post-response housekeeping (``_mute_post_response``),
         all non-forced output is suppressed.
+
+        ``suppress_status_output`` is a stricter CLI automation mode used by
+        parseable single-query flows such as ``hermes chat -q``. In that mode,
+        all status/diagnostic prints routed through ``_vprint`` are suppressed
+        so stdout stays machine-readable.
         """
+        if getattr(self, "suppress_status_output", False):
+            return
         if not force and getattr(self, "_mute_post_response", False):
             return
         if not force and self._has_stream_consumers() and not self._executing_tools:
@@ -1484,6 +1595,17 @@ class AIAgent:
         except (AttributeError, ValueError, OSError):
             return False
 
+    def _should_emit_quiet_tool_messages(self) -> bool:
+        """Return True when quiet-mode tool summaries should print directly.
+
+        When the caller provides ``tool_progress_callback`` (for example the CLI
+        TUI or a gateway progress renderer), that callback owns progress display.
+        Emitting quiet-mode summary lines here duplicates progress and leaks tool
+        previews into flows that are expected to stay silent, such as
+        ``hermes chat -q``.
+        """
+        return self.quiet_mode and not self.tool_progress_callback
+
     def _emit_status(self, message: str) -> None:
         """Emit a lifecycle status message to both CLI and gateway channels.
 
@@ -1855,19 +1977,14 @@ class AIAgent:
             except Exception as e:
                 logger.debug("Background memory/skill review failed: %s", e)
             finally:
-                # Explicitly close the OpenAI/httpx client so GC doesn't
-                # try to clean it up on a dead asyncio event loop (which
-                # produces "Event loop is closed" errors in the terminal).
+                # 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).
                 if review_agent is not None:
-                    client = getattr(review_agent, "client", None)
-                    if client is not None:
-                        try:
-                            review_agent._close_openai_client(
-                                client, reason="bg_review_done", shared=True
-                            )
-                            review_agent.client = None
-                        except Exception:
-                            pass
+                    try:
+                        review_agent.close()
+                    except Exception:
+                        pass
 
         t = threading.Thread(target=_run_review, daemon=True, name="bg-review")
         t.start()
@@ -2607,6 +2724,64 @@ class AIAgent:
             except Exception:
                 pass
     
+    def close(self) -> None:
+        """Release all resources held by this agent instance.
+
+        Cleans up subprocess resources that would otherwise become orphans:
+        - Background processes tracked in ProcessRegistry
+        - Terminal sandbox environments
+        - Browser daemon sessions
+        - Active child agents (subagent delegation)
+        - OpenAI/httpx client connections
+
+        Safe to call multiple times (idempotent).  Each cleanup step is
+        independently guarded so a failure in one does not prevent the rest.
+        """
+        task_id = getattr(self, "session_id", None) or ""
+
+        # 1. Kill background processes for this task
+        try:
+            from tools.process_registry import process_registry
+            process_registry.kill_all(task_id=task_id)
+        except Exception:
+            pass
+
+        # 2. Clean terminal sandbox environments
+        try:
+            from tools.terminal_tool import cleanup_vm
+            cleanup_vm(task_id)
+        except Exception:
+            pass
+
+        # 3. Clean browser daemon sessions
+        try:
+            from tools.browser_tool import cleanup_browser
+            cleanup_browser(task_id)
+        except Exception:
+            pass
+
+        # 4. Close active child agents
+        try:
+            with self._active_children_lock:
+                children = list(self._active_children)
+                self._active_children.clear()
+            for child in children:
+                try:
+                    child.close()
+                except Exception:
+                    pass
+        except Exception:
+            pass
+
+        # 5. Close the OpenAI/httpx client
+        try:
+            client = getattr(self, "client", None)
+            if client is not None:
+                self._close_openai_client(client, reason="agent_close", shared=True)
+                self.client = None
+        except Exception:
+            pass
+
     def _hydrate_todo_store(self, history: List[Dict[str, Any]]) -> None:
         """
         Recover todo state from conversation history.
@@ -2899,7 +3074,7 @@ class AIAgent:
 
     @staticmethod
     def _cap_delegate_task_calls(tool_calls: list) -> list:
-        """Truncate excess delegate_task calls to MAX_CONCURRENT_CHILDREN.
+        """Truncate excess delegate_task calls to max_concurrent_children.
 
         The delegate_tool caps the task list inside a single call, but the
         model can emit multiple separate delegate_task tool_calls in one
@@ -2907,23 +3082,24 @@ class AIAgent:
 
         Returns the original list if no truncation was needed.
         """
-        from tools.delegate_tool import MAX_CONCURRENT_CHILDREN
+        from tools.delegate_tool import _get_max_concurrent_children
+        max_children = _get_max_concurrent_children()
         delegate_count = sum(1 for tc in tool_calls if tc.function.name == "delegate_task")
-        if delegate_count <= MAX_CONCURRENT_CHILDREN:
+        if delegate_count <= max_children:
             return tool_calls
         kept_delegates = 0
         truncated = []
         for tc in tool_calls:
             if tc.function.name == "delegate_task":
-                if kept_delegates < MAX_CONCURRENT_CHILDREN:
+                if kept_delegates < max_children:
                     truncated.append(tc)
                     kept_delegates += 1
             else:
                 truncated.append(tc)
         logger.warning(
             "Truncated %d excess delegate_task call(s) to enforce "
-            "MAX_CONCURRENT_CHILDREN=%d limit",
-            delegate_count - MAX_CONCURRENT_CHILDREN, MAX_CONCURRENT_CHILDREN,
+            "max_concurrent_children=%d limit",
+            delegate_count - max_children, max_children,
         )
         return truncated
 
@@ -3322,7 +3498,7 @@ class AIAgent:
         allowed_keys = {
             "model", "instructions", "input", "tools", "store",
             "reasoning", "include", "max_output_tokens", "temperature",
-            "tool_choice", "parallel_tool_calls", "prompt_cache_key",
+            "tool_choice", "parallel_tool_calls", "prompt_cache_key", "service_tier",
         }
         normalized: Dict[str, Any] = {
             "model": model,
@@ -3340,6 +3516,9 @@ class AIAgent:
         include = api_kwargs.get("include")
         if isinstance(include, list):
             normalized["include"] = include
+        service_tier = api_kwargs.get("service_tier")
+        if isinstance(service_tier, str) and service_tier.strip():
+            normalized["service_tier"] = service_tier.strip()
 
         # Pass through max_output_tokens and temperature
         max_output_tokens = api_kwargs.get("max_output_tokens")
@@ -4152,7 +4331,7 @@ class AIAgent:
 
             self._client_kwargs["default_headers"] = copilot_default_headers()
         elif "api.kimi.com" in normalized:
-            self._client_kwargs["default_headers"] = {"User-Agent": "KimiCLI/1.3"}
+            self._client_kwargs["default_headers"] = {"User-Agent": "KimiCLI/1.30.0"}
         elif "portal.qwen.ai" in normalized:
             self._client_kwargs["default_headers"] = _qwen_portal_headers()
         else:
@@ -4190,49 +4369,80 @@ class AIAgent:
         *,
         status_code: Optional[int],
         has_retried_429: bool,
+        classified_reason: Optional[FailoverReason] = None,
         error_context: Optional[Dict[str, Any]] = None,
     ) -> tuple[bool, bool]:
         """Attempt credential recovery via pool rotation.
 
         Returns (recovered, has_retried_429).
-        On 429: first occurrence retries same credential (sets flag True).
-                second consecutive 429 rotates to next credential (resets flag).
-        On 402: immediately rotates (billing exhaustion won't resolve with retry).
-        On 401: attempts token refresh before rotating.
+        On rate limits: first occurrence retries same credential (sets flag True).
+                        second consecutive failure rotates to next credential.
+        On billing exhaustion: immediately rotates.
+        On auth failures: attempts token refresh before rotating.
+
+        `classified_reason` lets the recovery path honor the structured error
+        classifier instead of relying only on raw HTTP codes. This matters for
+        providers that surface billing/rate-limit/auth conditions under a
+        different status code, such as Anthropic returning HTTP 400 for
+        "out of extra usage".
         """
         pool = self._credential_pool
-        if pool is None or status_code is None:
+        if pool is None:
             return False, has_retried_429
 
-        if status_code == 402:
-            next_entry = pool.mark_exhausted_and_rotate(status_code=402, error_context=error_context)
+        effective_reason = classified_reason
+        if effective_reason is None:
+            if status_code == 402:
+                effective_reason = FailoverReason.billing
+            elif status_code == 429:
+                effective_reason = FailoverReason.rate_limit
+            elif status_code == 401:
+                effective_reason = FailoverReason.auth
+
+        if effective_reason == FailoverReason.billing:
+            rotate_status = status_code if status_code is not None else 402
+            next_entry = pool.mark_exhausted_and_rotate(status_code=rotate_status, error_context=error_context)
             if next_entry is not None:
-                logger.info(f"Credential 402 (billing) — rotated to pool entry {getattr(next_entry, 'id', '?')}")
+                logger.info(
+                    "Credential %s (billing) — rotated to pool entry %s",
+                    rotate_status,
+                    getattr(next_entry, "id", "?"),
+                )
                 self._swap_credential(next_entry)
                 return True, False
             return False, has_retried_429
 
-        if status_code == 429:
+        if effective_reason == FailoverReason.rate_limit:
             if not has_retried_429:
                 return False, True
-            next_entry = pool.mark_exhausted_and_rotate(status_code=429, error_context=error_context)
+            rotate_status = status_code if status_code is not None else 429
+            next_entry = pool.mark_exhausted_and_rotate(status_code=rotate_status, error_context=error_context)
             if next_entry is not None:
-                logger.info(f"Credential 429 (rate limit) — rotated to pool entry {getattr(next_entry, 'id', '?')}")
+                logger.info(
+                    "Credential %s (rate limit) — rotated to pool entry %s",
+                    rotate_status,
+                    getattr(next_entry, "id", "?"),
+                )
                 self._swap_credential(next_entry)
                 return True, False
             return False, True
 
-        if status_code == 401:
+        if effective_reason == FailoverReason.auth:
             refreshed = pool.try_refresh_current()
             if refreshed is not None:
-                logger.info(f"Credential 401 — refreshed pool entry {getattr(refreshed, 'id', '?')}")
+                logger.info(f"Credential auth failure — refreshed pool entry {getattr(refreshed, 'id', '?')}")
                 self._swap_credential(refreshed)
                 return True, has_retried_429
             # Refresh failed — rotate to next credential instead of giving up.
             # The failed entry is already marked exhausted by try_refresh_current().
-            next_entry = pool.mark_exhausted_and_rotate(status_code=401, error_context=error_context)
+            rotate_status = status_code if status_code is not None else 401
+            next_entry = pool.mark_exhausted_and_rotate(status_code=rotate_status, error_context=error_context)
             if next_entry is not None:
-                logger.info(f"Credential 401 (refresh failed) — rotated to pool entry {getattr(next_entry, 'id', '?')}")
+                logger.info(
+                    "Credential %s (auth refresh failed) — rotated to pool entry %s",
+                    rotate_status,
+                    getattr(next_entry, "id", "?"),
+                )
                 self._swap_credential(next_entry)
                 return True, False
 
@@ -4403,7 +4613,17 @@ class AIAgent:
             """Stream a chat completions response."""
             import httpx as _httpx
             _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
-            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 60.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            # Local providers (Ollama, llama.cpp, vLLM) can take minutes for
+            # prefill on large contexts before producing the first token.
+            # Auto-increase the httpx read timeout unless the user explicitly
+            # overrode HERMES_STREAM_READ_TIMEOUT.
+            if _stream_read_timeout == 120.0 and self.base_url and is_local_endpoint(self.base_url):
+                _stream_read_timeout = _base_timeout
+                logger.debug(
+                    "Local provider detected (%s) — stream read timeout raised to %.0fs",
+                    self.base_url, _stream_read_timeout,
+                )
             stream_kwargs = {
                 **api_kwargs,
                 "stream": True,
@@ -4557,20 +4777,31 @@ class AIAgent:
             # Build mock response matching non-streaming shape
             full_content = "".join(content_parts) or None
             mock_tool_calls = None
+            has_truncated_tool_args = False
             if tool_calls_acc:
                 mock_tool_calls = []
                 for idx in sorted(tool_calls_acc):
                     tc = tool_calls_acc[idx]
+                    arguments = tc["function"]["arguments"]
+                    if arguments and arguments.strip():
+                        try:
+                            json.loads(arguments)
+                        except json.JSONDecodeError:
+                            has_truncated_tool_args = True
                     mock_tool_calls.append(SimpleNamespace(
                         id=tc["id"],
                         type=tc["type"],
                         extra_content=tc.get("extra_content"),
                         function=SimpleNamespace(
                             name=tc["function"]["name"],
-                            arguments=tc["function"]["arguments"],
+                            arguments=arguments,
                         ),
                     ))
 
+            effective_finish_reason = finish_reason or "stop"
+            if has_truncated_tool_args:
+                effective_finish_reason = "length"
+
             full_reasoning = "".join(reasoning_parts) or None
             mock_message = SimpleNamespace(
                 role=role,
@@ -4581,7 +4812,7 @@ class AIAgent:
             mock_choice = SimpleNamespace(
                 index=0,
                 message=mock_message,
-                finish_reason=finish_reason or "stop",
+                finish_reason=effective_finish_reason,
             )
             return SimpleNamespace(
                 id="stream-" + str(uuid.uuid4()),
@@ -4605,6 +4836,14 @@ class AIAgent:
             # Use the Anthropic SDK's streaming context manager
             with self._anthropic_client.messages.stream(**api_kwargs) as stream:
                 for event in stream:
+                    # Update stale-stream timer on every event so the
+                    # outer poll loop knows data is flowing.  Without
+                    # this, the detector kills healthy long-running
+                    # Opus streams after 180 s even when events are
+                    # actively arriving (the chat_completions path
+                    # already does this at the top of its chunk loop).
+                    last_chunk_time["t"] = time.time()
+
                     if self._interrupt_requested:
                         break
 
@@ -4628,6 +4867,7 @@ class AIAgent:
                                 if text and not has_tool_use:
                                     _fire_first_delta()
                                     self._fire_stream_delta(text)
+                                    deltas_were_sent["yes"] = True
                             elif delta_type == "thinking_delta":
                                 thinking_text = getattr(delta, "thinking", "")
                                 if thinking_text:
@@ -4918,7 +5158,7 @@ class AIAgent:
             # when no explicit key is in the fallback config.
             if fb_base_url_hint and "ollama.com" in fb_base_url_hint.lower() and not fb_api_key_hint:
                 fb_api_key_hint = os.getenv("OLLAMA_API_KEY") or None
-            fb_client, _ = resolve_provider_client(
+            fb_client, _resolved_fb_model = resolve_provider_client(
                 fb_provider, model=fb_model, raw_codex=True,
                 explicit_base_url=fb_base_url_hint,
                 explicit_api_key=fb_api_key_hint)
@@ -4927,6 +5167,12 @@ class AIAgent:
                     "Fallback to %s failed: provider not configured",
                     fb_provider)
                 return self._try_activate_fallback()  # try next in chain
+            try:
+                from hermes_cli.model_normalize import normalize_model_for_provider
+
+                fb_model = normalize_model_for_provider(fb_model, fb_provider)
+            except Exception:
+                pass
 
             # Determine api_mode from provider / base URL
             fb_api_mode = "chat_completions"
@@ -5087,6 +5333,7 @@ class AIAgent:
     _TRANSIENT_TRANSPORT_ERRORS = frozenset({
         "ReadTimeout", "ConnectTimeout", "PoolTimeout",
         "ConnectError", "RemoteProtocolError",
+        "APIConnectionError", "APITimeoutError",
     })
 
     def _try_recover_primary_transport(
@@ -5410,6 +5657,7 @@ class AIAgent:
                 preserve_dots=self._anthropic_preserve_dots(),
                 context_length=ctx_len,
                 base_url=getattr(self, "_anthropic_base_url", None),
+                fast_mode=(self.request_overrides or {}).get("speed") == "fast",
             )
 
         if self.api_mode == "codex_responses":
@@ -5425,6 +5673,10 @@ class AIAgent:
                 "models.github.ai" in self.base_url.lower()
                 or "api.githubcopilot.com" in self.base_url.lower()
             )
+            is_codex_backend = (
+                self.provider == "openai-codex"
+                or "chatgpt.com/backend-api/codex" in self.base_url.lower()
+            )
 
             # Resolve reasoning effort: config > default (medium)
             reasoning_effort = "medium"
@@ -5462,7 +5714,10 @@ class AIAgent:
             elif not is_github_responses:
                 kwargs["include"] = []
 
-            if self.max_tokens is not None:
+            if self.request_overrides:
+                kwargs.update(self.request_overrides)
+
+            if self.max_tokens is not None and not is_codex_backend:
                 kwargs["max_output_tokens"] = self.max_tokens
 
             return kwargs
@@ -5557,20 +5812,20 @@ class AIAgent:
         if self.max_tokens is not None:
             if not self._is_qwen_portal():
                 api_kwargs.update(self._max_tokens_param(self.max_tokens))
-        elif self._is_openrouter_url() and "claude" in (self.model or "").lower():
-            # OpenRouter translates requests to Anthropic's Messages API,
-            # which requires max_tokens as a mandatory field.  When we omit
-            # it, OpenRouter picks a default that can be too low — the model
-            # spends its output budget on thinking and has almost nothing
-            # left for the actual response (especially large tool calls like
-            # write_file).  Sending the model's real output limit ensures
-            # full capacity.  Other providers handle the default fine.
+        elif (self._is_openrouter_url() or "nousresearch" in self._base_url_lower) and "claude" in (self.model or "").lower():
+            # OpenRouter and Nous Portal translate requests to Anthropic's
+            # Messages API, which requires max_tokens as a mandatory field.
+            # When we omit it, the proxy picks a default that can be too
+            # low — the model spends its output budget on thinking and has
+            # almost nothing left for the actual response (especially large
+            # tool calls like write_file).  Sending the model's real output
+            # limit ensures full capacity.
             try:
                 from agent.anthropic_adapter import _get_anthropic_max_output
                 _model_output_limit = _get_anthropic_max_output(self.model)
                 api_kwargs["max_tokens"] = _model_output_limit
             except Exception:
-                pass  # fail open — let OpenRouter pick its default
+                pass  # fail open — let the proxy pick its default
 
         extra_body = {}
 
@@ -5633,6 +5888,11 @@ class AIAgent:
         if "x.ai" in self._base_url_lower and hasattr(self, "session_id") and self.session_id:
             api_kwargs["extra_headers"] = {"x-grok-conv-id": self.session_id}
 
+        # Priority Processing / generic request overrides (e.g. service_tier).
+        # Applied last so overrides win over any defaults set above.
+        if self.request_overrides:
+            api_kwargs.update(self.request_overrides)
+
         return api_kwargs
 
     def _supports_reasoning_extra_body(self) -> bool:
@@ -6338,7 +6598,7 @@ class AIAgent:
 
         # Start spinner for CLI mode (skip when TUI handles tool progress)
         spinner = None
-        if self.quiet_mode and not self.tool_progress_callback and self._should_start_quiet_spinner():
+        if self._should_emit_quiet_tool_messages() and self._should_start_quiet_spinner():
             face = random.choice(KawaiiSpinner.KAWAII_WAITING)
             spinner = KawaiiSpinner(f"{face} ⚡ running {num_tools} tools concurrently", spinner_type='dots', print_fn=self._print_fn)
             spinner.start()
@@ -6388,7 +6648,7 @@ class AIAgent:
                     logging.debug(f"Tool result ({len(function_result)} chars): {function_result}")
 
             # Print cute message per tool
-            if self.quiet_mode:
+            if self._should_emit_quiet_tool_messages():
                 cute_msg = _get_cute_tool_message_impl(name, args, tool_duration, result=function_result)
                 self._safe_print(f"  {cute_msg}")
             elif not self.quiet_mode:
@@ -6545,7 +6805,7 @@ class AIAgent:
                     store=self._todo_store,
                 )
                 tool_duration = time.time() - tool_start_time
-                if self.quiet_mode:
+                if self._should_emit_quiet_tool_messages():
                     self._vprint(f"  {_get_cute_tool_message_impl('todo', function_args, tool_duration, result=function_result)}")
             elif function_name == "session_search":
                 if not self._session_db:
@@ -6560,7 +6820,7 @@ class AIAgent:
                         current_session_id=self.session_id,
                     )
                 tool_duration = time.time() - tool_start_time
-                if self.quiet_mode:
+                if self._should_emit_quiet_tool_messages():
                     self._vprint(f"  {_get_cute_tool_message_impl('session_search', function_args, tool_duration, result=function_result)}")
             elif function_name == "memory":
                 target = function_args.get("target", "memory")
@@ -6573,7 +6833,7 @@ class AIAgent:
                     store=self._memory_store,
                 )
                 tool_duration = time.time() - tool_start_time
-                if self.quiet_mode:
+                if self._should_emit_quiet_tool_messages():
                     self._vprint(f"  {_get_cute_tool_message_impl('memory', function_args, tool_duration, result=function_result)}")
             elif function_name == "clarify":
                 from tools.clarify_tool import clarify_tool as _clarify_tool
@@ -6583,7 +6843,7 @@ class AIAgent:
                     callback=self.clarify_callback,
                 )
                 tool_duration = time.time() - tool_start_time
-                if self.quiet_mode:
+                if self._should_emit_quiet_tool_messages():
                     self._vprint(f"  {_get_cute_tool_message_impl('clarify', function_args, tool_duration, result=function_result)}")
             elif function_name == "delegate_task":
                 from tools.delegate_tool import delegate_task as _delegate_task
@@ -6594,7 +6854,7 @@ class AIAgent:
                     goal_preview = (function_args.get("goal") or "")[:30]
                     spinner_label = f"🔀 {goal_preview}" if goal_preview else "🔀 delegating"
                 spinner = None
-                if self.quiet_mode and not self.tool_progress_callback and self._should_start_quiet_spinner():
+                if self._should_emit_quiet_tool_messages() and self._should_start_quiet_spinner():
                     face = random.choice(KawaiiSpinner.KAWAII_WAITING)
                     spinner = KawaiiSpinner(f"{face} {spinner_label}", spinner_type='dots', print_fn=self._print_fn)
                     spinner.start()
@@ -6616,13 +6876,13 @@ class AIAgent:
                     cute_msg = _get_cute_tool_message_impl('delegate_task', function_args, tool_duration, result=_delegate_result)
                     if spinner:
                         spinner.stop(cute_msg)
-                    elif self.quiet_mode:
+                    elif self._should_emit_quiet_tool_messages():
                         self._vprint(f"  {cute_msg}")
             elif self._memory_manager and self._memory_manager.has_tool(function_name):
                 # Memory provider tools (hindsight_retain, honcho_search, etc.)
                 # These are not in the tool registry — route through MemoryManager.
                 spinner = None
-                if self.quiet_mode and not self.tool_progress_callback:
+                if self._should_emit_quiet_tool_messages() and self._should_start_quiet_spinner():
                     face = random.choice(KawaiiSpinner.KAWAII_WAITING)
                     emoji = _get_tool_emoji(function_name)
                     preview = _build_tool_preview(function_name, function_args) or function_name
@@ -6640,11 +6900,11 @@ class AIAgent:
                     cute_msg = _get_cute_tool_message_impl(function_name, function_args, tool_duration, result=_mem_result)
                     if spinner:
                         spinner.stop(cute_msg)
-                    elif self.quiet_mode:
+                    elif self._should_emit_quiet_tool_messages():
                         self._vprint(f"  {cute_msg}")
             elif self.quiet_mode:
                 spinner = None
-                if not self.tool_progress_callback:
+                if self._should_emit_quiet_tool_messages() and self._should_start_quiet_spinner():
                     face = random.choice(KawaiiSpinner.KAWAII_WAITING)
                     emoji = _get_tool_emoji(function_name)
                     preview = _build_tool_preview(function_name, function_args) or function_name
@@ -6667,7 +6927,7 @@ class AIAgent:
                     cute_msg = _get_cute_tool_message_impl(function_name, function_args, tool_duration, result=_spinner_result)
                     if spinner:
                         spinner.stop(cute_msg)
-                    else:
+                    elif self._should_emit_quiet_tool_messages():
                         self._vprint(f"  {cute_msg}")
             else:
                 try:
@@ -7061,7 +7321,7 @@ class AIAgent:
         self._thinking_prefill_retries = 0
         self._last_content_with_tools = None
         self._mute_post_response = False
-        self._surrogate_sanitized = False
+        self._unicode_sanitization_passes = 0
 
         # Pre-turn connection health check: detect and clean up dead TCP
         # connections left over from provider outages or dropped streams.
@@ -7291,6 +7551,7 @@ class AIAgent:
         interrupted = False
         codex_ack_continuations = 0
         length_continue_retries = 0
+        truncated_tool_call_retries = 0
         truncated_response_prefix = ""
         compression_attempts = 0
         _turn_exit_reason = "unknown"  # Diagnostic: why the loop ended
@@ -7500,6 +7761,7 @@ class AIAgent:
 
             finish_reason = "stop"
             response = None  # Guard against UnboundLocalError if all retries fail
+            api_kwargs = None  # Guard against UnboundLocalError in except handler
 
             while retry_count < max_retries:
                 try:
@@ -7759,9 +8021,11 @@ class AIAgent:
                         # retries are pointless.  Detect this early and give a
                         # targeted error instead of wasting 3 API calls.
                         _trunc_content = None
+                        _trunc_has_tool_calls = False
                         if self.api_mode == "chat_completions":
                             _trunc_msg = response.choices[0].message if (hasattr(response, "choices") and response.choices) else None
                             _trunc_content = getattr(_trunc_msg, "content", None) if _trunc_msg else None
+                            _trunc_has_tool_calls = bool(getattr(_trunc_msg, "tool_calls", None)) if _trunc_msg else False
                         elif self.api_mode == "anthropic_messages":
                             # Anthropic response.content is a list of blocks
                             _text_parts = []
@@ -7771,9 +8035,11 @@ class AIAgent:
                             _trunc_content = "\n".join(_text_parts) if _text_parts else None
 
                         _thinking_exhausted = (
-                            _trunc_content is not None
-                            and not self._has_content_after_think_block(_trunc_content)
-                        ) or _trunc_content is None
+                            not _trunc_has_tool_calls and (
+                                (_trunc_content is not None and not self._has_content_after_think_block(_trunc_content))
+                                or _trunc_content is None
+                            )
+                        )
 
                         if _thinking_exhausted:
                             _exhaust_error = (
@@ -7849,6 +8115,34 @@ class AIAgent:
                                     "error": "Response remained truncated after 3 continuation attempts",
                                 }
 
+                        if self.api_mode == "chat_completions":
+                            assistant_message = response.choices[0].message
+                            if assistant_message.tool_calls:
+                                if truncated_tool_call_retries < 1:
+                                    truncated_tool_call_retries += 1
+                                    self._vprint(
+                                        f"{self.log_prefix}⚠️  Truncated tool call detected — retrying API call...",
+                                        force=True,
+                                    )
+                                    # Don't append the broken response to messages;
+                                    # just re-run the same API call from the current
+                                    # message state, giving the model another chance.
+                                    continue
+                                self._vprint(
+                                    f"{self.log_prefix}⚠️  Truncated tool call response detected again — refusing to execute incomplete tool arguments.",
+                                    force=True,
+                                )
+                                self._cleanup_task_resources(effective_task_id)
+                                self._persist_session(messages, conversation_history)
+                                return {
+                                    "final_response": None,
+                                    "messages": messages,
+                                    "api_calls": api_call_count,
+                                    "completed": False,
+                                    "partial": True,
+                                    "error": "Response truncated due to output length limit",
+                                }
+
                         # If we have prior messages, roll back to last complete state
                         if len(messages) > 1:
                             self._vprint(f"{self.log_prefix}   ⏪ Rolling back to last complete assistant turn")
@@ -8013,22 +8307,40 @@ class AIAgent:
                         self.thinking_callback("")
 
                     # -----------------------------------------------------------
-                    # Surrogate character recovery.  UnicodeEncodeError happens
-                    # when the messages contain lone surrogates (U+D800..U+DFFF)
-                    # that are invalid UTF-8.  Common source: clipboard paste
-                    # from Google Docs or similar rich-text editors.  We sanitize
-                    # the entire messages list in-place and retry once.
+                    # UnicodeEncodeError recovery.  Two common causes:
+                    #   1. Lone surrogates (U+D800..U+DFFF) from clipboard paste
+                    #      (Google Docs, rich-text editors) — sanitize and retry.
+                    #   2. ASCII codec on systems with LANG=C or non-UTF-8 locale
+                    #      (e.g. Chromebooks) — any non-ASCII character fails.
+                    #      Detect via the error message mentioning 'ascii' codec.
+                    # We sanitize messages in-place and may retry twice:
+                    # first to strip surrogates, then once more for pure
+                    # ASCII-only locale sanitization if needed.
                     # -----------------------------------------------------------
-                    if isinstance(api_error, UnicodeEncodeError) and not getattr(self, '_surrogate_sanitized', False):
-                        self._surrogate_sanitized = True
-                        if _sanitize_messages_surrogates(messages):
+                    if isinstance(api_error, UnicodeEncodeError) and getattr(self, '_unicode_sanitization_passes', 0) < 2:
+                        _err_str = str(api_error).lower()
+                        _is_ascii_codec = "'ascii'" in _err_str or "ascii" in _err_str
+                        _surrogates_found = _sanitize_messages_surrogates(messages)
+                        if _surrogates_found:
+                            self._unicode_sanitization_passes += 1
                             self._vprint(
                                 f"{self.log_prefix}⚠️  Stripped invalid surrogate characters from messages. Retrying...",
                                 force=True,
                             )
                             continue
-                        # Surrogates weren't in messages — might be in system
-                        # prompt or prefill.  Fall through to normal error path.
+                        if _is_ascii_codec:
+                            # ASCII codec: the system encoding can't handle
+                            # non-ASCII characters at all. Sanitize all
+                            # non-ASCII content from messages and retry.
+                            if _sanitize_messages_non_ascii(messages):
+                                self._unicode_sanitization_passes += 1
+                                self._vprint(
+                                    f"{self.log_prefix}⚠️  System encoding is ASCII — stripped non-ASCII characters from messages. Retrying...",
+                                    force=True,
+                                )
+                                continue
+                        # Nothing to sanitize in messages — might be in system
+                        # prompt or prefill. Fall through to normal error path.
 
                     status_code = getattr(api_error, "status_code", None)
                     error_context = self._extract_api_error_context(api_error)
@@ -8054,6 +8366,7 @@ class AIAgent:
                     recovered_with_pool, has_retried_429 = self._recover_with_credential_pool(
                         status_code=status_code,
                         has_retried_429=has_retried_429,
+                        classified_reason=classified.reason,
                         error_context=error_context,
                     )
                     if recovered_with_pool:
@@ -8161,7 +8474,33 @@ class AIAgent:
                         if _err_body_str:
                             self._vprint(f"{self.log_prefix}   📋 Details: {_err_body_str}", force=True)
                     self._vprint(f"{self.log_prefix}   ⏱️  Elapsed: {elapsed_time:.2f}s  Context: {len(api_messages)} msgs, ~{approx_tokens:,} tokens")
-                    
+
+                    # Actionable hint for OpenRouter "no tool endpoints" error.
+                    # This fires regardless of whether fallback succeeds — the
+                    # user needs to know WHY their model failed so they can fix
+                    # their provider routing, not just silently fall back.
+                    if (
+                        self._is_openrouter_url()
+                        and "support tool use" in error_msg
+                    ):
+                        self._vprint(
+                            f"{self.log_prefix}   💡 No OpenRouter providers for {_model} support tool calling with your current settings.",
+                            force=True,
+                        )
+                        if self.providers_allowed:
+                            self._vprint(
+                                f"{self.log_prefix}      Your provider_routing.only restriction is filtering out tool-capable providers.",
+                                force=True,
+                            )
+                            self._vprint(
+                                f"{self.log_prefix}      Try removing the restriction or adding providers that support tools for this model.",
+                                force=True,
+                            )
+                        self._vprint(
+                            f"{self.log_prefix}      Check which providers support tools: https://openrouter.ai/models/{_model}",
+                            force=True,
+                        )
+
                     # Check for interrupt before deciding to retry
                     if self._interrupt_requested:
                         self._vprint(f"{self.log_prefix}⚡ Interrupt detected during error handling, aborting retries.", force=True)
@@ -8217,6 +8556,10 @@ class AIAgent:
                                 approx_tokens=approx_tokens,
                                 task_id=effective_task_id,
                             )
+                            # Compression created a new session — clear history
+                            # so _flush_messages_to_session_db writes compressed
+                            # messages to the new session, not skipping them.
+                            conversation_history = None
                             if len(messages) < original_len or old_ctx > _reduced_ctx:
                                 self._emit_status(
                                     f"🗜️ Context reduced to {_reduced_ctx:,} tokens "
@@ -8274,6 +8617,10 @@ class AIAgent:
                             messages, system_message, approx_tokens=approx_tokens,
                             task_id=effective_task_id,
                         )
+                        # Compression created a new session — clear history
+                        # so _flush_messages_to_session_db writes compressed
+                        # messages to the new session, not skipping them.
+                        conversation_history = None
 
                         if len(messages) < original_len:
                             self._emit_status(f"🗜️ Compressed {original_len} → {len(messages)} messages, retrying...")
@@ -8392,6 +8739,10 @@ class AIAgent:
                             messages, system_message, approx_tokens=approx_tokens,
                             task_id=effective_task_id,
                         )
+                        # Compression created a new session — clear history
+                        # so _flush_messages_to_session_db writes compressed
+                        # messages to the new session, not skipping them.
+                        conversation_history = None
 
                         if len(messages) < original_len or new_ctx and new_ctx < old_ctx:
                             if len(messages) < original_len:
@@ -8445,9 +8796,10 @@ class AIAgent:
                         if self._try_activate_fallback():
                             retry_count = 0
                             continue
-                        self._dump_api_request_debug(
-                            api_kwargs, reason="non_retryable_client_error", error=api_error,
-                        )
+                        if api_kwargs is not None:
+                            self._dump_api_request_debug(
+                                api_kwargs, reason="non_retryable_client_error", error=api_error,
+                            )
                         self._emit_status(
                             f"❌ Non-retryable error (HTTP {status_code}): "
                             f"{self._summarize_api_error(api_error)}"
@@ -8550,9 +8902,10 @@ class AIAgent:
                             self.log_prefix, max_retries, _final_summary,
                             _provider, _model, len(api_messages), f"{approx_tokens:,}",
                         )
-                        self._dump_api_request_debug(
-                            api_kwargs, reason="max_retries_exhausted", error=api_error,
-                        )
+                        if api_kwargs is not None:
+                            self._dump_api_request_debug(
+                                api_kwargs, reason="max_retries_exhausted", error=api_error,
+                            )
                         self._persist_session(messages, conversation_history)
                         _final_response = f"API call failed after {max_retries} retries: {_final_summary}"
                         if _is_stream_drop:
@@ -8999,6 +9352,11 @@ class AIAgent:
 
                     self._execute_tool_calls(assistant_message, messages, effective_task_id, api_call_count)
 
+                    # Reset per-turn retry counters after successful tool
+                    # execution so a single truncation doesn't poison the
+                    # entire conversation.
+                    truncated_tool_call_retries = 0
+
                     # Signal that a paragraph break is needed before the next
                     # streamed text.  We don't emit it immediately because
                     # multiple consecutive tool iterations would stack up

scripts/install.sh

@@ -2,8 +2,8 @@
 # ============================================================================
 # Hermes Agent Installer
 # ============================================================================
-# Installation script for Linux and macOS.
-# Uses uv for fast Python provisioning and package management.
+# Installation script for Linux, macOS, and Android/Termux.
+# Uses uv for desktop/server installs and Python's stdlib venv + pip on Termux.
 #
 # Usage:
 #   curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
@@ -117,6 +117,36 @@ log_error() {
     echo -e "${RED}✗${NC} $1"
 }
 
+is_termux() {
+    [ -n "${TERMUX_VERSION:-}" ] || [[ "${PREFIX:-}" == *"com.termux/files/usr"* ]]
+}
+
+get_command_link_dir() {
+    if is_termux && [ -n "${PREFIX:-}" ]; then
+        echo "$PREFIX/bin"
+    else
+        echo "$HOME/.local/bin"
+    fi
+}
+
+get_command_link_display_dir() {
+    if is_termux && [ -n "${PREFIX:-}" ]; then
+        echo '$PREFIX/bin'
+    else
+        echo '~/.local/bin'
+    fi
+}
+
+get_hermes_command_path() {
+    local link_dir
+    link_dir="$(get_command_link_dir)"
+    if [ -x "$link_dir/hermes" ]; then
+        echo "$link_dir/hermes"
+    else
+        echo "hermes"
+    fi
+}
+
 # ============================================================================
 # System detection
 # ============================================================================
@@ -124,12 +154,17 @@ log_error() {
 detect_os() {
     case "$(uname -s)" in
         Linux*)
-            OS="linux"
-            if [ -f /etc/os-release ]; then
-                . /etc/os-release
-                DISTRO="$ID"
+            if is_termux; then
+                OS="android"
+                DISTRO="termux"
             else
-                DISTRO="unknown"
+                OS="linux"
+                if [ -f /etc/os-release ]; then
+                    . /etc/os-release
+                    DISTRO="$ID"
+                else
+                    DISTRO="unknown"
+                fi
             fi
             ;;
         Darwin*)
@@ -158,6 +193,12 @@ detect_os() {
 # ============================================================================
 
 install_uv() {
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Termux detected — using Python's stdlib venv + pip instead of uv"
+        UV_CMD=""
+        return 0
+    fi
+
     log_info "Checking for uv package manager..."
 
     # Check common locations for uv
@@ -209,6 +250,25 @@ install_uv() {
 }
 
 check_python() {
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Checking Termux Python..."
+        if command -v python >/dev/null 2>&1; then
+            PYTHON_PATH="$(command -v python)"
+            if "$PYTHON_PATH" -c 'import sys; raise SystemExit(0 if sys.version_info >= (3, 11) else 1)' 2>/dev/null; then
+                PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
+                log_success "Python found: $PYTHON_FOUND_VERSION"
+                return 0
+            fi
+        fi
+
+        log_info "Installing Python via pkg..."
+        pkg install -y python >/dev/null
+        PYTHON_PATH="$(command -v python)"
+        PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
+        log_success "Python installed: $PYTHON_FOUND_VERSION"
+        return 0
+    fi
+
     log_info "Checking Python $PYTHON_VERSION..."
 
     # Let uv handle Python — it can download and manage Python versions
@@ -243,6 +303,17 @@ check_git() {
     fi
 
     log_error "Git not found"
+
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Installing Git via pkg..."
+        pkg install -y git >/dev/null
+        if command -v git >/dev/null 2>&1; then
+            GIT_VERSION=$(git --version | awk '{print $3}')
+            log_success "Git $GIT_VERSION installed"
+            return 0
+        fi
+    fi
+
     log_info "Please install Git:"
 
     case "$OS" in
@@ -262,6 +333,9 @@ check_git() {
                     ;;
             esac
             ;;
+        android)
+            log_info "  pkg install git"
+            ;;
         macos)
             log_info "  xcode-select --install"
             log_info "  Or: brew install git"
@@ -290,11 +364,29 @@ check_node() {
         return 0
     fi
 
-    log_info "Node.js not found — installing Node.js $NODE_VERSION LTS..."
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Node.js not found — installing Node.js via pkg..."
+    else
+        log_info "Node.js not found — installing Node.js $NODE_VERSION LTS..."
+    fi
     install_node
 }
 
 install_node() {
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Installing Node.js via pkg..."
+        if pkg install -y nodejs >/dev/null; then
+            local installed_ver
+            installed_ver=$(node --version 2>/dev/null)
+            log_success "Node.js $installed_ver installed via pkg"
+            HAS_NODE=true
+        else
+            log_warn "Failed to install Node.js via pkg"
+            HAS_NODE=false
+        fi
+        return 0
+    fi
+
     local arch=$(uname -m)
     local node_arch
     case "$arch" in
@@ -413,6 +505,30 @@ install_system_packages() {
         need_ffmpeg=true
     fi
 
+    # Termux always needs the Android build toolchain for the tested pip path,
+    # even when ripgrep/ffmpeg are already present.
+    if [ "$DISTRO" = "termux" ]; then
+        local termux_pkgs=(clang rust make pkg-config libffi openssl)
+        if [ "$need_ripgrep" = true ]; then
+            termux_pkgs+=("ripgrep")
+        fi
+        if [ "$need_ffmpeg" = true ]; then
+            termux_pkgs+=("ffmpeg")
+        fi
+
+        log_info "Installing Termux packages: ${termux_pkgs[*]}"
+        if pkg install -y "${termux_pkgs[@]}" >/dev/null; then
+            [ "$need_ripgrep" = true ] && HAS_RIPGREP=true && log_success "ripgrep installed"
+            [ "$need_ffmpeg" = true ]  && HAS_FFMPEG=true  && log_success "ffmpeg installed"
+            log_success "Termux build dependencies installed"
+            return 0
+        fi
+
+        log_warn "Could not auto-install all Termux packages"
+        log_info "Install manually: pkg install ${termux_pkgs[*]}"
+        return 0
+    fi
+
     # Nothing to install — done
     if [ "$need_ripgrep" = false ] && [ "$need_ffmpeg" = false ]; then
         return 0
@@ -550,6 +666,9 @@ show_manual_install_hint() {
                 *)             log_info "  Use your package manager or visit the project homepage" ;;
             esac
             ;;
+        android)
+            log_info "  pkg install $pkg"
+            ;;
         macos) log_info "  brew install $pkg" ;;
     esac
 }
@@ -646,6 +765,19 @@ setup_venv() {
         return 0
     fi
 
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Creating virtual environment with Termux Python..."
+
+        if [ -d "venv" ]; then
+            log_info "Virtual environment already exists, recreating..."
+            rm -rf venv
+        fi
+
+        "$PYTHON_PATH" -m venv venv
+        log_success "Virtual environment ready ($(./venv/bin/python --version 2>/dev/null))"
+        return 0
+    fi
+
     log_info "Creating virtual environment with Python $PYTHON_VERSION..."
 
     if [ -d "venv" ]; then
@@ -662,6 +794,46 @@ setup_venv() {
 install_deps() {
     log_info "Installing dependencies..."
 
+    if [ "$DISTRO" = "termux" ]; then
+        if [ "$USE_VENV" = true ]; then
+            export VIRTUAL_ENV="$INSTALL_DIR/venv"
+            PIP_PYTHON="$INSTALL_DIR/venv/bin/python"
+        else
+            PIP_PYTHON="$PYTHON_PATH"
+        fi
+
+        if [ -z "${ANDROID_API_LEVEL:-}" ]; then
+            ANDROID_API_LEVEL="$(getprop ro.build.version.sdk 2>/dev/null || true)"
+            if [ -z "$ANDROID_API_LEVEL" ]; then
+                ANDROID_API_LEVEL=24
+            fi
+            export ANDROID_API_LEVEL
+            log_info "Using ANDROID_API_LEVEL=$ANDROID_API_LEVEL for Android wheel builds"
+        fi
+
+        "$PIP_PYTHON" -m pip install --upgrade pip setuptools wheel >/dev/null
+        if ! "$PIP_PYTHON" -m pip install -e '.[termux]' -c constraints-termux.txt; then
+            log_warn "Termux feature install (.[termux]) failed, trying base install..."
+            if ! "$PIP_PYTHON" -m pip install -e '.' -c constraints-termux.txt; then
+                log_error "Package installation failed on Termux."
+                log_info "Ensure these packages are installed: pkg install clang rust make pkg-config libffi openssl"
+                log_info "Then re-run: cd $INSTALL_DIR && python -m pip install -e '.[termux]' -c constraints-termux.txt"
+                exit 1
+            fi
+        fi
+
+        log_success "Main package installed"
+        log_info "Termux note: browser/WhatsApp tooling is not installed by default; see the Termux guide for optional follow-up steps."
+
+        if [ -d "tinker-atropos" ] && [ -f "tinker-atropos/pyproject.toml" ]; then
+            log_info "tinker-atropos submodule found — skipping install (optional, for RL training)"
+            log_info "  To install later: $PIP_PYTHON -m pip install -e \"./tinker-atropos\""
+        fi
+
+        log_success "All dependencies installed"
+        return 0
+    fi
+
     if [ "$USE_VENV" = true ]; then
         # Tell uv to install into our venv (no need to activate)
         export VIRTUAL_ENV="$INSTALL_DIR/venv"
@@ -743,19 +915,35 @@ setup_path() {
     if [ ! -x "$HERMES_BIN" ]; then
         log_warn "hermes entry point not found at $HERMES_BIN"
         log_info "This usually means the pip install didn't complete successfully."
-        log_info "Try: cd $INSTALL_DIR && uv pip install -e '.[all]'"
+        if [ "$DISTRO" = "termux" ]; then
+            log_info "Try: cd $INSTALL_DIR && python -m pip install -e '.[termux]' -c constraints-termux.txt"
+        else
+            log_info "Try: cd $INSTALL_DIR && uv pip install -e '.[all]'"
+        fi
         return 0
     fi
 
-    # Create symlink in ~/.local/bin (standard user binary location, usually on PATH)
-    mkdir -p "$HOME/.local/bin"
-    ln -sf "$HERMES_BIN" "$HOME/.local/bin/hermes"
-    log_success "Symlinked hermes → ~/.local/bin/hermes"
+    local command_link_dir
+    local command_link_display_dir
+    command_link_dir="$(get_command_link_dir)"
+    command_link_display_dir="$(get_command_link_display_dir)"
+
+    # Create a user-facing shim for the hermes command.
+    mkdir -p "$command_link_dir"
+    ln -sf "$HERMES_BIN" "$command_link_dir/hermes"
+    log_success "Symlinked hermes → $command_link_display_dir/hermes"
+
+    if [ "$DISTRO" = "termux" ]; then
+        export PATH="$command_link_dir:$PATH"
+        log_info "$command_link_display_dir is the native Termux command path"
+        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).
-    if ! echo "$PATH" | tr ':' '\n' | grep -q "^$HOME/.local/bin$"; then
+    if ! echo "$PATH" | tr ':' '\n' | grep -q "^$command_link_dir$"; then
         SHELL_CONFIGS=()
         LOGIN_SHELL="$(basename "${SHELL:-/bin/bash}")"
         case "$LOGIN_SHELL" in
@@ -801,7 +989,7 @@ setup_path() {
     fi
 
     # Export for current session so hermes works immediately
-    export PATH="$HOME/.local/bin:$PATH"
+    export PATH="$command_link_dir:$PATH"
 
     log_success "hermes command ready"
 }
@@ -878,6 +1066,13 @@ install_node_deps() {
         return 0
     fi
 
+    if [ "$DISTRO" = "termux" ]; then
+        log_info "Skipping automatic Node/browser dependency setup on Termux"
+        log_info "Browser automation and WhatsApp bridge are not part of the tested Termux install path yet."
+        log_info "If you want to experiment manually later, run: cd $INSTALL_DIR && npm install"
+        return 0
+    fi
+
     if [ -f "$INSTALL_DIR/package.json" ]; then
         log_info "Installing Node.js dependencies (browser tools)..."
         cd "$INSTALL_DIR"
@@ -887,10 +1082,19 @@ install_node_deps() {
         log_success "Node.js dependencies installed"
 
         # Install Playwright browser + system dependencies.
-        # Playwright's install-deps only supports apt/dnf/zypper natively.
+        # Playwright's --with-deps only supports apt-based systems natively.
         # For Arch/Manjaro we install the system libs via pacman first.
+        # Other systems must install Chromium dependencies manually.
         log_info "Installing browser engine (Playwright Chromium)..."
         case "$DISTRO" in
+            ubuntu|debian|raspbian|pop|linuxmint|elementary|zorin|kali|parrot)
+                log_info "Playwright may request sudo to install browser system dependencies (shared libraries)."
+                log_info "This is standard Playwright setup — Hermes itself does not require root access."
+                cd "$INSTALL_DIR" && npx playwright install --with-deps chromium 2>/dev/null || {
+                    log_warn "Playwright browser installation failed — browser tools will not work."
+                    log_warn "Try running manually: cd $INSTALL_DIR && npx playwright install --with-deps chromium"
+                }
+                ;;
             arch|manjaro)
                 if command -v pacman &> /dev/null; then
                     log_info "Arch/Manjaro detected — installing Chromium system dependencies via pacman..."
@@ -905,15 +1109,35 @@ install_node_deps() {
                         log_warn "  sudo pacman -S nss atk at-spi2-core cups libdrm libxkbcommon mesa pango cairo alsa-lib"
                     fi
                 fi
-                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || true
+                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || {
+                    log_warn "Playwright browser installation failed — browser tools will not work."
+                }
+                ;;
+            fedora|rhel|centos|rocky|alma)
+                log_warn "Playwright does not support automatic dependency installation on RPM-based systems."
+                log_info "Install Chromium system dependencies manually before using browser tools:"
+                log_info "  sudo dnf install nss atk at-spi2-core cups-libs libdrm libxkbcommon mesa-libgbm pango cairo alsa-lib"
+                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || {
+                    log_warn "Playwright browser installation failed — install dependencies above and retry."
+                }
+                ;;
+            opensuse*|sles)
+                log_warn "Playwright does not support automatic dependency installation on zypper-based systems."
+                log_info "Install Chromium system dependencies manually before using browser tools:"
+                log_info "  sudo zypper install mozilla-nss libatk-1_0-0 at-spi2-core cups-libs libdrm2 libxkbcommon0 Mesa-libgbm1 pango cairo libasound2"
+                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || {
+                    log_warn "Playwright browser installation failed — install dependencies above and retry."
+                }
                 ;;
             *)
-                log_info "Playwright may request sudo to install browser system dependencies (shared libraries)."
-                log_info "This is standard Playwright setup — Hermes itself does not require root access."
-                cd "$INSTALL_DIR" && npx playwright install --with-deps chromium 2>/dev/null || true
+                log_warn "Playwright does not support automatic dependency installation on $DISTRO."
+                log_info "Install Chromium/browser system dependencies for your distribution, then run:"
+                log_info "  cd $INSTALL_DIR && npx playwright install chromium"
+                log_info "Browser tools will not work until dependencies are installed."
+                cd "$INSTALL_DIR" && npx playwright install chromium 2>/dev/null || true
                 ;;
         esac
-        log_success "Browser engine installed"
+        log_success "Browser engine setup complete"
     fi
 
     # Install WhatsApp bridge dependencies
@@ -992,8 +1216,7 @@ maybe_start_gateway() {
             read -p "Pair WhatsApp now? [Y/n] " -n 1 -r
             echo
             if [[ $REPLY =~ ^[Yy]$ ]] || [[ -z $REPLY ]]; then
-                HERMES_CMD="$HOME/.local/bin/hermes"
-                [ ! -x "$HERMES_CMD" ] && HERMES_CMD="hermes"
+                HERMES_CMD="$(get_hermes_command_path)"
                 $HERMES_CMD whatsapp || true
             fi
         else
@@ -1007,16 +1230,17 @@ maybe_start_gateway() {
     fi
 
     echo ""
-    read -p "Would you like to install the gateway as a background service? [Y/n] " -n 1 -r < /dev/tty
+    if [ "$DISTRO" = "termux" ]; then
+        read -p "Would you like to start the gateway in the background? [Y/n] " -n 1 -r < /dev/tty
+    else
+        read -p "Would you like to install the gateway as a background service? [Y/n] " -n 1 -r < /dev/tty
+    fi
     echo
 
     if [[ $REPLY =~ ^[Yy]$ ]] || [[ -z $REPLY ]]; then
-        HERMES_CMD="$HOME/.local/bin/hermes"
-        if [ ! -x "$HERMES_CMD" ]; then
-            HERMES_CMD="hermes"
-        fi
+        HERMES_CMD="$(get_hermes_command_path)"
 
-        if command -v systemctl &> /dev/null; then
+        if [ "$DISTRO" != "termux" ] && command -v systemctl &> /dev/null; then
             log_info "Installing systemd service..."
             if $HERMES_CMD gateway install 2>/dev/null; then
                 log_success "Gateway service installed"
@@ -1029,12 +1253,19 @@ maybe_start_gateway() {
                 log_warn "Systemd install failed. You can start manually: hermes gateway"
             fi
         else
-            log_info "systemd not available — starting gateway in background..."
+            if [ "$DISTRO" = "termux" ]; then
+                log_info "Termux detected — starting gateway in best-effort background mode..."
+            else
+                log_info "systemd not available — starting gateway in background..."
+            fi
             nohup $HERMES_CMD gateway > "$HERMES_HOME/logs/gateway.log" 2>&1 &
             GATEWAY_PID=$!
             log_success "Gateway started (PID $GATEWAY_PID). Logs: ~/.hermes/logs/gateway.log"
             log_info "To stop: kill $GATEWAY_PID"
             log_info "To restart later: hermes gateway"
+            if [ "$DISTRO" = "termux" ]; then
+                log_warn "Android may stop background processes when Termux is suspended or the system reclaims resources."
+            fi
         fi
     else
         log_info "Skipped. Start the gateway later with: hermes gateway"
@@ -1073,24 +1304,33 @@ print_success() {
 
     echo -e "${CYAN}─────────────────────────────────────────────────────────${NC}"
     echo ""
-    echo -e "${YELLOW}⚡ Reload your shell to use 'hermes' command:${NC}"
-    echo ""
-    LOGIN_SHELL="$(basename "${SHELL:-/bin/bash}")"
-    if [ "$LOGIN_SHELL" = "zsh" ]; then
-        echo "   source ~/.zshrc"
-    elif [ "$LOGIN_SHELL" = "bash" ]; then
-        echo "   source ~/.bashrc"
+    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 ""
     else
-        echo "   source ~/.bashrc   # or ~/.zshrc"
+        echo -e "${YELLOW}⚡ Reload your shell to use 'hermes' command:${NC}"
+        echo ""
+        LOGIN_SHELL="$(basename "${SHELL:-/bin/bash}")"
+        if [ "$LOGIN_SHELL" = "zsh" ]; then
+            echo "   source ~/.zshrc"
+        elif [ "$LOGIN_SHELL" = "bash" ]; then
+            echo "   source ~/.bashrc"
+        else
+            echo "   source ~/.bashrc   # or ~/.zshrc"
+        fi
+        echo ""
     fi
-    echo ""
 
     # Show Node.js warning if auto-install failed
     if [ "$HAS_NODE" = false ]; then
         echo -e "${YELLOW}"
         echo "Note: Node.js could not be installed automatically."
         echo "Browser tools need Node.js. Install manually:"
-        echo "  https://nodejs.org/en/download/"
+        if [ "$DISTRO" = "termux" ]; then
+            echo "  pkg install nodejs"
+        else
+            echo "  https://nodejs.org/en/download/"
+        fi
         echo -e "${NC}"
     fi
 
@@ -1099,7 +1339,11 @@ print_success() {
         echo -e "${YELLOW}"
         echo "Note: ripgrep (rg) was not found. File search will use"
         echo "grep as a fallback. For faster search in large codebases,"
-        echo "install ripgrep: sudo apt install ripgrep (or brew install ripgrep)"
+        if [ "$DISTRO" = "termux" ]; then
+            echo "install ripgrep: pkg install ripgrep"
+        else
+            echo "install ripgrep: sudo apt install ripgrep (or brew install ripgrep)"
+        fi
         echo -e "${NC}"
     fi
 }

setup-hermes.sh

@@ -3,17 +3,17 @@
 # Hermes Agent Setup Script
 # ============================================================================
 # Quick setup for developers who cloned the repo manually.
-# Uses uv for fast Python provisioning and package management.
+# Uses uv for desktop/server setup and Python's stdlib venv + pip on Termux.
 #
 # Usage:
 #   ./setup-hermes.sh
 #
 # This script:
-# 1. Installs uv if not present
-# 2. Creates a virtual environment with Python 3.11 via uv
-# 3. Installs all dependencies (main package + submodules)
+# 1. Detects desktop/server vs Android/Termux setup path
+# 2. Creates a Python 3.11 virtual environment
+# 3. Installs the appropriate dependency set for the platform
 # 4. Creates .env from template (if not exists)
-# 5. Symlinks the 'hermes' CLI command into ~/.local/bin
+# 5. Symlinks the 'hermes' CLI command into a user-facing bin dir
 # 6. Runs the setup wizard (optional)
 # ============================================================================
 
@@ -31,6 +31,26 @@ cd "$SCRIPT_DIR"
 
 PYTHON_VERSION="3.11"
 
+is_termux() {
+    [ -n "${TERMUX_VERSION:-}" ] || [[ "${PREFIX:-}" == *"com.termux/files/usr"* ]]
+}
+
+get_command_link_dir() {
+    if is_termux && [ -n "${PREFIX:-}" ]; then
+        echo "$PREFIX/bin"
+    else
+        echo "$HOME/.local/bin"
+    fi
+}
+
+get_command_link_display_dir() {
+    if is_termux && [ -n "${PREFIX:-}" ]; then
+        echo '$PREFIX/bin'
+    else
+        echo '~/.local/bin'
+    fi
+}
+
 echo ""
 echo -e "${CYAN}⚕ Hermes Agent Setup${NC}"
 echo ""
@@ -42,36 +62,40 @@ echo ""
 echo -e "${CYAN}→${NC} Checking for uv..."
 
 UV_CMD=""
-if command -v uv &> /dev/null; then
-    UV_CMD="uv"
-elif [ -x "$HOME/.local/bin/uv" ]; then
-    UV_CMD="$HOME/.local/bin/uv"
-elif [ -x "$HOME/.cargo/bin/uv" ]; then
-    UV_CMD="$HOME/.cargo/bin/uv"
-fi
-
-if [ -n "$UV_CMD" ]; then
-    UV_VERSION=$($UV_CMD --version 2>/dev/null)
-    echo -e "${GREEN}✓${NC} uv found ($UV_VERSION)"
+if is_termux; then
+    echo -e "${CYAN}→${NC} Termux detected — using Python's stdlib venv + pip instead of uv"
 else
-    echo -e "${CYAN}→${NC} Installing uv..."
-    if curl -LsSf https://astral.sh/uv/install.sh | sh 2>/dev/null; then
-        if [ -x "$HOME/.local/bin/uv" ]; then
-            UV_CMD="$HOME/.local/bin/uv"
-        elif [ -x "$HOME/.cargo/bin/uv" ]; then
-            UV_CMD="$HOME/.cargo/bin/uv"
-        fi
-        
-        if [ -n "$UV_CMD" ]; then
-            UV_VERSION=$($UV_CMD --version 2>/dev/null)
-            echo -e "${GREEN}✓${NC} uv installed ($UV_VERSION)"
+    if command -v uv &> /dev/null; then
+        UV_CMD="uv"
+    elif [ -x "$HOME/.local/bin/uv" ]; then
+        UV_CMD="$HOME/.local/bin/uv"
+    elif [ -x "$HOME/.cargo/bin/uv" ]; then
+        UV_CMD="$HOME/.cargo/bin/uv"
+    fi
+
+    if [ -n "$UV_CMD" ]; then
+        UV_VERSION=$($UV_CMD --version 2>/dev/null)
+        echo -e "${GREEN}✓${NC} uv found ($UV_VERSION)"
+    else
+        echo -e "${CYAN}→${NC} Installing uv..."
+        if curl -LsSf https://astral.sh/uv/install.sh | sh 2>/dev/null; then
+            if [ -x "$HOME/.local/bin/uv" ]; then
+                UV_CMD="$HOME/.local/bin/uv"
+            elif [ -x "$HOME/.cargo/bin/uv" ]; then
+                UV_CMD="$HOME/.cargo/bin/uv"
+            fi
+
+            if [ -n "$UV_CMD" ]; then
+                UV_VERSION=$($UV_CMD --version 2>/dev/null)
+                echo -e "${GREEN}✓${NC} uv installed ($UV_VERSION)"
+            else
+                echo -e "${RED}✗${NC} uv installed but not found. Add ~/.local/bin to PATH and retry."
+                exit 1
+            fi
         else
-            echo -e "${RED}✗${NC} uv installed but not found. Add ~/.local/bin to PATH and retry."
+            echo -e "${RED}✗${NC} Failed to install uv. Visit https://docs.astral.sh/uv/"
             exit 1
         fi
-    else
-        echo -e "${RED}✗${NC} Failed to install uv. Visit https://docs.astral.sh/uv/"
-        exit 1
     fi
 fi
 
@@ -81,16 +105,34 @@ fi
 
 echo -e "${CYAN}→${NC} Checking Python $PYTHON_VERSION..."
 
-if $UV_CMD python find "$PYTHON_VERSION" &> /dev/null; then
-    PYTHON_PATH=$($UV_CMD python find "$PYTHON_VERSION")
-    PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
-    echo -e "${GREEN}✓${NC} $PYTHON_FOUND_VERSION found"
+if is_termux; then
+    if command -v python >/dev/null 2>&1; then
+        PYTHON_PATH="$(command -v python)"
+        if "$PYTHON_PATH" -c 'import sys; raise SystemExit(0 if sys.version_info >= (3, 11) else 1)' 2>/dev/null; then
+            PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
+            echo -e "${GREEN}✓${NC} $PYTHON_FOUND_VERSION found"
+        else
+            echo -e "${RED}✗${NC} Termux Python must be 3.11+"
+            echo "    Run: pkg install python"
+            exit 1
+        fi
+    else
+        echo -e "${RED}✗${NC} Python not found in Termux"
+        echo "    Run: pkg install python"
+        exit 1
+    fi
 else
-    echo -e "${CYAN}→${NC} Python $PYTHON_VERSION not found, installing via uv..."
-    $UV_CMD python install "$PYTHON_VERSION"
-    PYTHON_PATH=$($UV_CMD python find "$PYTHON_VERSION")
-    PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
-    echo -e "${GREEN}✓${NC} $PYTHON_FOUND_VERSION installed"
+    if $UV_CMD python find "$PYTHON_VERSION" &> /dev/null; then
+        PYTHON_PATH=$($UV_CMD python find "$PYTHON_VERSION")
+        PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
+        echo -e "${GREEN}✓${NC} $PYTHON_FOUND_VERSION found"
+    else
+        echo -e "${CYAN}→${NC} Python $PYTHON_VERSION not found, installing via uv..."
+        $UV_CMD python install "$PYTHON_VERSION"
+        PYTHON_PATH=$($UV_CMD python find "$PYTHON_VERSION")
+        PYTHON_FOUND_VERSION=$($PYTHON_PATH --version 2>/dev/null)
+        echo -e "${GREEN}✓${NC} $PYTHON_FOUND_VERSION installed"
+    fi
 fi
 
 # ============================================================================
@@ -104,11 +146,16 @@ if [ -d "venv" ]; then
     rm -rf venv
 fi
 
-$UV_CMD venv venv --python "$PYTHON_VERSION"
-echo -e "${GREEN}✓${NC} venv created (Python $PYTHON_VERSION)"
+if is_termux; then
+    "$PYTHON_PATH" -m venv venv
+    echo -e "${GREEN}✓${NC} venv created with stdlib venv"
+else
+    $UV_CMD venv venv --python "$PYTHON_VERSION"
+    echo -e "${GREEN}✓${NC} venv created (Python $PYTHON_VERSION)"
+fi
 
-# Tell uv to install into this venv (no activation needed for uv)
 export VIRTUAL_ENV="$SCRIPT_DIR/venv"
+SETUP_PYTHON="$SCRIPT_DIR/venv/bin/python"
 
 # ============================================================================
 # Dependencies
@@ -116,19 +163,34 @@ export VIRTUAL_ENV="$SCRIPT_DIR/venv"
 
 echo -e "${CYAN}→${NC} Installing dependencies..."
 
-# Prefer uv sync with lockfile (hash-verified installs) when available,
-# fall back to pip install for compatibility or when lockfile is stale.
-if [ -f "uv.lock" ]; then
-    echo -e "${CYAN}→${NC} Using uv.lock for hash-verified installation..."
-    UV_PROJECT_ENVIRONMENT="$SCRIPT_DIR/venv" $UV_CMD sync --all-extras --locked 2>/dev/null && \
-        echo -e "${GREEN}✓${NC} Dependencies installed (lockfile verified)" || {
-        echo -e "${YELLOW}⚠${NC} Lockfile install failed (may be outdated), falling back to pip install..."
+if is_termux; then
+    export ANDROID_API_LEVEL="$(getprop ro.build.version.sdk 2>/dev/null || printf '%s' "${ANDROID_API_LEVEL:-}")"
+    echo -e "${CYAN}→${NC} Termux detected — installing the tested Android bundle"
+    "$SETUP_PYTHON" -m pip install --upgrade pip setuptools wheel
+    if [ -f "constraints-termux.txt" ]; then
+        "$SETUP_PYTHON" -m pip install -e ".[termux]" -c constraints-termux.txt || {
+            echo -e "${YELLOW}⚠${NC} Termux bundle install failed, falling back to base install..."
+            "$SETUP_PYTHON" -m pip install -e "." -c constraints-termux.txt
+        }
+    else
+        "$SETUP_PYTHON" -m pip install -e ".[termux]" || "$SETUP_PYTHON" -m pip install -e "."
+    fi
+    echo -e "${GREEN}✓${NC} Dependencies installed"
+else
+    # Prefer uv sync with lockfile (hash-verified installs) when available,
+    # fall back to pip install for compatibility or when lockfile is stale.
+    if [ -f "uv.lock" ]; then
+        echo -e "${CYAN}→${NC} Using uv.lock for hash-verified installation..."
+        UV_PROJECT_ENVIRONMENT="$SCRIPT_DIR/venv" $UV_CMD sync --all-extras --locked 2>/dev/null && \
+            echo -e "${GREEN}✓${NC} Dependencies installed (lockfile verified)" || {
+            echo -e "${YELLOW}⚠${NC} Lockfile install failed (may be outdated), falling back to pip install..."
+            $UV_CMD pip install -e ".[all]" || $UV_CMD pip install -e "."
+            echo -e "${GREEN}✓${NC} Dependencies installed"
+        }
+    else
         $UV_CMD pip install -e ".[all]" || $UV_CMD pip install -e "."
         echo -e "${GREEN}✓${NC} Dependencies installed"
-    }
-else
-    $UV_CMD pip install -e ".[all]" || $UV_CMD pip install -e "."
-    echo -e "${GREEN}✓${NC} Dependencies installed"
+    fi
 fi
 
 # ============================================================================
@@ -138,7 +200,9 @@ fi
 echo -e "${CYAN}→${NC} Installing optional submodules..."
 
 # tinker-atropos (RL training backend)
-if [ -d "tinker-atropos" ] && [ -f "tinker-atropos/pyproject.toml" ]; then
+if is_termux; then
+    echo -e "${CYAN}→${NC} Skipping tinker-atropos on Termux (not part of the tested Android path)"
+elif [ -d "tinker-atropos" ] && [ -f "tinker-atropos/pyproject.toml" ]; then
     $UV_CMD pip install -e "./tinker-atropos" && \
         echo -e "${GREEN}✓${NC} tinker-atropos installed" || \
         echo -e "${YELLOW}⚠${NC} tinker-atropos install failed (RL tools may not work)"
@@ -160,34 +224,42 @@ else
     echo
     if [[ $REPLY =~ ^[Yy]$ ]] || [[ -z $REPLY ]]; then
         INSTALLED=false
-        
-        # Check if sudo is available
-        if command -v sudo &> /dev/null && sudo -n true 2>/dev/null; then
-            if command -v apt &> /dev/null; then
-                sudo apt install -y ripgrep && INSTALLED=true
-            elif command -v dnf &> /dev/null; then
-                sudo dnf install -y ripgrep && INSTALLED=true
+
+        if is_termux; then
+            pkg install -y ripgrep && INSTALLED=true
+        else
+            # Check if sudo is available
+            if command -v sudo &> /dev/null && sudo -n true 2>/dev/null; then
+                if command -v apt &> /dev/null; then
+                    sudo apt install -y ripgrep && INSTALLED=true
+                elif command -v dnf &> /dev/null; then
+                    sudo dnf install -y ripgrep && INSTALLED=true
+                fi
+            fi
+
+            # Try brew (no sudo needed)
+            if [ "$INSTALLED" = false ] && command -v brew &> /dev/null; then
+                brew install ripgrep && INSTALLED=true
+            fi
+
+            # Try cargo (no sudo needed)
+            if [ "$INSTALLED" = false ] && command -v cargo &> /dev/null; then
+                echo -e "${CYAN}→${NC} Trying cargo install (no sudo required)..."
+                cargo install ripgrep && INSTALLED=true
             fi
         fi
-        
-        # Try brew (no sudo needed)
-        if [ "$INSTALLED" = false ] && command -v brew &> /dev/null; then
-            brew install ripgrep && INSTALLED=true
-        fi
-        
-        # Try cargo (no sudo needed)
-        if [ "$INSTALLED" = false ] && command -v cargo &> /dev/null; then
-            echo -e "${CYAN}→${NC} Trying cargo install (no sudo required)..."
-            cargo install ripgrep && INSTALLED=true
-        fi
-        
+
         if [ "$INSTALLED" = true ]; then
             echo -e "${GREEN}✓${NC} ripgrep installed"
         else
             echo -e "${YELLOW}⚠${NC} Auto-install failed. Install options:"
-            echo "    sudo apt install ripgrep     # Debian/Ubuntu"
-            echo "    brew install ripgrep         # macOS"
-            echo "    cargo install ripgrep        # With Rust (no sudo)"
+            if is_termux; then
+                echo "    pkg install ripgrep          # Termux / Android"
+            else
+                echo "    sudo apt install ripgrep     # Debian/Ubuntu"
+                echo "    brew install ripgrep         # macOS"
+                echo "    cargo install ripgrep        # With Rust (no sudo)"
+            fi
             echo "    https://github.com/BurntSushi/ripgrep#installation"
         fi
     fi
@@ -207,49 +279,56 @@ else
 fi
 
 # ============================================================================
-# PATH setup — symlink hermes into ~/.local/bin
+# PATH setup — symlink hermes into a user-facing bin dir
 # ============================================================================
 
 echo -e "${CYAN}→${NC} Setting up hermes command..."
 
 HERMES_BIN="$SCRIPT_DIR/venv/bin/hermes"
-mkdir -p "$HOME/.local/bin"
-ln -sf "$HERMES_BIN" "$HOME/.local/bin/hermes"
-echo -e "${GREEN}✓${NC} Symlinked hermes → ~/.local/bin/hermes"
+COMMAND_LINK_DIR="$(get_command_link_dir)"
+COMMAND_LINK_DISPLAY_DIR="$(get_command_link_display_dir)"
+mkdir -p "$COMMAND_LINK_DIR"
+ln -sf "$HERMES_BIN" "$COMMAND_LINK_DIR/hermes"
+echo -e "${GREEN}✓${NC} Symlinked hermes → $COMMAND_LINK_DISPLAY_DIR/hermes"
 
-# Determine the appropriate shell config file
-SHELL_CONFIG=""
-if [[ "$SHELL" == *"zsh"* ]]; then
-    SHELL_CONFIG="$HOME/.zshrc"
-elif [[ "$SHELL" == *"bash"* ]]; then
-    SHELL_CONFIG="$HOME/.bashrc"
-    [ ! -f "$SHELL_CONFIG" ] && SHELL_CONFIG="$HOME/.bash_profile"
+if is_termux; then
+    export PATH="$COMMAND_LINK_DIR:$PATH"
+    echo -e "${GREEN}✓${NC} $COMMAND_LINK_DISPLAY_DIR is already on PATH in Termux"
 else
-    # Fallback to checking existing files
-    if [ -f "$HOME/.zshrc" ]; then
+    # Determine the appropriate shell config file
+    SHELL_CONFIG=""
+    if [[ "$SHELL" == *"zsh"* ]]; then
         SHELL_CONFIG="$HOME/.zshrc"
-    elif [ -f "$HOME/.bashrc" ]; then
+    elif [[ "$SHELL" == *"bash"* ]]; then
         SHELL_CONFIG="$HOME/.bashrc"
-    elif [ -f "$HOME/.bash_profile" ]; then
-        SHELL_CONFIG="$HOME/.bash_profile"
-    fi
-fi
-
-if [ -n "$SHELL_CONFIG" ]; then
-    # Touch the file just in case it doesn't exist yet but was selected
-    touch "$SHELL_CONFIG" 2>/dev/null || true
-    
-    if ! echo "$PATH" | tr ':' '\n' | grep -q "^$HOME/.local/bin$"; then
-        if ! grep -q '\.local/bin' "$SHELL_CONFIG" 2>/dev/null; then
-            echo "" >> "$SHELL_CONFIG"
-            echo "# Hermes Agent — ensure ~/.local/bin is on PATH" >> "$SHELL_CONFIG"
-            echo 'export PATH="$HOME/.local/bin:$PATH"' >> "$SHELL_CONFIG"
-            echo -e "${GREEN}✓${NC} Added ~/.local/bin to PATH in $SHELL_CONFIG"
-        else
-            echo -e "${GREEN}✓${NC} ~/.local/bin already in $SHELL_CONFIG"
-        fi
+        [ ! -f "$SHELL_CONFIG" ] && SHELL_CONFIG="$HOME/.bash_profile"
     else
-        echo -e "${GREEN}✓${NC} ~/.local/bin already on PATH"
+        # Fallback to checking existing files
+        if [ -f "$HOME/.zshrc" ]; then
+            SHELL_CONFIG="$HOME/.zshrc"
+        elif [ -f "$HOME/.bashrc" ]; then
+            SHELL_CONFIG="$HOME/.bashrc"
+        elif [ -f "$HOME/.bash_profile" ]; then
+            SHELL_CONFIG="$HOME/.bash_profile"
+        fi
+    fi
+
+    if [ -n "$SHELL_CONFIG" ]; then
+        # Touch the file just in case it doesn't exist yet but was selected
+        touch "$SHELL_CONFIG" 2>/dev/null || true
+
+        if ! echo "$PATH" | tr ':' '\n' | grep -q "^$HOME/.local/bin$"; then
+            if ! grep -q '\.local/bin' "$SHELL_CONFIG" 2>/dev/null; then
+                echo "" >> "$SHELL_CONFIG"
+                echo "# Hermes Agent — ensure ~/.local/bin is on PATH" >> "$SHELL_CONFIG"
+                echo 'export PATH="$HOME/.local/bin:$PATH"' >> "$SHELL_CONFIG"
+                echo -e "${GREEN}✓${NC} Added ~/.local/bin to PATH in $SHELL_CONFIG"
+            else
+                echo -e "${GREEN}✓${NC} ~/.local/bin already in $SHELL_CONFIG"
+            fi
+        else
+            echo -e "${GREEN}✓${NC} ~/.local/bin already on PATH"
+        fi
     fi
 fi
 
@@ -281,18 +360,31 @@ echo -e "${GREEN}✓ Setup complete!${NC}"
 echo ""
 echo "Next steps:"
 echo ""
-echo "  1. Reload your shell:"
-echo "     source $SHELL_CONFIG"
-echo ""
-echo "  2. Run the setup wizard to configure API keys:"
-echo "     hermes setup"
-echo ""
-echo "  3. Start chatting:"
-echo "     hermes"
-echo ""
+if is_termux; then
+    echo "  1. Run the setup wizard to configure API keys:"
+    echo "     hermes setup"
+    echo ""
+    echo "  2. Start chatting:"
+    echo "     hermes"
+    echo ""
+else
+    echo "  1. Reload your shell:"
+    echo "     source $SHELL_CONFIG"
+    echo ""
+    echo "  2. Run the setup wizard to configure API keys:"
+    echo "     hermes setup"
+    echo ""
+    echo "  3. Start chatting:"
+    echo "     hermes"
+    echo ""
+fi
 echo "Other commands:"
 echo "  hermes status        # Check configuration"
-echo "  hermes gateway install # Install gateway service (messaging + cron)"
+if is_termux; then
+    echo "  hermes gateway       # Run gateway in foreground"
+else
+    echo "  hermes gateway install # Install gateway service (messaging + cron)"
+fi
 echo "  hermes cron list     # View scheduled jobs"
 echo "  hermes doctor        # Diagnose issues"
 echo ""

tests/acp/test_server.py

@@ -68,9 +68,22 @@ class TestInitialize:
         resp = await agent.initialize(protocol_version=1)
         caps = resp.agent_capabilities
         assert isinstance(caps, AgentCapabilities)
+        assert caps.load_session is True
         assert caps.session_capabilities is not None
         assert caps.session_capabilities.fork is not None
         assert caps.session_capabilities.list is not None
+        assert caps.session_capabilities.resume is not None
+
+    @pytest.mark.asyncio
+    async def test_initialize_capabilities_wire_format(self, agent):
+        """Verify the JSON wire format uses correct aliases so ACP clients see the right keys."""
+        resp = await agent.initialize(protocol_version=1)
+        payload = resp.agent_capabilities.model_dump(by_alias=True, exclude_none=True)
+        assert payload["loadSession"] is True
+        session_caps = payload["sessionCapabilities"]
+        assert "fork" in session_caps
+        assert "list" in session_caps
+        assert "resume" in session_caps
 
 
 # ---------------------------------------------------------------------------
@@ -410,6 +423,37 @@ class TestPrompt:
         update = last_call[1].get("update") or last_call[0][1]
         assert update.session_update == "agent_message_chunk"
 
+    @pytest.mark.asyncio
+    async def test_prompt_populates_usage_from_top_level_run_conversation_fields(self, agent):
+        """ACP should map top-level token fields into PromptResponse.usage."""
+        new_resp = await agent.new_session(cwd=".")
+        state = agent.session_manager.get_session(new_resp.session_id)
+
+        state.agent.run_conversation = MagicMock(return_value={
+            "final_response": "usage attached",
+            "messages": [],
+            "prompt_tokens": 123,
+            "completion_tokens": 45,
+            "total_tokens": 168,
+            "reasoning_tokens": 7,
+            "cache_read_tokens": 11,
+        })
+
+        mock_conn = MagicMock(spec=acp.Client)
+        mock_conn.session_update = AsyncMock()
+        agent._conn = mock_conn
+
+        prompt = [TextContentBlock(type="text", text="show usage")]
+        resp = await agent.prompt(prompt=prompt, session_id=new_resp.session_id)
+
+        assert isinstance(resp, PromptResponse)
+        assert resp.usage is not None
+        assert resp.usage.input_tokens == 123
+        assert resp.usage.output_tokens == 45
+        assert resp.usage.total_tokens == 168
+        assert resp.usage.thought_tokens == 7
+        assert resp.usage.cached_read_tokens == 11
+
     @pytest.mark.asyncio
     async def test_prompt_cancelled_returns_cancelled_stop_reason(self, agent):
         """If cancel is called during prompt, stop_reason should be 'cancelled'."""

tests/agent/test_anthropic_adapter.py

@@ -80,6 +80,9 @@ class TestBuildAnthropicClient:
             build_anthropic_client("sk-ant-api03-x", base_url="https://custom.api.com")
             kwargs = mock_sdk.Anthropic.call_args[1]
             assert kwargs["base_url"] == "https://custom.api.com"
+            assert kwargs["default_headers"] == {
+                "anthropic-beta": "interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14"
+            }
 
     def test_minimax_anthropic_endpoint_uses_bearer_auth_for_regular_api_keys(self):
         with patch("agent.anthropic_adapter._anthropic_sdk") as mock_sdk:
@@ -91,7 +94,20 @@ class TestBuildAnthropicClient:
             assert kwargs["auth_token"] == "minimax-secret-123"
             assert "api_key" not in kwargs
             assert kwargs["default_headers"] == {
-                "anthropic-beta": "interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14"
+                "anthropic-beta": "interleaved-thinking-2025-05-14"
+            }
+
+    def test_minimax_cn_anthropic_endpoint_omits_tool_streaming_beta(self):
+        with patch("agent.anthropic_adapter._anthropic_sdk") as mock_sdk:
+            build_anthropic_client(
+                "minimax-cn-secret-123",
+                base_url="https://api.minimaxi.com/anthropic",
+            )
+            kwargs = mock_sdk.Anthropic.call_args[1]
+            assert kwargs["auth_token"] == "minimax-cn-secret-123"
+            assert "api_key" not in kwargs
+            assert kwargs["default_headers"] == {
+                "anthropic-beta": "interleaved-thinking-2025-05-14"
             }
 
 

tests/agent/test_auxiliary_client.py

@@ -658,6 +658,19 @@ class TestGetTextAuxiliaryClient:
         assert client is None
         assert model is None
 
+    def test_custom_endpoint_uses_codex_wrapper_when_runtime_requests_responses_api(self):
+        with patch("agent.auxiliary_client._resolve_custom_runtime",
+                   return_value=("https://api.openai.com/v1", "sk-test", "codex_responses")), \
+             patch("agent.auxiliary_client._read_main_model", return_value="gpt-5.3-codex"), \
+             patch("agent.auxiliary_client.OpenAI") as mock_openai:
+            client, model = get_text_auxiliary_client()
+
+        from agent.auxiliary_client import CodexAuxiliaryClient
+        assert isinstance(client, CodexAuxiliaryClient)
+        assert model == "gpt-5.3-codex"
+        assert mock_openai.call_args.kwargs["base_url"] == "https://api.openai.com/v1"
+        assert mock_openai.call_args.kwargs["api_key"] == "sk-test"
+
 
 class TestVisionClientFallback:
     """Vision client auto mode resolves known-good multimodal backends."""
@@ -1111,3 +1124,45 @@ class TestCallLlmPaymentFallback:
                     task="compression",
                     messages=[{"role": "user", "content": "hello"}],
                 )
+
+
+# ---------------------------------------------------------------------------
+# Gate: _resolve_api_key_provider must skip anthropic when not configured
+# ---------------------------------------------------------------------------
+
+
+def test_resolve_api_key_provider_skips_unconfigured_anthropic(monkeypatch):
+    """_resolve_api_key_provider must not try anthropic when user never configured it."""
+    from collections import OrderedDict
+    from hermes_cli.auth import ProviderConfig
+
+    # Build a minimal registry with only "anthropic" so the loop is guaranteed
+    # to reach it without being short-circuited by earlier providers.
+    fake_registry = OrderedDict({
+        "anthropic": ProviderConfig(
+            id="anthropic",
+            name="Anthropic",
+            auth_type="api_key",
+            inference_base_url="https://api.anthropic.com",
+            api_key_env_vars=("ANTHROPIC_API_KEY",),
+        ),
+    })
+
+    called = []
+
+    def mock_try_anthropic():
+        called.append("anthropic")
+        return None, None
+
+    monkeypatch.setattr("agent.auxiliary_client._try_anthropic", mock_try_anthropic)
+    monkeypatch.setattr("hermes_cli.auth.PROVIDER_REGISTRY", fake_registry)
+    monkeypatch.setattr(
+        "hermes_cli.auth.is_provider_explicitly_configured",
+        lambda pid: False,
+    )
+
+    from agent.auxiliary_client import _resolve_api_key_provider
+    _resolve_api_key_provider()
+
+    assert "anthropic" not in called, \
+        "_try_anthropic() should not be called when anthropic is not explicitly configured"

tests/agent/test_auxiliary_named_custom_providers.py

@@ -12,6 +12,17 @@ def _isolate(tmp_path, monkeypatch):
     hermes_home = tmp_path / ".hermes"
     hermes_home.mkdir()
     monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+    for env_var in (
+        "AUXILIARY_VISION_PROVIDER",
+        "AUXILIARY_VISION_MODEL",
+        "AUXILIARY_VISION_BASE_URL",
+        "AUXILIARY_VISION_API_KEY",
+        "CONTEXT_VISION_PROVIDER",
+        "CONTEXT_VISION_MODEL",
+        "CONTEXT_VISION_BASE_URL",
+        "CONTEXT_VISION_API_KEY",
+    ):
+        monkeypatch.delenv(env_var, raising=False)
     # Write a minimal config so load_config doesn't fail
     (hermes_home / "config.yaml").write_text("model:\n  default: test-model\n")
 
@@ -149,3 +160,83 @@ class TestResolveProviderClientNamedCustom:
         # "coffee" doesn't exist in custom_providers
         client, model = resolve_provider_client("coffee", "test")
         assert client is None
+
+
+class TestResolveProviderClientModelNormalization:
+    """Direct-provider auxiliary routing should normalize models like main runtime."""
+
+    def test_matching_native_prefix_is_stripped_for_main_provider(self, tmp_path):
+        _write_config(tmp_path, {
+            "model": {"default": "zai/glm-5.1", "provider": "zai"},
+        })
+        with (
+            patch("hermes_cli.auth.resolve_api_key_provider_credentials", return_value={
+                "api_key": "glm-key",
+                "base_url": "https://api.z.ai/api/paas/v4",
+            }),
+            patch("agent.auxiliary_client.OpenAI") as mock_openai,
+        ):
+            mock_openai.return_value = MagicMock()
+            from agent.auxiliary_client import resolve_provider_client
+
+            client, model = resolve_provider_client("main", "zai/glm-5.1")
+
+        assert client is not None
+        assert model == "glm-5.1"
+
+    def test_non_matching_prefix_is_preserved_for_direct_provider(self, tmp_path):
+        _write_config(tmp_path, {
+            "model": {"default": "zai/glm-5.1", "provider": "zai"},
+        })
+        with (
+            patch("hermes_cli.auth.resolve_api_key_provider_credentials", return_value={
+                "api_key": "glm-key",
+                "base_url": "https://api.z.ai/api/paas/v4",
+            }),
+            patch("agent.auxiliary_client.OpenAI") as mock_openai,
+        ):
+            mock_openai.return_value = MagicMock()
+            from agent.auxiliary_client import resolve_provider_client
+
+            client, model = resolve_provider_client("zai", "google/gemini-2.5-pro")
+
+        assert client is not None
+        assert model == "google/gemini-2.5-pro"
+
+    def test_aggregator_vendor_slug_is_preserved(self, monkeypatch):
+        monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
+        with patch("agent.auxiliary_client.OpenAI") as mock_openai:
+            mock_openai.return_value = MagicMock()
+            from agent.auxiliary_client import resolve_provider_client
+
+            client, model = resolve_provider_client(
+                "openrouter", "anthropic/claude-sonnet-4.6"
+            )
+
+        assert client is not None
+        assert model == "anthropic/claude-sonnet-4.6"
+
+
+class TestResolveVisionProviderClientModelNormalization:
+    """Vision auto-routing should reuse the same provider-specific normalization."""
+
+    def test_vision_auto_strips_matching_main_provider_prefix(self, tmp_path):
+        _write_config(tmp_path, {
+            "model": {"default": "zai/glm-5.1", "provider": "zai"},
+        })
+        with (
+            patch("agent.auxiliary_client._read_nous_auth", return_value=None),
+            patch("hermes_cli.auth.resolve_api_key_provider_credentials", return_value={
+                "api_key": "glm-key",
+                "base_url": "https://api.z.ai/api/paas/v4",
+            }),
+            patch("agent.auxiliary_client.OpenAI") as mock_openai,
+        ):
+            mock_openai.return_value = MagicMock()
+            from agent.auxiliary_client import resolve_vision_provider_client
+
+            provider, client, model = resolve_vision_provider_client()
+
+        assert provider == "zai"
+        assert client is not None
+        assert model == "glm-5.1"

tests/agent/test_context_references.py

@@ -83,6 +83,24 @@ def test_parse_references_strips_trailing_punctuation():
     assert refs[1].target == "https://example.com/docs"
 
 
+def test_parse_quoted_references_with_spaces_and_preserve_unquoted_ranges():
+    from agent.context_references import parse_context_references
+
+    refs = parse_context_references(
+        'review @file:"C:\\Users\\Simba\\My Project\\main.py":7-9 '
+        'and @folder:"docs and specs" plus @file:src/main.py:1-2'
+    )
+
+    assert [ref.kind for ref in refs] == ["file", "folder", "file"]
+    assert refs[0].target == r"C:\Users\Simba\My Project\main.py"
+    assert refs[0].line_start == 7
+    assert refs[0].line_end == 9
+    assert refs[1].target == "docs and specs"
+    assert refs[2].target == "src/main.py"
+    assert refs[2].line_start == 1
+    assert refs[2].line_end == 2
+
+
 def test_expand_file_range_and_folder_listing(sample_repo: Path):
     from agent.context_references import preprocess_context_references
 
@@ -106,6 +124,30 @@ def test_expand_file_range_and_folder_listing(sample_repo: Path):
     assert not result.warnings
 
 
+def test_expand_quoted_file_reference_with_spaces(tmp_path: Path):
+    from agent.context_references import preprocess_context_references
+
+    workspace = tmp_path / "repo"
+    folder = workspace / "docs and specs"
+    folder.mkdir(parents=True)
+    file_path = folder / "release notes.txt"
+    file_path.write_text("line 1\nline 2\nline 3\n", encoding="utf-8")
+
+    result = preprocess_context_references(
+        'Review @file:"docs and specs/release notes.txt":2-3',
+        cwd=workspace,
+        context_length=100_000,
+    )
+
+    assert result.expanded
+    assert result.message.startswith("Review")
+    assert "line 1" not in result.message
+    assert "line 2" in result.message
+    assert "line 3" in result.message
+    assert "release notes.txt" in result.message
+    assert not result.warnings
+
+
 def test_expand_git_diff_staged_and_log(sample_repo: Path):
     from agent.context_references import preprocess_context_references
 

tests/agent/test_credential_pool.py

@@ -567,6 +567,7 @@ def test_singleton_seed_does_not_clobber_manual_oauth_entry(tmp_path, monkeypatc
     monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
     monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
     monkeypatch.delenv("CLAUDE_CODE_OAUTH_TOKEN", raising=False)
+    monkeypatch.setattr("hermes_cli.auth.is_provider_explicitly_configured", lambda pid: True)
     _write_auth_store(
         tmp_path,
         {
@@ -702,53 +703,6 @@ def test_least_used_strategy_selects_lowest_count(tmp_path, monkeypatch):
     assert entry.access_token == "sk-or-light"
 
 
-def test_mark_used_increments_request_count(tmp_path, monkeypatch):
-    """mark_used should increment the request_count of the current entry."""
-    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
-    monkeypatch.setattr(
-        "agent.credential_pool.get_pool_strategy",
-        lambda _provider: "fill_first",
-    )
-    monkeypatch.setattr(
-        "agent.credential_pool._seed_from_singletons",
-        lambda provider, entries: (False, set()),
-    )
-    monkeypatch.setattr(
-        "agent.credential_pool._seed_from_env",
-        lambda provider, entries: (False, set()),
-    )
-    _write_auth_store(
-        tmp_path,
-        {
-            "version": 1,
-            "credential_pool": {
-                "openrouter": [
-                    {
-                        "id": "key-a",
-                        "label": "test",
-                        "auth_type": "api_key",
-                        "priority": 0,
-                        "source": "manual",
-                        "access_token": "sk-or-test",
-                        "request_count": 5,
-                    },
-                ]
-            },
-        },
-    )
-
-    from agent.credential_pool import load_pool
-
-    pool = load_pool("openrouter")
-    entry = pool.select()
-    assert entry is not None
-    assert entry.request_count == 5
-    pool.mark_used()
-    updated = pool.current()
-    assert updated is not None
-    assert updated.request_count == 6
-
-
 def test_thread_safety_concurrent_select(tmp_path, monkeypatch):
     """Concurrent select() calls should not corrupt pool state."""
     import threading as _threading
@@ -798,7 +752,6 @@ def test_thread_safety_concurrent_select(tmp_path, monkeypatch):
                 entry = pool.select()
                 if entry:
                     results.append(entry.id)
-                    pool.mark_used(entry.id)
         except Exception as exc:
             errors.append(exc)
 
@@ -1056,8 +1009,8 @@ def test_acquire_lease_prefers_unleased_entry(tmp_path, monkeypatch):
 
     assert first == "cred-1"
     assert second == "cred-2"
-    assert pool.active_lease_count("cred-1") == 1
-    assert pool.active_lease_count("cred-2") == 1
+    assert pool._active_leases.get("cred-1", 0) == 1
+    assert pool._active_leases.get("cred-2", 0) == 1
 
 
 
@@ -1087,7 +1040,34 @@ def test_release_lease_decrements_counter(tmp_path, monkeypatch):
     pool = load_pool("openrouter")
     leased = pool.acquire_lease()
     assert leased == "cred-1"
-    assert pool.active_lease_count("cred-1") == 1
+    assert pool._active_leases.get("cred-1", 0) == 1
 
     pool.release_lease("cred-1")
-    assert pool.active_lease_count("cred-1") == 0
+    assert pool._active_leases.get("cred-1", 0) == 0
+
+
+def test_load_pool_does_not_seed_claude_code_when_anthropic_not_configured(tmp_path, monkeypatch):
+    """Claude Code credentials must not be auto-seeded when the user never selected anthropic."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    _write_auth_store(tmp_path, {"version": 1, "credential_pool": {}})
+
+    # Claude Code credentials exist on disk
+    monkeypatch.setattr(
+        "agent.anthropic_adapter.read_claude_code_credentials",
+        lambda: {"accessToken": "sk-ant...oken", "refreshToken": "rt", "expiresAt": 9999999999999},
+    )
+    monkeypatch.setattr(
+        "agent.anthropic_adapter.read_hermes_oauth_credentials",
+        lambda: None,
+    )
+    # User configured kimi-coding, NOT anthropic
+    monkeypatch.setattr(
+        "hermes_cli.auth.is_provider_explicitly_configured",
+        lambda pid: pid == "kimi-coding",
+    )
+
+    from agent.credential_pool import load_pool
+    pool = load_pool("anthropic")
+
+    # Should NOT have seeded the claude_code entry
+    assert pool.entries() == []

tests/agent/test_error_classifier.py

@@ -75,28 +75,6 @@ class TestClassifiedError:
         e3 = ClassifiedError(reason=FailoverReason.billing)
         assert e3.is_auth is False
 
-    def test_is_transient_property(self):
-        transient_reasons = [
-            FailoverReason.rate_limit,
-            FailoverReason.overloaded,
-            FailoverReason.server_error,
-            FailoverReason.timeout,
-            FailoverReason.unknown,
-        ]
-        for reason in transient_reasons:
-            e = ClassifiedError(reason=reason)
-            assert e.is_transient is True, f"{reason} should be transient"
-
-        non_transient = [
-            FailoverReason.auth,
-            FailoverReason.billing,
-            FailoverReason.model_not_found,
-            FailoverReason.format_error,
-        ]
-        for reason in non_transient:
-            e = ClassifiedError(reason=reason)
-            assert e.is_transient is False, f"{reason} should NOT be transient"
-
     def test_defaults(self):
         e = ClassifiedError(reason=FailoverReason.unknown)
         assert e.retryable is True
@@ -271,6 +249,22 @@ class TestClassifyApiError:
         assert result.reason == FailoverReason.rate_limit
         assert result.should_fallback is True
 
+    def test_alibaba_rate_increased_too_quickly(self):
+        """Alibaba/DashScope returns a unique throttling message.
+
+        Port from anomalyco/opencode#21355.
+        """
+        msg = (
+            "Upstream error from Alibaba: Request rate increased too quickly. "
+            "To ensure system stability, please adjust your client logic to "
+            "scale requests more smoothly over time."
+        )
+        e = MockAPIError(msg, status_code=400)
+        result = classify_api_error(e)
+        assert result.reason == FailoverReason.rate_limit
+        assert result.retryable is True
+        assert result.should_rotate_credential is True
+
     # ── Server errors ──
 
     def test_500_server_error(self):
@@ -480,6 +474,39 @@ class TestClassifyApiError:
         result = classify_api_error(e)
         assert result.reason == FailoverReason.context_overflow
 
+    # ── Message-only usage limit disambiguation (no status code) ──
+
+    def test_message_usage_limit_transient_is_rate_limit(self):
+        """'usage limit' + 'try again' with no status code → rate_limit, not billing."""
+        e = Exception("usage limit exceeded, try again in 5 minutes")
+        result = classify_api_error(e)
+        assert result.reason == FailoverReason.rate_limit
+        assert result.retryable is True
+        assert result.should_rotate_credential is True
+        assert result.should_fallback is True
+
+    def test_message_usage_limit_no_retry_signal_is_billing(self):
+        """'usage limit' with no transient signal and no status code → billing."""
+        e = Exception("usage limit reached")
+        result = classify_api_error(e)
+        assert result.reason == FailoverReason.billing
+        assert result.retryable is False
+        assert result.should_rotate_credential is True
+
+    def test_message_quota_with_reset_window_is_rate_limit(self):
+        """'quota' + 'resets at' with no status code → rate_limit."""
+        e = Exception("quota exceeded, resets at midnight UTC")
+        result = classify_api_error(e)
+        assert result.reason == FailoverReason.rate_limit
+        assert result.retryable is True
+
+    def test_message_limit_exceeded_with_wait_is_rate_limit(self):
+        """'limit exceeded' + 'wait' with no status code → rate_limit."""
+        e = Exception("key limit exceeded, please wait before retrying")
+        result = classify_api_error(e)
+        assert result.reason == FailoverReason.rate_limit
+        assert result.retryable is True
+
     # ── Unknown / fallback ──
 
     def test_generic_exception_is_unknown(self):

tests/agent/test_local_stream_timeout.py

@@ -0,0 +1,70 @@
+"""Tests for local provider stream read timeout auto-detection.
+
+When a local LLM provider is detected (Ollama, llama.cpp, vLLM, etc.),
+the httpx stream read timeout should be automatically increased from the
+default 60s to HERMES_API_TIMEOUT (1800s) to avoid premature connection
+kills during long prefill phases.
+"""
+
+import os
+import pytest
+from unittest.mock import patch
+
+from agent.model_metadata import is_local_endpoint
+
+
+class TestLocalStreamReadTimeout:
+    """Verify stream read timeout auto-detection logic."""
+
+    @pytest.mark.parametrize("base_url", [
+        "http://localhost:11434",
+        "http://127.0.0.1:8080",
+        "http://0.0.0.0:5000",
+        "http://192.168.1.100:8000",
+        "http://10.0.0.5:1234",
+    ])
+    def test_local_endpoint_bumps_read_timeout(self, base_url):
+        """Local endpoint + default timeout -> bumps to base_timeout."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 1800.0
+
+    def test_user_override_respected_for_local(self):
+        """User sets HERMES_STREAM_READ_TIMEOUT -> keep their value even for local."""
+        with patch.dict(os.environ, {"HERMES_STREAM_READ_TIMEOUT": "300"}, clear=False):
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            base_url = "http://localhost:11434"
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 300.0
+
+    @pytest.mark.parametrize("base_url", [
+        "https://api.openai.com",
+        "https://openrouter.ai/api",
+        "https://api.anthropic.com",
+    ])
+    def test_remote_endpoint_keeps_default(self, base_url):
+        """Remote endpoint -> keep 120s default."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 120.0
+
+    def test_empty_base_url_keeps_default(self):
+        """No base_url set -> keep 120s default."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            base_url = ""
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 120.0

tests/agent/test_minimax_provider.py

@@ -1,4 +1,6 @@
-"""Tests for MiniMax provider hardening — context lengths, thinking guard, catalog."""
+"""Tests for MiniMax provider hardening — context lengths, thinking guard, catalog, beta headers."""
+
+from unittest.mock import patch
 
 
 class TestMinimaxContextLengths:
@@ -103,3 +105,100 @@ class TestMinimaxModelCatalog:
             models = _PROVIDER_MODELS[provider]
             assert "MiniMax-M2.7-highspeed" not in models
             assert "MiniMax-M2.5-highspeed" not in models
+
+
+class TestMinimaxBetaHeaders:
+    """MiniMax Anthropic-compat endpoints reject fine-grained-tool-streaming beta.
+
+    Verify that build_anthropic_client omits the tool-streaming beta for MiniMax
+    (both global and China domains) while keeping it for native Anthropic and
+    other third-party endpoints.  Covers the fix for #6510 / #6555.
+    """
+
+    _TOOL_BETA = "fine-grained-tool-streaming-2025-05-14"
+    _THINKING_BETA = "interleaved-thinking-2025-05-14"
+
+    # -- helper ----------------------------------------------------------
+
+    def _build_and_get_betas(self, api_key, base_url=None):
+        """Build client, return the anthropic-beta header string."""
+        from agent.anthropic_adapter import build_anthropic_client
+        with patch("agent.anthropic_adapter._anthropic_sdk") as mock_sdk:
+            build_anthropic_client(api_key, base_url=base_url)
+            kwargs = mock_sdk.Anthropic.call_args[1]
+            headers = kwargs.get("default_headers", {})
+            return headers.get("anthropic-beta", "")
+
+    # -- MiniMax global --------------------------------------------------
+
+    def test_minimax_global_omits_tool_streaming(self):
+        betas = self._build_and_get_betas(
+            "mm-key-123", base_url="https://api.minimax.io/anthropic"
+        )
+        assert self._TOOL_BETA not in betas
+        assert self._THINKING_BETA in betas
+
+    def test_minimax_global_trailing_slash(self):
+        betas = self._build_and_get_betas(
+            "mm-key-123", base_url="https://api.minimax.io/anthropic/"
+        )
+        assert self._TOOL_BETA not in betas
+
+    # -- MiniMax China ---------------------------------------------------
+
+    def test_minimax_cn_omits_tool_streaming(self):
+        betas = self._build_and_get_betas(
+            "mm-cn-key-456", base_url="https://api.minimaxi.com/anthropic"
+        )
+        assert self._TOOL_BETA not in betas
+        assert self._THINKING_BETA in betas
+
+    def test_minimax_cn_trailing_slash(self):
+        betas = self._build_and_get_betas(
+            "mm-cn-key-456", base_url="https://api.minimaxi.com/anthropic/"
+        )
+        assert self._TOOL_BETA not in betas
+
+    # -- Non-MiniMax keeps full betas ------------------------------------
+
+    def test_native_anthropic_keeps_tool_streaming(self):
+        betas = self._build_and_get_betas("sk-ant-api03-real-key-here")
+        assert self._TOOL_BETA in betas
+        assert self._THINKING_BETA in betas
+
+    def test_third_party_proxy_keeps_tool_streaming(self):
+        betas = self._build_and_get_betas(
+            "custom-key", base_url="https://my-proxy.example.com/anthropic"
+        )
+        assert self._TOOL_BETA in betas
+
+    def test_custom_base_url_keeps_tool_streaming(self):
+        betas = self._build_and_get_betas(
+            "custom-key", base_url="https://custom.api.com"
+        )
+        assert self._TOOL_BETA in betas
+
+    # -- _common_betas_for_base_url unit tests ---------------------------
+
+    def test_common_betas_none_url(self):
+        from agent.anthropic_adapter import _common_betas_for_base_url, _COMMON_BETAS
+        assert _common_betas_for_base_url(None) == _COMMON_BETAS
+
+    def test_common_betas_empty_url(self):
+        from agent.anthropic_adapter import _common_betas_for_base_url, _COMMON_BETAS
+        assert _common_betas_for_base_url("") == _COMMON_BETAS
+
+    def test_common_betas_minimax_url(self):
+        from agent.anthropic_adapter import _common_betas_for_base_url, _TOOL_STREAMING_BETA
+        betas = _common_betas_for_base_url("https://api.minimax.io/anthropic")
+        assert _TOOL_STREAMING_BETA not in betas
+        assert len(betas) > 0  # still has other betas
+
+    def test_common_betas_minimax_cn_url(self):
+        from agent.anthropic_adapter import _common_betas_for_base_url, _TOOL_STREAMING_BETA
+        betas = _common_betas_for_base_url("https://api.minimaxi.com/anthropic")
+        assert _TOOL_STREAMING_BETA not in betas
+
+    def test_common_betas_regular_url(self):
+        from agent.anthropic_adapter import _common_betas_for_base_url, _COMMON_BETAS
+        assert _common_betas_for_base_url("https://api.anthropic.com") == _COMMON_BETAS

tests/agent/test_model_metadata.py

@@ -132,6 +132,61 @@ class TestDefaultContextLengths:
             if "gemini" in key:
                 assert value == 1048576, f"{key} should be 1048576"
 
+    def test_grok_models_context_lengths(self):
+        # xAI /v1/models does not return context_length metadata, so
+        # DEFAULT_CONTEXT_LENGTHS must cover the Grok family explicitly.
+        # Values sourced from models.dev (2026-04).
+        expected = {
+            "grok-4.20": 2000000,
+            "grok-4-1-fast": 2000000,
+            "grok-4-fast": 2000000,
+            "grok-4": 256000,
+            "grok-code-fast": 256000,
+            "grok-3": 131072,
+            "grok-2": 131072,
+            "grok-2-vision": 8192,
+            "grok": 131072,
+        }
+        for key, value in expected.items():
+            assert key in DEFAULT_CONTEXT_LENGTHS, f"{key} missing from DEFAULT_CONTEXT_LENGTHS"
+            assert DEFAULT_CONTEXT_LENGTHS[key] == value, (
+                f"{key} should be {value}, got {DEFAULT_CONTEXT_LENGTHS[key]}"
+            )
+
+    def test_grok_substring_matching(self):
+        # Longest-first substring matching must resolve the real xAI model
+        # IDs to the correct fallback entries without 128k probe-down.
+        from agent.model_metadata import get_model_context_length
+        from unittest.mock import patch as mock_patch
+
+        # Fake the provider/API/cache layers so the lookup falls through
+        # to DEFAULT_CONTEXT_LENGTHS.
+        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 = [
+                ("grok-4.20-0309-reasoning", 2000000),
+                ("grok-4.20-0309-non-reasoning", 2000000),
+                ("grok-4.20-multi-agent-0309", 2000000),
+                ("grok-4-1-fast-reasoning", 2000000),
+                ("grok-4-1-fast-non-reasoning", 2000000),
+                ("grok-4-fast-reasoning", 2000000),
+                ("grok-4-fast-non-reasoning", 2000000),
+                ("grok-4", 256000),
+                ("grok-4-0709", 256000),
+                ("grok-code-fast-1", 256000),
+                ("grok-3", 131072),
+                ("grok-3-mini", 131072),
+                ("grok-3-mini-fast", 131072),
+                ("grok-2", 131072),
+                ("grok-2-vision", 8192),
+                ("grok-2-vision-1212", 8192),
+                ("grok-beta", 131072),
+            ]
+            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"

tests/cli/test_cli_file_drop.py

@@ -147,6 +147,20 @@ class TestEscapedSpaces:
         assert result["path"] == tmp_image_with_spaces
         assert result["remainder"] == "what is this?"
 
+    def test_tilde_prefixed_path(self, tmp_path, monkeypatch):
+        home = tmp_path / "home"
+        img = home / "storage" / "shared" / "Pictures" / "cat.png"
+        img.parent.mkdir(parents=True, exist_ok=True)
+        img.write_bytes(b"\x89PNG\r\n\x1a\n")
+        monkeypatch.setenv("HOME", str(home))
+
+        result = _detect_file_drop("~/storage/shared/Pictures/cat.png what is this?")
+
+        assert result is not None
+        assert result["path"] == img
+        assert result["is_image"] is True
+        assert result["remainder"] == "what is this?"
+
 
 # ---------------------------------------------------------------------------
 # Tests: edge cases

tests/cli/test_cli_image_command.py

@@ -0,0 +1,109 @@
+from pathlib import Path
+from unittest.mock import patch
+
+from cli import (
+    HermesCLI,
+    _collect_query_images,
+    _format_image_attachment_badges,
+    _termux_example_image_path,
+)
+
+
+def _make_cli():
+    cli_obj = HermesCLI.__new__(HermesCLI)
+    cli_obj._attached_images = []
+    return cli_obj
+
+
+def _make_image(path: Path) -> Path:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_bytes(b"\x89PNG\r\n\x1a\n")
+    return path
+
+
+class TestImageCommand:
+    def test_handle_image_command_attaches_local_image(self, tmp_path):
+        img = _make_image(tmp_path / "photo.png")
+        cli_obj = _make_cli()
+
+        with patch("cli._cprint"):
+            cli_obj._handle_image_command(f"/image {img}")
+
+        assert cli_obj._attached_images == [img]
+
+    def test_handle_image_command_supports_quoted_path_with_spaces(self, tmp_path):
+        img = _make_image(tmp_path / "my photo.png")
+        cli_obj = _make_cli()
+
+        with patch("cli._cprint"):
+            cli_obj._handle_image_command(f'/image "{img}"')
+
+        assert cli_obj._attached_images == [img]
+
+    def test_handle_image_command_rejects_non_image_file(self, tmp_path):
+        file_path = tmp_path / "notes.txt"
+        file_path.write_text("hello\n", encoding="utf-8")
+        cli_obj = _make_cli()
+
+        with patch("cli._cprint") as mock_print:
+            cli_obj._handle_image_command(f"/image {file_path}")
+
+        assert cli_obj._attached_images == []
+        rendered = " ".join(str(arg) for call in mock_print.call_args_list for arg in call.args)
+        assert "Not a supported image file" in rendered
+
+
+class TestCollectQueryImages:
+    def test_collect_query_images_accepts_explicit_image_arg(self, tmp_path):
+        img = _make_image(tmp_path / "diagram.png")
+
+        message, images = _collect_query_images("describe this", str(img))
+
+        assert message == "describe this"
+        assert images == [img]
+
+    def test_collect_query_images_extracts_leading_path(self, tmp_path):
+        img = _make_image(tmp_path / "camera.png")
+
+        message, images = _collect_query_images(f"{img} what do you see?")
+
+        assert message == "what do you see?"
+        assert images == [img]
+
+    def test_collect_query_images_supports_tilde_paths(self, tmp_path, monkeypatch):
+        home = tmp_path / "home"
+        img = _make_image(home / "storage" / "shared" / "Pictures" / "cat.png")
+        monkeypatch.setenv("HOME", str(home))
+
+        message, images = _collect_query_images("describe this", "~/storage/shared/Pictures/cat.png")
+
+        assert message == "describe this"
+        assert images == [img]
+
+
+class TestTermuxImageHints:
+    def test_termux_example_image_path_prefers_real_shared_storage_root(self, monkeypatch):
+        existing = {"/sdcard", "/storage/emulated/0"}
+        monkeypatch.setattr("cli.os.path.isdir", lambda path: path in existing)
+
+        hint = _termux_example_image_path()
+
+        assert hint == "/sdcard/Pictures/cat.png"
+
+
+class TestImageBadgeFormatting:
+    def test_compact_badges_use_filename_on_narrow_terminals(self, tmp_path):
+        img = _make_image(tmp_path / "Screenshot 2026-04-09 at 11.22.33 AM.png")
+
+        badges = _format_image_attachment_badges([img], image_counter=1, width=40)
+
+        assert badges.startswith("[📎 ")
+        assert "Image #1" not in badges
+
+    def test_compact_badges_summarize_multiple_images(self, tmp_path):
+        img1 = _make_image(tmp_path / "one.png")
+        img2 = _make_image(tmp_path / "two.png")
+
+        badges = _format_image_attachment_badges([img1, img2], image_counter=2, width=45)
+
+        assert badges == "[📎 2 images attached]"

tests/cli/test_cli_skin_integration.py

@@ -49,6 +49,25 @@ class TestCliSkinPromptIntegration:
         set_active_skin("ares")
         assert cli._get_tui_prompt_fragments() == [("class:sudo-prompt", "🔑 ❯ ")]
 
+    def test_narrow_terminals_compact_voice_prompt_fragments(self):
+        cli = _make_cli_stub()
+        cli._voice_mode = True
+
+        with patch.object(HermesCLI, "_get_tui_terminal_width", return_value=50):
+            assert cli._get_tui_prompt_fragments() == [("class:voice-prompt", "🎤 ")]
+
+    def test_narrow_terminals_compact_voice_recording_prompt_fragments(self):
+        cli = _make_cli_stub()
+        cli._voice_recording = True
+        cli._voice_recorder = SimpleNamespace(current_rms=3000)
+
+        with patch.object(HermesCLI, "_get_tui_terminal_width", return_value=50):
+            frags = cli._get_tui_prompt_fragments()
+
+        assert frags[0][0] == "class:voice-recording"
+        assert frags[0][1].startswith("●")
+        assert "❯" not in frags[0][1]
+
     def test_icon_only_skin_symbol_still_visible_in_special_states(self):
         cli = _make_cli_stub()
         cli._secret_state = {"response_queue": object()}

tests/cli/test_cli_status_bar.py

@@ -206,6 +206,59 @@ class TestCLIStatusBar:
         assert "⚕" in text
         assert "claude-sonnet-4-20250514" in text
 
+    def test_minimal_tui_chrome_threshold(self):
+        cli_obj = _make_cli()
+
+        assert cli_obj._use_minimal_tui_chrome(width=63) is True
+        assert cli_obj._use_minimal_tui_chrome(width=64) is False
+
+    def test_bottom_input_rule_hides_on_narrow_terminals(self):
+        cli_obj = _make_cli()
+
+        assert cli_obj._tui_input_rule_height("top", width=50) == 1
+        assert cli_obj._tui_input_rule_height("bottom", width=50) == 0
+        assert cli_obj._tui_input_rule_height("bottom", width=90) == 1
+
+    def test_agent_spacer_reclaimed_on_narrow_terminals(self):
+        cli_obj = _make_cli()
+        cli_obj._agent_running = True
+
+        assert cli_obj._agent_spacer_height(width=50) == 0
+        assert cli_obj._agent_spacer_height(width=90) == 1
+        cli_obj._agent_running = False
+        assert cli_obj._agent_spacer_height(width=90) == 0
+
+    def test_spinner_line_hidden_on_narrow_terminals(self):
+        cli_obj = _make_cli()
+        cli_obj._spinner_text = "thinking"
+
+        assert cli_obj._spinner_widget_height(width=50) == 0
+        assert cli_obj._spinner_widget_height(width=90) == 1
+        cli_obj._spinner_text = ""
+        assert cli_obj._spinner_widget_height(width=90) == 0
+
+    def test_voice_status_bar_compacts_on_narrow_terminals(self):
+        cli_obj = _make_cli()
+        cli_obj._voice_mode = True
+        cli_obj._voice_recording = False
+        cli_obj._voice_processing = False
+        cli_obj._voice_tts = True
+        cli_obj._voice_continuous = True
+
+        fragments = cli_obj._get_voice_status_fragments(width=50)
+
+        assert fragments == [("class:voice-status", " 🎤 Ctrl+B ")]
+
+    def test_voice_recording_status_bar_compacts_on_narrow_terminals(self):
+        cli_obj = _make_cli()
+        cli_obj._voice_mode = True
+        cli_obj._voice_recording = True
+        cli_obj._voice_processing = False
+
+        fragments = cli_obj._get_voice_status_fragments(width=50)
+
+        assert fragments == [("class:voice-status-recording", " ● REC ")]
+
 
 class TestCLIUsageReport:
     def test_show_usage_includes_estimated_cost(self, capsys):

tests/cli/test_cli_status_command.py

@@ -0,0 +1,85 @@
+"""Tests for CLI /status command behavior."""
+from datetime import datetime
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+from cli import HermesCLI
+from hermes_cli.commands import resolve_command
+
+
+def _make_cli():
+    cli_obj = HermesCLI.__new__(HermesCLI)
+    cli_obj.config = {}
+    cli_obj.console = MagicMock()
+    cli_obj.agent = None
+    cli_obj.conversation_history = []
+    cli_obj.session_id = "session-123"
+    cli_obj._pending_input = MagicMock()
+    cli_obj._status_bar_visible = True
+    cli_obj.model = "openai/gpt-5.4"
+    cli_obj.provider = "openai"
+    cli_obj.session_start = datetime(2026, 4, 9, 19, 24)
+    cli_obj._agent_running = False
+    cli_obj._session_db = MagicMock()
+    cli_obj._session_db.get_session.return_value = None
+    return cli_obj
+
+
+def test_status_command_is_available_in_cli_registry():
+    cmd = resolve_command("status")
+    assert cmd is not None
+    assert cmd.gateway_only is False
+
+
+def test_process_command_status_dispatches_without_toggling_status_bar():
+    cli_obj = _make_cli()
+
+    with patch.object(cli_obj, "_show_session_status", create=True) as mock_status:
+        assert cli_obj.process_command("/status") is True
+
+    mock_status.assert_called_once_with()
+    assert cli_obj._status_bar_visible is True
+
+
+def test_statusbar_still_toggles_visibility():
+    cli_obj = _make_cli()
+
+    assert cli_obj.process_command("/statusbar") is True
+    assert cli_obj._status_bar_visible is False
+
+
+def test_status_prefix_prefers_status_command_over_statusbar_toggle():
+    cli_obj = _make_cli()
+
+    with patch.object(cli_obj, "_show_session_status") as mock_status:
+        assert cli_obj.process_command("/sta") is True
+
+    mock_status.assert_called_once_with()
+    assert cli_obj._status_bar_visible is True
+
+
+def test_show_session_status_prints_gateway_style_summary():
+    cli_obj = _make_cli()
+    cli_obj.agent = SimpleNamespace(
+        session_total_tokens=321,
+        session_api_calls=4,
+    )
+    cli_obj._session_db.get_session.return_value = {
+        "title": "My titled session",
+        "started_at": 1775791440,
+    }
+
+    with patch("cli.display_hermes_home", return_value="~/.hermes"):
+        cli_obj._show_session_status()
+
+    printed = "\n".join(str(call.args[0]) for call in cli_obj.console.print.call_args_list)
+    assert "Hermes CLI Status" in printed
+    assert "Session ID: session-123" in printed
+    assert "Path: ~/.hermes" in printed
+    assert "Title: My titled session" in printed
+    assert "Model: openai/gpt-5.4 (openai)" in printed
+    assert "Tokens: 321" in printed
+    assert "Agent Running: No" in printed
+    _, kwargs = cli_obj.console.print.call_args
+    assert kwargs.get("highlight") is False
+    assert kwargs.get("markup") is False

tests/cli/test_fast_command.py

@@ -0,0 +1,413 @@
+"""Tests for the /fast CLI command and service-tier config handling."""
+
+import unittest
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+
+def _import_cli():
+    import hermes_cli.config as config_mod
+
+    if not hasattr(config_mod, "save_env_value_secure"):
+        config_mod.save_env_value_secure = lambda key, value: {
+            "success": True,
+            "stored_as": key,
+            "validated": False,
+        }
+
+    import cli as cli_mod
+
+    return cli_mod
+
+
+class TestParseServiceTierConfig(unittest.TestCase):
+    def _parse(self, raw):
+        cli_mod = _import_cli()
+        return cli_mod._parse_service_tier_config(raw)
+
+    def test_fast_maps_to_priority(self):
+        self.assertEqual(self._parse("fast"), "priority")
+        self.assertEqual(self._parse("priority"), "priority")
+
+    def test_normal_disables_service_tier(self):
+        self.assertIsNone(self._parse("normal"))
+        self.assertIsNone(self._parse("off"))
+        self.assertIsNone(self._parse(""))
+
+
+class TestHandleFastCommand(unittest.TestCase):
+    def _make_cli(self, service_tier=None):
+        return SimpleNamespace(
+            service_tier=service_tier,
+            provider="openai-codex",
+            requested_provider="openai-codex",
+            model="gpt-5.4",
+            _fast_command_available=lambda: True,
+            agent=MagicMock(),
+        )
+
+    def test_no_args_shows_status(self):
+        cli_mod = _import_cli()
+        stub = self._make_cli(service_tier=None)
+        with (
+            patch.object(cli_mod, "_cprint") as mock_cprint,
+            patch.object(cli_mod, "save_config_value") as mock_save,
+        ):
+            cli_mod.HermesCLI._handle_fast_command(stub, "/fast")
+
+        # Bare /fast shows status, does not change config
+        mock_save.assert_not_called()
+        # Should have printed the status line
+        printed = " ".join(str(c) for c in mock_cprint.call_args_list)
+        self.assertIn("normal", printed)
+
+    def test_no_args_shows_fast_when_enabled(self):
+        cli_mod = _import_cli()
+        stub = self._make_cli(service_tier="priority")
+        with (
+            patch.object(cli_mod, "_cprint") as mock_cprint,
+            patch.object(cli_mod, "save_config_value") as mock_save,
+        ):
+            cli_mod.HermesCLI._handle_fast_command(stub, "/fast")
+
+        mock_save.assert_not_called()
+        printed = " ".join(str(c) for c in mock_cprint.call_args_list)
+        self.assertIn("fast", printed)
+
+    def test_normal_argument_clears_service_tier(self):
+        cli_mod = _import_cli()
+        stub = self._make_cli(service_tier="priority")
+        with (
+            patch.object(cli_mod, "_cprint"),
+            patch.object(cli_mod, "save_config_value", return_value=True) as mock_save,
+        ):
+            cli_mod.HermesCLI._handle_fast_command(stub, "/fast normal")
+
+        mock_save.assert_called_once_with("agent.service_tier", "normal")
+        self.assertIsNone(stub.service_tier)
+        self.assertIsNone(stub.agent)
+
+    def test_unsupported_model_does_not_expose_fast(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            service_tier=None,
+            provider="openai-codex",
+            requested_provider="openai-codex",
+            model="gpt-5.3-codex",
+            _fast_command_available=lambda: False,
+            agent=MagicMock(),
+        )
+
+        with (
+            patch.object(cli_mod, "_cprint") as mock_cprint,
+            patch.object(cli_mod, "save_config_value") as mock_save,
+        ):
+            cli_mod.HermesCLI._handle_fast_command(stub, "/fast")
+
+        mock_save.assert_not_called()
+        self.assertTrue(mock_cprint.called)
+
+
+class TestPriorityProcessingModels(unittest.TestCase):
+    """Verify the expanded Priority Processing model registry."""
+
+    def test_all_documented_models_supported(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        # All models from OpenAI's Priority Processing pricing table
+        supported = [
+            "gpt-5.4", "gpt-5.4-mini", "gpt-5.2",
+            "gpt-5.1", "gpt-5", "gpt-5-mini",
+            "gpt-4.1", "gpt-4.1-mini", "gpt-4.1-nano",
+            "gpt-4o", "gpt-4o-mini",
+            "o3", "o4-mini",
+        ]
+        for model in supported:
+            assert model_supports_fast_mode(model), f"{model} should support fast mode"
+
+    def test_vendor_prefix_stripped(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        assert model_supports_fast_mode("openai/gpt-5.4") is True
+        assert model_supports_fast_mode("openai/gpt-4.1") is True
+        assert model_supports_fast_mode("openai/o3") is True
+
+    def test_non_priority_models_rejected(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        assert model_supports_fast_mode("gpt-5.3-codex") is False
+        assert model_supports_fast_mode("claude-sonnet-4") is False
+        assert model_supports_fast_mode("") is False
+        assert model_supports_fast_mode(None) is False
+
+    def test_resolve_overrides_returns_service_tier(self):
+        from hermes_cli.models import resolve_fast_mode_overrides
+
+        result = resolve_fast_mode_overrides("gpt-5.4")
+        assert result == {"service_tier": "priority"}
+
+        result = resolve_fast_mode_overrides("gpt-4.1")
+        assert result == {"service_tier": "priority"}
+
+    def test_resolve_overrides_none_for_unsupported(self):
+        from hermes_cli.models import resolve_fast_mode_overrides
+
+        assert resolve_fast_mode_overrides("gpt-5.3-codex") is None
+        assert resolve_fast_mode_overrides("claude-sonnet-4") is None
+
+
+class TestFastModeRouting(unittest.TestCase):
+    def test_fast_command_exposed_for_model_even_when_provider_is_auto(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(provider="auto", requested_provider="auto", model="gpt-5.4", agent=None)
+
+        assert cli_mod.HermesCLI._fast_command_available(stub) is True
+
+    def test_fast_command_exposed_for_non_codex_models(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(provider="openai", requested_provider="openai", model="gpt-4.1", agent=None)
+        assert cli_mod.HermesCLI._fast_command_available(stub) is True
+
+        stub = SimpleNamespace(provider="openrouter", requested_provider="openrouter", model="o3", agent=None)
+        assert cli_mod.HermesCLI._fast_command_available(stub) is True
+
+    def test_turn_route_injects_overrides_without_provider_switch(self):
+        """Fast mode should add request_overrides but NOT change the provider/runtime."""
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            model="gpt-5.4",
+            api_key="primary-key",
+            base_url="https://openrouter.ai/api/v1",
+            provider="openrouter",
+            api_mode="chat_completions",
+            acp_command=None,
+            acp_args=[],
+            _credential_pool=None,
+            _smart_model_routing={},
+            service_tier="priority",
+        )
+
+        original_runtime = {
+            "api_key": "***",
+            "base_url": "https://openrouter.ai/api/v1",
+            "provider": "openrouter",
+            "api_mode": "chat_completions",
+            "command": None,
+            "args": [],
+            "credential_pool": None,
+        }
+
+        with patch("agent.smart_model_routing.resolve_turn_route", return_value={
+            "model": "gpt-5.4",
+            "runtime": dict(original_runtime),
+            "label": None,
+            "signature": ("gpt-5.4", "openrouter", "https://openrouter.ai/api/v1", "chat_completions", None, ()),
+        }):
+            route = cli_mod.HermesCLI._resolve_turn_agent_config(stub, "hi")
+
+        # Provider should NOT have changed
+        assert route["runtime"]["provider"] == "openrouter"
+        assert route["runtime"]["api_mode"] == "chat_completions"
+        # But request_overrides should be set
+        assert route["request_overrides"] == {"service_tier": "priority"}
+
+    def test_turn_route_keeps_primary_runtime_when_model_has_no_fast_backend(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            model="gpt-5.3-codex",
+            api_key="primary-key",
+            base_url="https://openrouter.ai/api/v1",
+            provider="openrouter",
+            api_mode="chat_completions",
+            acp_command=None,
+            acp_args=[],
+            _credential_pool=None,
+            _smart_model_routing={},
+            service_tier="priority",
+        )
+
+        primary_route = {
+            "model": "gpt-5.3-codex",
+            "runtime": {
+                "api_key": "***",
+                "base_url": "https://openrouter.ai/api/v1",
+                "provider": "openrouter",
+                "api_mode": "chat_completions",
+                "command": None,
+                "args": [],
+                "credential_pool": None,
+            },
+            "label": None,
+            "signature": ("gpt-5.3-codex", "openrouter", "https://openrouter.ai/api/v1", "chat_completions", None, ()),
+        }
+        with patch("agent.smart_model_routing.resolve_turn_route", return_value=primary_route):
+            route = cli_mod.HermesCLI._resolve_turn_agent_config(stub, "hi")
+
+        assert route["runtime"]["provider"] == "openrouter"
+        assert route.get("request_overrides") is None
+
+
+class TestAnthropicFastMode(unittest.TestCase):
+    """Verify Anthropic Fast Mode model support and override resolution."""
+
+    def test_anthropic_opus_supported(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        # Native Anthropic format (hyphens)
+        assert model_supports_fast_mode("claude-opus-4-6") is True
+        # OpenRouter format (dots)
+        assert model_supports_fast_mode("claude-opus-4.6") is True
+        # With vendor prefix
+        assert model_supports_fast_mode("anthropic/claude-opus-4-6") is True
+        assert model_supports_fast_mode("anthropic/claude-opus-4.6") is True
+
+    def test_anthropic_non_opus_rejected(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        assert model_supports_fast_mode("claude-sonnet-4-6") is False
+        assert model_supports_fast_mode("claude-sonnet-4.6") is False
+        assert model_supports_fast_mode("claude-haiku-4-5") is False
+        assert model_supports_fast_mode("anthropic/claude-sonnet-4.6") is False
+
+    def test_anthropic_variant_tags_stripped(self):
+        from hermes_cli.models import model_supports_fast_mode
+
+        # OpenRouter variant tags after colon should be stripped
+        assert model_supports_fast_mode("claude-opus-4.6:fast") is True
+        assert model_supports_fast_mode("claude-opus-4.6:beta") is True
+
+    def test_resolve_overrides_returns_speed_for_anthropic(self):
+        from hermes_cli.models import resolve_fast_mode_overrides
+
+        result = resolve_fast_mode_overrides("claude-opus-4-6")
+        assert result == {"speed": "fast"}
+
+        result = resolve_fast_mode_overrides("anthropic/claude-opus-4.6")
+        assert result == {"speed": "fast"}
+
+    def test_resolve_overrides_returns_service_tier_for_openai(self):
+        """OpenAI models should still get service_tier, not speed."""
+        from hermes_cli.models import resolve_fast_mode_overrides
+
+        result = resolve_fast_mode_overrides("gpt-5.4")
+        assert result == {"service_tier": "priority"}
+
+    def test_is_anthropic_fast_model(self):
+        from hermes_cli.models import _is_anthropic_fast_model
+
+        assert _is_anthropic_fast_model("claude-opus-4-6") is True
+        assert _is_anthropic_fast_model("claude-opus-4.6") is True
+        assert _is_anthropic_fast_model("anthropic/claude-opus-4-6") is True
+        assert _is_anthropic_fast_model("gpt-5.4") is False
+        assert _is_anthropic_fast_model("claude-sonnet-4-6") is False
+
+    def test_fast_command_exposed_for_anthropic_model(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            provider="anthropic", requested_provider="anthropic",
+            model="claude-opus-4-6", agent=None,
+        )
+        assert cli_mod.HermesCLI._fast_command_available(stub) is True
+
+    def test_fast_command_hidden_for_anthropic_sonnet(self):
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            provider="anthropic", requested_provider="anthropic",
+            model="claude-sonnet-4-6", agent=None,
+        )
+        assert cli_mod.HermesCLI._fast_command_available(stub) is False
+
+    def test_turn_route_injects_speed_for_anthropic(self):
+        """Anthropic models should get speed:'fast' override, not service_tier."""
+        cli_mod = _import_cli()
+        stub = SimpleNamespace(
+            model="claude-opus-4-6",
+            api_key="sk-ant-test",
+            base_url="https://api.anthropic.com",
+            provider="anthropic",
+            api_mode="anthropic_messages",
+            acp_command=None,
+            acp_args=[],
+            _credential_pool=None,
+            _smart_model_routing={},
+            service_tier="priority",
+        )
+
+        original_runtime = {
+            "api_key": "***",
+            "base_url": "https://api.anthropic.com",
+            "provider": "anthropic",
+            "api_mode": "anthropic_messages",
+            "command": None,
+            "args": [],
+            "credential_pool": None,
+        }
+
+        with patch("agent.smart_model_routing.resolve_turn_route", return_value={
+            "model": "claude-opus-4-6",
+            "runtime": dict(original_runtime),
+            "label": None,
+            "signature": ("claude-opus-4-6", "anthropic", "https://api.anthropic.com", "anthropic_messages", None, ()),
+        }):
+            route = cli_mod.HermesCLI._resolve_turn_agent_config(stub, "hi")
+
+        assert route["runtime"]["provider"] == "anthropic"
+        assert route["request_overrides"] == {"speed": "fast"}
+
+
+class TestAnthropicFastModeAdapter(unittest.TestCase):
+    """Verify build_anthropic_kwargs handles fast_mode parameter."""
+
+    def test_fast_mode_adds_speed_and_beta(self):
+        from agent.anthropic_adapter import build_anthropic_kwargs, _FAST_MODE_BETA
+
+        kwargs = build_anthropic_kwargs(
+            model="claude-opus-4-6",
+            messages=[{"role": "user", "content": [{"type": "text", "text": "hi"}]}],
+            tools=None,
+            max_tokens=None,
+            reasoning_config=None,
+            fast_mode=True,
+        )
+        assert kwargs.get("speed") == "fast"
+        assert "extra_headers" in kwargs
+        assert _FAST_MODE_BETA in kwargs["extra_headers"].get("anthropic-beta", "")
+
+    def test_fast_mode_off_no_speed(self):
+        from agent.anthropic_adapter import build_anthropic_kwargs
+
+        kwargs = build_anthropic_kwargs(
+            model="claude-opus-4-6",
+            messages=[{"role": "user", "content": [{"type": "text", "text": "hi"}]}],
+            tools=None,
+            max_tokens=None,
+            reasoning_config=None,
+            fast_mode=False,
+        )
+        assert "speed" not in kwargs
+        assert "extra_headers" not in kwargs
+
+    def test_fast_mode_skipped_for_third_party_endpoint(self):
+        from agent.anthropic_adapter import build_anthropic_kwargs
+
+        kwargs = build_anthropic_kwargs(
+            model="claude-opus-4-6",
+            messages=[{"role": "user", "content": [{"type": "text", "text": "hi"}]}],
+            tools=None,
+            max_tokens=None,
+            reasoning_config=None,
+            fast_mode=True,
+            base_url="https://api.minimax.io/anthropic/v1",
+        )
+        # Third-party endpoints should NOT get speed or fast-mode beta
+        assert "speed" not in kwargs
+        assert "extra_headers" not in kwargs
+
+
+class TestConfigDefault(unittest.TestCase):
+    def test_default_config_has_service_tier(self):
+        from hermes_cli.config import DEFAULT_CONFIG
+
+        agent = DEFAULT_CONFIG.get("agent", {})
+        self.assertIn("service_tier", agent)
+        self.assertEqual(agent["service_tier"], "")

tests/cli/test_stream_delta_think_tag.py

@@ -0,0 +1,138 @@
+"""Tests for _stream_delta's handling of <think> tags in prose vs real reasoning blocks."""
+import sys
+import os
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", ".."))
+
+import pytest
+
+
+def _make_cli_stub():
+    """Create a minimal HermesCLI-like object with stream state."""
+    from cli import HermesCLI
+
+    cli = HermesCLI.__new__(HermesCLI)
+    cli.show_reasoning = False
+    cli._stream_buf = ""
+    cli._stream_started = False
+    cli._stream_box_opened = False
+    cli._stream_prefilt = ""
+    cli._in_reasoning_block = False
+    cli._reasoning_stream_started = False
+    cli._reasoning_box_opened = False
+    cli._reasoning_buf = ""
+    cli._reasoning_preview_buf = ""
+    cli._deferred_content = ""
+    cli._stream_text_ansi = ""
+    cli._stream_needs_break = False
+    cli._emitted = []
+
+    # Mock _emit_stream_text to capture output
+    def mock_emit(text):
+        cli._emitted.append(text)
+    cli._emit_stream_text = mock_emit
+
+    # Mock _stream_reasoning_delta
+    cli._reasoning_emitted = []
+    def mock_reasoning(text):
+        cli._reasoning_emitted.append(text)
+    cli._stream_reasoning_delta = mock_reasoning
+
+    return cli
+
+
+class TestThinkTagInProse:
+    """<think> mentioned in prose should NOT trigger reasoning suppression."""
+
+    def test_think_tag_mid_sentence(self):
+        """'(/think not producing <think> tags)' should pass through."""
+        cli = _make_cli_stub()
+        tokens = [
+            "  1. Fix reasoning mode in eval ",
+            "(/think not producing ",
+            "<think>",
+            " tags — ~2% gap)",
+            "\n  2. Launch production",
+        ]
+        for t in tokens:
+            cli._stream_delta(t)
+        assert not cli._in_reasoning_block, "<think> in prose should not enter reasoning block"
+        full = "".join(cli._emitted)
+        assert "<think>" in full, "The literal <think> tag should be in the emitted text"
+        assert "Launch production" in full
+
+    def test_think_tag_after_text_on_same_line(self):
+        """'some text <think>' should NOT trigger reasoning."""
+        cli = _make_cli_stub()
+        cli._stream_delta("Here is the <think> tag explanation")
+        assert not cli._in_reasoning_block
+        full = "".join(cli._emitted)
+        assert "<think>" in full
+
+    def test_think_tag_in_backticks(self):
+        """'`<think>`' should NOT trigger reasoning."""
+        cli = _make_cli_stub()
+        cli._stream_delta("Use the `<think>` tag for reasoning")
+        assert not cli._in_reasoning_block
+
+
+class TestRealReasoningBlock:
+    """Real <think> tags at block boundaries should still be caught."""
+
+    def test_think_at_start_of_stream(self):
+        """'<think>reasoning</think>answer' should suppress reasoning."""
+        cli = _make_cli_stub()
+        cli._stream_delta("<think>")
+        assert cli._in_reasoning_block
+        cli._stream_delta("I need to analyze this")
+        cli._stream_delta("</think>")
+        assert not cli._in_reasoning_block
+        cli._stream_delta("Here is my answer")
+        full = "".join(cli._emitted)
+        assert "Here is my answer" in full
+        assert "I need to analyze" not in full  # reasoning was suppressed
+
+    def test_think_after_newline(self):
+        """'text\\n<think>' should trigger reasoning block."""
+        cli = _make_cli_stub()
+        cli._stream_delta("Some preamble\n<think>")
+        assert cli._in_reasoning_block
+        full = "".join(cli._emitted)
+        assert "Some preamble" in full
+
+    def test_think_after_newline_with_whitespace(self):
+        """'text\\n  <think>' should trigger reasoning block."""
+        cli = _make_cli_stub()
+        cli._stream_delta("Some preamble\n  <think>")
+        assert cli._in_reasoning_block
+
+    def test_think_with_only_whitespace_before(self):
+        """'   <think>' (whitespace only prefix) should trigger."""
+        cli = _make_cli_stub()
+        cli._stream_delta("   <think>")
+        assert cli._in_reasoning_block
+
+
+class TestFlushRecovery:
+    """_flush_stream should recover content from false-positive reasoning blocks."""
+
+    def test_flush_recovers_buffered_content(self):
+        """If somehow in reasoning block at flush, content is recovered."""
+        cli = _make_cli_stub()
+        # Manually set up a false-positive state
+        cli._in_reasoning_block = True
+        cli._stream_prefilt = " tags — ~2% gap)\n  2. Launch production"
+        cli._stream_box_opened = True
+
+        # Mock _close_reasoning_box and box closing
+        cli._close_reasoning_box = lambda: None
+
+        # Call flush
+        from unittest.mock import patch
+        import shutil
+        with patch.object(shutil, "get_terminal_size", return_value=os.terminal_size((80, 24))):
+            with patch("cli._cprint"):
+                cli._flush_stream()
+
+        assert not cli._in_reasoning_block
+        full = "".join(cli._emitted)
+        assert "Launch production" in full

tests/cron/test_scheduler.py

@@ -173,6 +173,40 @@ class TestResolveDeliveryTarget:
             "thread_id": None,
         }
 
+    def test_explicit_discord_topic_target_with_thread_id(self):
+        """deliver: 'discord:chat_id:thread_id' parses correctly."""
+        job = {
+            "deliver": "discord:-1001234567890:17585",
+        }
+        assert _resolve_delivery_target(job) == {
+            "platform": "discord",
+            "chat_id": "-1001234567890",
+            "thread_id": "17585",
+        }
+
+    def test_explicit_discord_chat_id_without_thread_id(self):
+        """deliver: 'discord:chat_id' sets thread_id to None."""
+        job = {
+            "deliver": "discord:9876543210",
+        }
+        assert _resolve_delivery_target(job) == {
+            "platform": "discord",
+            "chat_id": "9876543210",
+            "thread_id": None,
+        }
+
+    def test_explicit_discord_channel_without_thread(self):
+        """deliver: 'discord:1001234567890' resolves via explicit platform:chat_id path."""
+        job = {
+            "deliver": "discord:1001234567890",
+        }
+        result = _resolve_delivery_target(job)
+        assert result == {
+            "platform": "discord",
+            "chat_id": "1001234567890",
+            "thread_id": None,
+        }
+
 
 class TestDeliverResultWrapping:
     """Verify that cron deliveries are wrapped with header/footer and no longer mirrored."""

tests/e2e/conftest.py

@@ -1,4 +1,4 @@
-"""Shared fixtures for Telegram gateway e2e tests.
+"""Shared fixtures for gateway e2e tests (Telegram, Discord).
 
 These tests exercise the full async message flow:
     adapter.handle_message(event)
@@ -14,19 +14,22 @@ import sys
 import uuid
 from datetime import datetime
 from types import SimpleNamespace
-from unittest.mock import AsyncMock, MagicMock
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
 
 from gateway.config import GatewayConfig, Platform, PlatformConfig
 from gateway.platforms.base import MessageEvent, SendResult
 from gateway.session import SessionEntry, SessionSource, build_session_key
 
 
-#Ensure telegram module is available (mock it if not installed)
+# Platform library mocks
 
+# Ensure telegram module is available (mock it if not installed)
 def _ensure_telegram_mock():
     """Install mock telegram modules so TelegramAdapter can be imported."""
     if "telegram" in sys.modules and hasattr(sys.modules["telegram"], "__file__"):
-        return  # Real library installed
+        return # Real library installed
 
     telegram_mod = MagicMock()
     telegram_mod.Update = MagicMock()
@@ -51,24 +54,118 @@ def _ensure_telegram_mock():
         sys.modules.setdefault(name, telegram_mod)
 
 
-_ensure_telegram_mock()
+# Ensure discord module is available (mock it if not installed)
+def _ensure_discord_mock():
+    """Install mock discord modules so DiscordAdapter can be imported."""
+    if "discord" in sys.modules and hasattr(sys.modules["discord"], "__file__"):
+        return # Real library installed
 
+    discord_mod = MagicMock()
+    discord_mod.Intents.default.return_value = MagicMock()
+    discord_mod.DMChannel = type("DMChannel", (), {})
+    discord_mod.Thread = type("Thread", (), {})
+    discord_mod.ForumChannel = type("ForumChannel", (), {})
+    discord_mod.Interaction = object
+    discord_mod.app_commands = SimpleNamespace(
+        describe=lambda **kwargs: (lambda fn: fn),
+        choices=lambda **kwargs: (lambda fn: fn),
+        Choice=lambda **kwargs: SimpleNamespace(**kwargs),
+    )
+    discord_mod.opus.is_loaded.return_value = True
+
+    ext_mod = MagicMock()
+    commands_mod = MagicMock()
+    commands_mod.Bot = MagicMock
+    ext_mod.commands = commands_mod
+
+    sys.modules.setdefault("discord", discord_mod)
+    sys.modules.setdefault("discord.ext", ext_mod)
+    sys.modules.setdefault("discord.ext.commands", commands_mod)
+    sys.modules.setdefault("discord.opus", discord_mod.opus)
+
+
+def _ensure_slack_mock():
+    """Install mock slack modules so SlackAdapter can be imported."""
+    if "slack_bolt" in sys.modules and hasattr(sys.modules["slack_bolt"], "__file__"):
+        return  # Real library installed
+
+    slack_bolt = MagicMock()
+    slack_bolt.async_app.AsyncApp = MagicMock
+    slack_bolt.adapter.socket_mode.async_handler.AsyncSocketModeHandler = MagicMock
+
+    slack_sdk = MagicMock()
+    slack_sdk.web.async_client.AsyncWebClient = MagicMock
+
+    for name, mod in [
+        ("slack_bolt", slack_bolt),
+        ("slack_bolt.async_app", slack_bolt.async_app),
+        ("slack_bolt.adapter", slack_bolt.adapter),
+        ("slack_bolt.adapter.socket_mode", slack_bolt.adapter.socket_mode),
+        ("slack_bolt.adapter.socket_mode.async_handler", slack_bolt.adapter.socket_mode.async_handler),
+        ("slack_sdk", slack_sdk),
+        ("slack_sdk.web", slack_sdk.web),
+        ("slack_sdk.web.async_client", slack_sdk.web.async_client),
+    ]:
+        sys.modules.setdefault(name, mod)
+
+
+_ensure_telegram_mock()
+_ensure_discord_mock()
+_ensure_slack_mock()
+
+from gateway.platforms.discord import DiscordAdapter   # noqa: E402
 from gateway.platforms.telegram import TelegramAdapter  # noqa: E402
 
+import gateway.platforms.slack as _slack_mod  # noqa: E402
+_slack_mod.SLACK_AVAILABLE = True
+from gateway.platforms.slack import SlackAdapter  # noqa: E402
 
-#GatewayRunner factory (based on tests/gateway/test_status_command.py)
 
-def make_runner(session_entry: SessionEntry) -> "GatewayRunner":
+# Platform-generic factories
+
+def make_source(platform: Platform, chat_id: str = "e2e-chat-1", user_id: str = "e2e-user-1") -> SessionSource:
+    return SessionSource(
+        platform=platform,
+        chat_id=chat_id,
+        user_id=user_id,
+        user_name="e2e_tester",
+        chat_type="dm",
+    )
+
+
+def make_session_entry(platform: Platform, source: SessionSource = None) -> SessionEntry:
+    source = source or make_source(platform)
+    return SessionEntry(
+        session_key=build_session_key(source),
+        session_id=f"sess-{uuid.uuid4().hex[:8]}",
+        created_at=datetime.now(),
+        updated_at=datetime.now(),
+        platform=platform,
+        chat_type="dm",
+    )
+
+
+def make_event(platform: Platform, text: str = "/help", chat_id: str = "e2e-chat-1", user_id: str = "e2e-user-1") -> MessageEvent:
+    return MessageEvent(
+        text=text,
+        source=make_source(platform, chat_id, user_id),
+        message_id=f"msg-{uuid.uuid4().hex[:8]}",
+    )
+
+
+def make_runner(platform: Platform, session_entry: SessionEntry = None) -> "GatewayRunner":
     """Create a GatewayRunner with mocked internals for e2e testing.
 
     Skips __init__ to avoid filesystem/network side effects.
-    All command-dispatch dependencies are wired manually.
     """
     from gateway.run import GatewayRunner
 
+    if session_entry is None:
+        session_entry = make_session_entry(platform)
+
     runner = object.__new__(GatewayRunner)
     runner.config = GatewayConfig(
-        platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="e2e-test-token")}
+        platforms={platform: PlatformConfig(enabled=True, token="e2e-test-token")}
     )
     runner.adapters = {}
     runner._voice_mode = {}
@@ -99,7 +196,6 @@ def make_runner(session_entry: SessionEntry) -> "GatewayRunner":
     runner._capture_gateway_honcho_if_configured = lambda *a, **kw: None
     runner._emit_gateway_run_progress = AsyncMock()
 
-    # Pairing store (used by authorization rejection path)
     runner.pairing_store = MagicMock()
     runner.pairing_store._is_rate_limited = MagicMock(return_value=False)
     runner.pairing_store.generate_code = MagicMock(return_value="ABC123")
@@ -107,67 +203,63 @@ def make_runner(session_entry: SessionEntry) -> "GatewayRunner":
     return runner
 
 
-#TelegramAdapter factory
+def make_adapter(platform: Platform, runner=None):
+    """Create a platform adapter wired to *runner*, with send methods mocked."""
+    if runner is None:
+        runner = make_runner(platform)
 
-def make_adapter(runner) -> TelegramAdapter:
-    """Create a TelegramAdapter wired to *runner*, with send methods mocked.
-
-    connect() is NOT called — no polling, no token lock, no real HTTP.
-    """
     config = PlatformConfig(enabled=True, token="e2e-test-token")
-    adapter = TelegramAdapter(config)
 
-    # Mock outbound methods so tests can capture what was sent
+    if platform == Platform.DISCORD:
+        with patch.object(DiscordAdapter, "_load_participated_threads", return_value=set()):
+            adapter = DiscordAdapter(config)
+        platform_key = Platform.DISCORD
+    elif platform == Platform.SLACK:
+        adapter = SlackAdapter(config)
+        platform_key = Platform.SLACK
+    else:
+        adapter = TelegramAdapter(config)
+        platform_key = Platform.TELEGRAM
+
     adapter.send = AsyncMock(return_value=SendResult(success=True, message_id="e2e-resp-1"))
     adapter.send_typing = AsyncMock()
 
-    # Wire adapter ↔ runner
     adapter.set_message_handler(runner._handle_message)
-    runner.adapters[Platform.TELEGRAM] = adapter
+    runner.adapters[platform_key] = adapter
 
     return adapter
 
 
-#Helpers
-
-def make_source(chat_id: str = "e2e-chat-1", user_id: str = "e2e-user-1") -> SessionSource:
-    return SessionSource(
-        platform=Platform.TELEGRAM,
-        chat_id=chat_id,
-        user_id=user_id,
-        user_name="e2e_tester",
-        chat_type="dm",
-    )
-
-
-def make_event(text: str, chat_id: str = "e2e-chat-1", user_id: str = "e2e-user-1") -> MessageEvent:
-    return MessageEvent(
-        text=text,
-        source=make_source(chat_id, user_id),
-        message_id=f"msg-{uuid.uuid4().hex[:8]}",
-    )
-
-
-def make_session_entry(source: SessionSource = None) -> SessionEntry:
-    source = source or make_source()
-    return SessionEntry(
-        session_key=build_session_key(source),
-        session_id=f"sess-{uuid.uuid4().hex[:8]}",
-        created_at=datetime.now(),
-        updated_at=datetime.now(),
-        platform=Platform.TELEGRAM,
-        chat_type="dm",
-    )
-
-
-async def send_and_capture(adapter: TelegramAdapter, text: str, **event_kwargs) -> AsyncMock:
-    """Send a message through the full e2e flow and return the send mock.
-
-    Drives: adapter.handle_message → background task → runner dispatch → adapter.send.
-    """
-    event = make_event(text, **event_kwargs)
+async def send_and_capture(adapter, text: str, platform: Platform, **event_kwargs) -> AsyncMock:
+    """Send a message through the full e2e flow and return the send mock."""
+    event = make_event(platform, text, **event_kwargs)
     adapter.send.reset_mock()
     await adapter.handle_message(event)
-    # Let the background task complete
     await asyncio.sleep(0.3)
     return adapter.send
+
+
+# Parametrized fixtures for platform-generic tests
+@pytest.fixture(params=[Platform.TELEGRAM, Platform.DISCORD, Platform.SLACK], ids=["telegram", "discord", "slack"])
+def platform(request):
+    return request.param
+
+
+@pytest.fixture()
+def source(platform):
+    return make_source(platform)
+
+
+@pytest.fixture()
+def session_entry(platform, source):
+    return make_session_entry(platform, source)
+
+
+@pytest.fixture()
+def runner(platform, session_entry):
+    return make_runner(platform, session_entry)
+
+
+@pytest.fixture()
+def adapter(platform, runner):
+    return make_adapter(platform, runner)

tests/e2e/test_platform_commands.py

@@ -1,4 +1,4 @@
-"""E2E tests for Telegram gateway slash commands.
+"""E2E tests for gateway slash commands (Telegram, Discord).
 
 Each test drives a message through the full async pipeline:
     adapter.handle_message(event)
@@ -7,6 +7,7 @@ Each test drives a message through the full async pipeline:
         → adapter.send() (captured for assertions)
 
 No LLM involved — only gateway-level commands are tested.
+Tests are parametrized over platforms via the ``platform`` fixture in conftest.
 """
 
 import asyncio
@@ -15,46 +16,15 @@ from unittest.mock import AsyncMock
 import pytest
 
 from gateway.platforms.base import SendResult
-from tests.e2e.conftest import (
-    make_adapter,
-    make_event,
-    make_runner,
-    make_session_entry,
-    make_source,
-    send_and_capture,
-)
+from tests.e2e.conftest import make_event, send_and_capture
 
 
-#Fixtures
-
-@pytest.fixture()
-def source():
-    return make_source()
-
-
-@pytest.fixture()
-def session_entry(source):
-    return make_session_entry(source)
-
-
-@pytest.fixture()
-def runner(session_entry):
-    return make_runner(session_entry)
-
-
-@pytest.fixture()
-def adapter(runner):
-    return make_adapter(runner)
-
-
-#Tests
-
-class TestTelegramSlashCommands:
+class TestSlashCommands:
     """Gateway slash commands dispatched through the full adapter pipeline."""
 
     @pytest.mark.asyncio
-    async def test_help_returns_command_list(self, adapter):
-        send = await send_and_capture(adapter, "/help")
+    async def test_help_returns_command_list(self, adapter, platform):
+        send = await send_and_capture(adapter, "/help", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
@@ -62,24 +32,23 @@ class TestTelegramSlashCommands:
         assert "/status" in response_text
 
     @pytest.mark.asyncio
-    async def test_status_shows_session_info(self, adapter):
-        send = await send_and_capture(adapter, "/status")
+    async def test_status_shows_session_info(self, adapter, platform):
+        send = await send_and_capture(adapter, "/status", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
-        # Status output includes session metadata
         assert "session" in response_text.lower() or "Session" in response_text
 
     @pytest.mark.asyncio
-    async def test_new_resets_session(self, adapter, runner):
-        send = await send_and_capture(adapter, "/new")
+    async def test_new_resets_session(self, adapter, runner, platform):
+        send = await send_and_capture(adapter, "/new", platform)
 
         send.assert_called_once()
         runner.session_store.reset_session.assert_called_once()
 
     @pytest.mark.asyncio
-    async def test_stop_when_no_agent_running(self, adapter):
-        send = await send_and_capture(adapter, "/stop")
+    async def test_stop_when_no_agent_running(self, adapter, platform):
+        send = await send_and_capture(adapter, "/stop", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
@@ -87,8 +56,8 @@ class TestTelegramSlashCommands:
         assert "no" in response_lower or "stop" in response_lower or "not running" in response_lower
 
     @pytest.mark.asyncio
-    async def test_commands_shows_listing(self, adapter):
-        send = await send_and_capture(adapter, "/commands")
+    async def test_commands_shows_listing(self, adapter, platform):
+        send = await send_and_capture(adapter, "/commands", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
@@ -96,29 +65,25 @@ class TestTelegramSlashCommands:
         assert "/" in response_text
 
     @pytest.mark.asyncio
-    async def test_sequential_commands_share_session(self, adapter):
+    async def test_sequential_commands_share_session(self, adapter, platform):
         """Two commands from the same chat_id should both succeed."""
-        send_help = await send_and_capture(adapter, "/help")
+        send_help = await send_and_capture(adapter, "/help", platform)
         send_help.assert_called_once()
 
-        send_status = await send_and_capture(adapter, "/status")
+        send_status = await send_and_capture(adapter, "/status", platform)
         send_status.assert_called_once()
 
     @pytest.mark.asyncio
-    @pytest.mark.xfail(
-        reason="Bug: _handle_provider_command references unbound model_cfg when config.yaml is absent",
-        strict=False,
-    )
-    async def test_provider_shows_current_provider(self, adapter):
-        send = await send_and_capture(adapter, "/provider")
+    async def test_provider_shows_current_provider(self, adapter, platform):
+        send = await send_and_capture(adapter, "/provider", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
         assert "provider" in response_text.lower()
 
     @pytest.mark.asyncio
-    async def test_verbose_responds(self, adapter):
-        send = await send_and_capture(adapter, "/verbose")
+    async def test_verbose_responds(self, adapter, platform):
+        send = await send_and_capture(adapter, "/verbose", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
@@ -126,42 +91,50 @@ class TestTelegramSlashCommands:
         assert "verbose" in response_text.lower() or "tool_progress" in response_text
 
     @pytest.mark.asyncio
-    async def test_personality_lists_options(self, adapter):
-        send = await send_and_capture(adapter, "/personality")
+    async def test_personality_lists_options(self, adapter, platform):
+        send = await send_and_capture(adapter, "/personality", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
         assert "personalit" in response_text.lower()  # matches "personality" or "personalities"
 
     @pytest.mark.asyncio
-    async def test_yolo_toggles_mode(self, adapter):
-        send = await send_and_capture(adapter, "/yolo")
+    async def test_yolo_toggles_mode(self, adapter, platform):
+        send = await send_and_capture(adapter, "/yolo", platform)
 
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
         assert "yolo" in response_text.lower()
 
+    @pytest.mark.asyncio
+    async def test_compress_command(self, adapter, platform):
+        send = await send_and_capture(adapter, "/compress", platform)
+
+        send.assert_called_once()
+        response_text = send.call_args[1].get("content") or send.call_args[0][1]
+        assert "compress" in response_text.lower() or "context" in response_text.lower()
+
 
 class TestSessionLifecycle:
     """Verify session state changes across command sequences."""
 
     @pytest.mark.asyncio
-    async def test_new_then_status_reflects_reset(self, adapter, runner, session_entry):
+    async def test_new_then_status_reflects_reset(self, adapter, runner, session_entry, platform):
         """After /new, /status should report the fresh session."""
-        await send_and_capture(adapter, "/new")
+        await send_and_capture(adapter, "/new", platform)
         runner.session_store.reset_session.assert_called_once()
 
-        send = await send_and_capture(adapter, "/status")
+        send = await send_and_capture(adapter, "/status", platform)
         send.assert_called_once()
         response_text = send.call_args[1].get("content") or send.call_args[0][1]
         # Session ID from the entry should appear in the status output
         assert session_entry.session_id[:8] in response_text
 
     @pytest.mark.asyncio
-    async def test_new_is_idempotent(self, adapter, runner):
+    async def test_new_is_idempotent(self, adapter, runner, platform):
         """/new called twice should not crash."""
-        await send_and_capture(adapter, "/new")
-        await send_and_capture(adapter, "/new")
+        await send_and_capture(adapter, "/new", platform)
+        await send_and_capture(adapter, "/new", platform)
         assert runner.session_store.reset_session.call_count == 2
 
 
@@ -169,11 +142,11 @@ class TestAuthorization:
     """Verify the pipeline handles unauthorized users."""
 
     @pytest.mark.asyncio
-    async def test_unauthorized_user_gets_pairing_response(self, adapter, runner):
+    async def test_unauthorized_user_gets_pairing_response(self, adapter, runner, platform):
         """Unauthorized DM should trigger pairing code, not a command response."""
         runner._is_user_authorized = lambda _source: False
 
-        event = make_event("/help")
+        event = make_event(platform, "/help")
         adapter.send.reset_mock()
         await adapter.handle_message(event)
         await asyncio.sleep(0.3)
@@ -185,11 +158,11 @@ class TestAuthorization:
         assert "recognize" in response_text.lower() or "pair" in response_text.lower() or "ABC123" in response_text
 
     @pytest.mark.asyncio
-    async def test_unauthorized_user_does_not_get_help(self, adapter, runner):
+    async def test_unauthorized_user_does_not_get_help(self, adapter, runner, platform):
         """Unauthorized user should NOT see the help command output."""
         runner._is_user_authorized = lambda _source: False
 
-        event = make_event("/help")
+        event = make_event(platform, "/help")
         adapter.send.reset_mock()
         await adapter.handle_message(event)
         await asyncio.sleep(0.3)
@@ -204,12 +177,12 @@ class TestSendFailureResilience:
     """Verify the pipeline handles send failures gracefully."""
 
     @pytest.mark.asyncio
-    async def test_send_failure_does_not_crash_pipeline(self, adapter):
+    async def test_send_failure_does_not_crash_pipeline(self, adapter, platform):
         """If send() returns failure, the pipeline should not raise."""
         adapter.send = AsyncMock(return_value=SendResult(success=False, error="network timeout"))
-        adapter.set_message_handler(adapter._message_handler)  # re-wire with same handler
+        adapter.set_message_handler(adapter._message_handler) # re-wire with same handler
 
-        event = make_event("/help")
+        event = make_event(platform, "/help")
         # Should not raise — pipeline handles send failures internally
         await adapter.handle_message(event)
         await asyncio.sleep(0.3)

tests/gateway/test_api_server.py

@@ -26,6 +26,7 @@ from gateway.platforms.api_server import (
     APIServerAdapter,
     ResponseStore,
     _CORS_HEADERS,
+    _derive_chat_session_id,
     check_api_server_requirements,
     cors_middleware,
     security_headers_middleware,
@@ -294,6 +295,40 @@ class TestModelsEndpoint:
             assert data["data"][0]["id"] == "hermes-agent"
             assert data["data"][0]["owned_by"] == "hermes"
 
+    @pytest.mark.asyncio
+    async def test_models_returns_profile_name(self):
+        """When running under a named profile, /v1/models advertises the profile name."""
+        with patch("gateway.platforms.api_server.APIServerAdapter._resolve_model_name", return_value="lucas"):
+            adapter = _make_adapter()
+        app = _create_app(adapter)
+        async with TestClient(TestServer(app)) as cli:
+            resp = await cli.get("/v1/models")
+            assert resp.status == 200
+            data = await resp.json()
+            assert data["data"][0]["id"] == "lucas"
+            assert data["data"][0]["root"] == "lucas"
+
+    @pytest.mark.asyncio
+    async def test_models_returns_explicit_model_name(self):
+        """Explicit model_name in config overrides profile name."""
+        extra = {"model_name": "my-custom-agent"}
+        config = PlatformConfig(enabled=True, extra=extra)
+        adapter = APIServerAdapter(config)
+        assert adapter._model_name == "my-custom-agent"
+
+    def test_resolve_model_name_explicit(self):
+        assert APIServerAdapter._resolve_model_name("my-bot") == "my-bot"
+
+    def test_resolve_model_name_default_profile(self):
+        """Default profile falls back to 'hermes-agent'."""
+        with patch("hermes_cli.profiles.get_active_profile_name", return_value="default"):
+            assert APIServerAdapter._resolve_model_name("") == "hermes-agent"
+
+    def test_resolve_model_name_named_profile(self):
+        """Named profile uses the profile name as model name."""
+        with patch("hermes_cli.profiles.get_active_profile_name", return_value="lucas"):
+            assert APIServerAdapter._resolve_model_name("") == "lucas"
+
     @pytest.mark.asyncio
     async def test_models_requires_auth(self, auth_adapter):
         app = _create_app(auth_adapter)
@@ -429,7 +464,7 @@ class TestChatCompletionsEndpoint:
 
     @pytest.mark.asyncio
     async def test_stream_includes_tool_progress(self, adapter):
-        """tool_progress_callback fires → progress appears in the SSE stream."""
+        """tool_progress_callback fires → progress appears as custom SSE event, not in delta.content."""
         import asyncio
 
         app = _create_app(adapter)
@@ -460,8 +495,26 @@ class TestChatCompletionsEndpoint:
                 assert resp.status == 200
                 body = await resp.text()
                 assert "[DONE]" in body
-                # Tool progress message must appear in the stream
-                assert "ls -la" in body
+                # Tool progress must appear as a custom SSE event, not in
+                # delta.content — prevents model from learning to imitate
+                # markers instead of calling tools (#6972).
+                assert "event: hermes.tool.progress" in body
+                assert '"tool": "terminal"' in body
+                assert '"label": "ls -la"' in body
+                # The progress marker must NOT appear inside any
+                # chat.completion.chunk delta.content field.
+                import json as _json
+                for line in body.splitlines():
+                    if line.startswith("data: ") and line.strip() != "data: [DONE]":
+                        try:
+                            chunk = _json.loads(line[len("data: "):])
+                        except _json.JSONDecodeError:
+                            continue
+                        if chunk.get("object") == "chat.completion.chunk":
+                            for choice in chunk.get("choices", []):
+                                content = choice.get("delta", {}).get("content", "")
+                                # Tool emoji markers must never leak into content
+                                assert "ls -la" not in content or content == "Here are the files."
                 # Final content must also be present
                 assert "Here are the files." in body
 
@@ -497,10 +550,12 @@ class TestChatCompletionsEndpoint:
                 )
                 assert resp.status == 200
                 body = await resp.text()
-                # Internal _thinking event should NOT appear
+                # Internal _thinking event should NOT appear anywhere
                 assert "some internal state" not in body
-                # Real tool progress should appear
-                assert "Python docs" in body
+                # Real tool progress should appear as custom SSE event
+                assert "event: hermes.tool.progress" in body
+                assert '"tool": "web_search"' in body
+                assert '"label": "Python docs"' in body
 
     @pytest.mark.asyncio
     async def test_no_user_message_returns_400(self, adapter):
@@ -624,6 +679,98 @@ class TestChatCompletionsEndpoint:
             data = await resp.json()
             assert "Provider failed" in data["error"]["message"]
 
+    @pytest.mark.asyncio
+    async def test_stable_session_id_across_turns(self, adapter):
+        """Same conversation (same first user message) produces the same session_id."""
+        mock_result = {"final_response": "ok", "messages": [], "api_calls": 1}
+
+        app = _create_app(adapter)
+        session_ids = []
+        async with TestClient(TestServer(app)) as cli:
+            # Turn 1: single user message
+            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
+                mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
+                await cli.post(
+                    "/v1/chat/completions",
+                    json={
+                        "model": "hermes-agent",
+                        "messages": [{"role": "user", "content": "Hello"}],
+                    },
+                )
+                session_ids.append(mock_run.call_args.kwargs["session_id"])
+
+            # Turn 2: same first message, conversation grew
+            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
+                mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
+                await cli.post(
+                    "/v1/chat/completions",
+                    json={
+                        "model": "hermes-agent",
+                        "messages": [
+                            {"role": "user", "content": "Hello"},
+                            {"role": "assistant", "content": "Hi there!"},
+                            {"role": "user", "content": "How are you?"},
+                        ],
+                    },
+                )
+                session_ids.append(mock_run.call_args.kwargs["session_id"])
+
+        assert session_ids[0] == session_ids[1], "Session ID should be stable across turns"
+        assert session_ids[0].startswith("api-"), "Derived session IDs should have api- prefix"
+
+    @pytest.mark.asyncio
+    async def test_different_conversations_get_different_session_ids(self, adapter):
+        """Different first messages produce different session_ids."""
+        mock_result = {"final_response": "ok", "messages": [], "api_calls": 1}
+
+        app = _create_app(adapter)
+        session_ids = []
+        async with TestClient(TestServer(app)) as cli:
+            for first_msg in ["Hello", "Goodbye"]:
+                with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
+                    mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
+                    await cli.post(
+                        "/v1/chat/completions",
+                        json={
+                            "model": "hermes-agent",
+                            "messages": [{"role": "user", "content": first_msg}],
+                        },
+                    )
+                    session_ids.append(mock_run.call_args.kwargs["session_id"])
+
+        assert session_ids[0] != session_ids[1]
+
+
+# ---------------------------------------------------------------------------
+# _derive_chat_session_id unit tests
+# ---------------------------------------------------------------------------
+
+
+class TestDeriveChatSessionId:
+    def test_deterministic(self):
+        """Same inputs always produce the same session ID."""
+        a = _derive_chat_session_id("sys", "hello")
+        b = _derive_chat_session_id("sys", "hello")
+        assert a == b
+
+    def test_prefix(self):
+        assert _derive_chat_session_id(None, "hi").startswith("api-")
+
+    def test_different_system_prompt(self):
+        a = _derive_chat_session_id("You are a pirate.", "Hello")
+        b = _derive_chat_session_id("You are a robot.", "Hello")
+        assert a != b
+
+    def test_different_first_message(self):
+        a = _derive_chat_session_id(None, "Hello")
+        b = _derive_chat_session_id(None, "Goodbye")
+        assert a != b
+
+    def test_none_system_prompt(self):
+        """None system prompt doesn't crash."""
+        sid = _derive_chat_session_id(None, "test")
+        assert isinstance(sid, str) and len(sid) > 4
+
 
 # ---------------------------------------------------------------------------
 # /v1/responses endpoint
@@ -1600,7 +1747,7 @@ class TestSessionIdHeader:
             assert resp.headers.get("X-Hermes-Session-Id") is not None
 
     @pytest.mark.asyncio
-    async def test_provided_session_id_is_used_and_echoed(self, adapter):
+    async def test_provided_session_id_is_used_and_echoed(self, auth_adapter):
         """When X-Hermes-Session-Id is provided, it's passed to the agent and echoed in the response."""
         mock_result = {"final_response": "Continuing!", "messages": [], "api_calls": 1}
         mock_db = MagicMock()
@@ -1608,15 +1755,15 @@ class TestSessionIdHeader:
             {"role": "user", "content": "previous message"},
             {"role": "assistant", "content": "previous reply"},
         ]
-        adapter._session_db = mock_db
-        app = _create_app(adapter)
+        auth_adapter._session_db = mock_db
+        app = _create_app(auth_adapter)
         async with TestClient(TestServer(app)) as cli:
-            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
+            with patch.object(auth_adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
                 mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
 
                 resp = await cli.post(
                     "/v1/chat/completions",
-                    headers={"X-Hermes-Session-Id": "my-session-123"},
+                    headers={"X-Hermes-Session-Id": "my-session-123", "Authorization": "Bearer sk-secret"},
                     json={"model": "hermes-agent", "messages": [{"role": "user", "content": "Continue"}]},
                 )
 
@@ -1626,7 +1773,7 @@ class TestSessionIdHeader:
             assert call_kwargs["session_id"] == "my-session-123"
 
     @pytest.mark.asyncio
-    async def test_provided_session_id_loads_history_from_db(self, adapter):
+    async def test_provided_session_id_loads_history_from_db(self, auth_adapter):
         """When X-Hermes-Session-Id is provided, history comes from SessionDB not request body."""
         mock_result = {"final_response": "OK", "messages": [], "api_calls": 1}
         db_history = [
@@ -1635,15 +1782,15 @@ class TestSessionIdHeader:
         ]
         mock_db = MagicMock()
         mock_db.get_messages_as_conversation.return_value = db_history
-        adapter._session_db = mock_db
-        app = _create_app(adapter)
+        auth_adapter._session_db = mock_db
+        app = _create_app(auth_adapter)
         async with TestClient(TestServer(app)) as cli:
-            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
+            with patch.object(auth_adapter, "_run_agent", new_callable=AsyncMock) as mock_run:
                 mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
 
                 resp = await cli.post(
                     "/v1/chat/completions",
-                    headers={"X-Hermes-Session-Id": "existing-session"},
+                    headers={"X-Hermes-Session-Id": "existing-session", "Authorization": "Bearer sk-secret"},
                     # Request body has different history — should be ignored
                     json={
                         "model": "hermes-agent",
@@ -1662,20 +1809,20 @@ class TestSessionIdHeader:
             assert call_kwargs["user_message"] == "new question"
 
     @pytest.mark.asyncio
-    async def test_db_failure_falls_back_to_empty_history(self, adapter):
+    async def test_db_failure_falls_back_to_empty_history(self, auth_adapter):
         """If SessionDB raises, history falls back to empty and request still succeeds."""
         mock_result = {"final_response": "OK", "messages": [], "api_calls": 1}
         # Simulate DB failure: _session_db is None and SessionDB() constructor raises
-        adapter._session_db = None
-        app = _create_app(adapter)
+        auth_adapter._session_db = None
+        app = _create_app(auth_adapter)
         async with TestClient(TestServer(app)) as cli:
-            with patch.object(adapter, "_run_agent", new_callable=AsyncMock) as mock_run, \
+            with patch.object(auth_adapter, "_run_agent", new_callable=AsyncMock) as mock_run, \
                  patch("hermes_state.SessionDB", side_effect=Exception("DB unavailable")):
                 mock_run.return_value = (mock_result, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
 
                 resp = await cli.post(
                     "/v1/chat/completions",
-                    headers={"X-Hermes-Session-Id": "some-session"},
+                    headers={"X-Hermes-Session-Id": "some-session", "Authorization": "Bearer sk-secret"},
                     json={"model": "hermes-agent", "messages": [{"role": "user", "content": "Hi"}]},
                 )
 

tests/gateway/test_api_server_bind_guard.py

@@ -0,0 +1,132 @@
+"""Tests for the API server bind-address startup guard.
+
+Validates that is_network_accessible() correctly classifies addresses and
+that connect() refuses to start on non-loopback without API_SERVER_KEY.
+"""
+
+import socket
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from gateway.config import PlatformConfig
+from gateway.platforms.api_server import APIServerAdapter
+from gateway.platforms.base import is_network_accessible
+
+
+# ---------------------------------------------------------------------------
+# Unit tests: is_network_accessible()
+# ---------------------------------------------------------------------------
+
+
+class TestIsNetworkAccessible:
+    """Direct tests for the address classification helper."""
+
+    # -- Loopback (safe, should return False) --
+
+    def test_ipv4_loopback(self):
+        assert is_network_accessible("127.0.0.1") is False
+
+    def test_ipv6_loopback(self):
+        assert is_network_accessible("::1") is False
+
+    def test_ipv4_mapped_loopback(self):
+        # ::ffff:127.0.0.1 — Python's is_loopback returns False for mapped
+        # addresses; the helper must unwrap and check ipv4_mapped.
+        assert is_network_accessible("::ffff:127.0.0.1") is False
+
+    # -- Network-accessible (should return True) --
+
+    def test_ipv4_wildcard(self):
+        assert is_network_accessible("0.0.0.0") is True
+
+    def test_ipv6_wildcard(self):
+        # This is the bypass vector that the string-based check missed.
+        assert is_network_accessible("::") is True
+
+    def test_ipv4_mapped_unspecified(self):
+        assert is_network_accessible("::ffff:0.0.0.0") is True
+
+    def test_private_ipv4(self):
+        assert is_network_accessible("10.0.0.1") is True
+
+    def test_private_ipv4_class_c(self):
+        assert is_network_accessible("192.168.1.1") is True
+
+    def test_public_ipv4(self):
+        assert is_network_accessible("8.8.8.8") is True
+
+    # -- Hostname resolution --
+
+    def test_localhost_resolves_to_loopback(self):
+        loopback_result = [
+            (socket.AF_INET, socket.SOCK_STREAM, 0, "", ("127.0.0.1", 0)),
+        ]
+        with patch("gateway.platforms.base._socket.getaddrinfo", return_value=loopback_result):
+            assert is_network_accessible("localhost") is False
+
+    def test_hostname_resolving_to_non_loopback(self):
+        non_loopback_result = [
+            (socket.AF_INET, socket.SOCK_STREAM, 0, "", ("10.0.0.1", 0)),
+        ]
+        with patch("gateway.platforms.base._socket.getaddrinfo", return_value=non_loopback_result):
+            assert is_network_accessible("my-server.local") is True
+
+    def test_hostname_mixed_resolution(self):
+        """If a hostname resolves to both loopback and non-loopback, it's
+        network-accessible (any non-loopback address is enough)."""
+        mixed_result = [
+            (socket.AF_INET, socket.SOCK_STREAM, 0, "", ("127.0.0.1", 0)),
+            (socket.AF_INET, socket.SOCK_STREAM, 0, "", ("10.0.0.1", 0)),
+        ]
+        with patch("gateway.platforms.base._socket.getaddrinfo", return_value=mixed_result):
+            assert is_network_accessible("dual-host.local") is True
+
+    def test_dns_failure_fails_closed(self):
+        """Unresolvable hostnames should require an API key (fail closed)."""
+        with patch(
+            "gateway.platforms.base._socket.getaddrinfo",
+            side_effect=socket.gaierror("Name resolution failed"),
+        ):
+            assert is_network_accessible("nonexistent.invalid") is True
+
+
+# ---------------------------------------------------------------------------
+# Integration tests: connect() startup guard
+# ---------------------------------------------------------------------------
+
+
+class TestConnectBindGuard:
+    """Verify that connect() refuses dangerous configurations."""
+
+    @pytest.mark.asyncio
+    async def test_refuses_ipv4_wildcard_without_key(self):
+        adapter = APIServerAdapter(PlatformConfig(enabled=True, extra={"host": "0.0.0.0"}))
+        result = await adapter.connect()
+        assert result is False
+
+    @pytest.mark.asyncio
+    async def test_refuses_ipv6_wildcard_without_key(self):
+        adapter = APIServerAdapter(PlatformConfig(enabled=True, extra={"host": "::"}))
+        result = await adapter.connect()
+        assert result is False
+
+    def test_allows_loopback_without_key(self):
+        """Loopback with no key should pass the guard."""
+        adapter = APIServerAdapter(PlatformConfig(enabled=True, extra={"host": "127.0.0.1"}))
+        assert adapter._api_key == ""
+        # The guard condition: is_network_accessible(host) AND NOT api_key
+        # For loopback, is_network_accessible is False so the guard does not block.
+        assert is_network_accessible(adapter._host) is False
+
+    @pytest.mark.asyncio
+    async def test_allows_wildcard_with_key(self):
+        """Non-loopback with a key should pass the guard."""
+        adapter = APIServerAdapter(
+            PlatformConfig(enabled=True, extra={"host": "0.0.0.0", "key": "sk-test"})
+        )
+        # The guard checks: is_network_accessible(host) AND NOT api_key
+        # With a key set, the guard should not block.
+        assert adapter._api_key == "sk-test"
+        assert is_network_accessible("0.0.0.0") is True
+        # Combined: the guard condition is False (key is set), so it passes

tests/gateway/test_background_command.py

@@ -308,6 +308,7 @@ class TestBackgroundInCLICommands:
 
     def test_background_autocompletes(self):
         """The /background command appears in autocomplete results."""
+        pytest.importorskip("prompt_toolkit")
         from hermes_cli.commands import SlashCommandCompleter
         from prompt_toolkit.document import Document
 

tests/gateway/test_base_topic_sessions.py

@@ -6,7 +6,7 @@ from types import SimpleNamespace
 import pytest
 
 from gateway.config import Platform, PlatformConfig
-from gateway.platforms.base import BasePlatformAdapter, MessageEvent, SendResult
+from gateway.platforms.base import BasePlatformAdapter, MessageEvent, ProcessingOutcome, SendResult
 from gateway.session import SessionSource, build_session_key
 
 
@@ -44,8 +44,8 @@ class DummyTelegramAdapter(BasePlatformAdapter):
     async def on_processing_start(self, event: MessageEvent) -> None:
         self.processing_hooks.append(("start", event.message_id))
 
-    async def on_processing_complete(self, event: MessageEvent, success: bool) -> None:
-        self.processing_hooks.append(("complete", event.message_id, success))
+    async def on_processing_complete(self, event: MessageEvent, outcome: ProcessingOutcome) -> None:
+        self.processing_hooks.append(("complete", event.message_id, outcome))
 
 
 def _make_event(chat_id: str, thread_id: str, message_id: str = "1") -> MessageEvent:
@@ -142,7 +142,7 @@ class TestBasePlatformTopicSessions:
         ]
         assert adapter.processing_hooks == [
             ("start", "1"),
-            ("complete", "1", True),
+            ("complete", "1", ProcessingOutcome.SUCCESS),
         ]
 
     @pytest.mark.asyncio
@@ -168,7 +168,7 @@ class TestBasePlatformTopicSessions:
 
         assert adapter.processing_hooks == [
             ("start", "1"),
-            ("complete", "1", False),
+            ("complete", "1", ProcessingOutcome.FAILURE),
         ]
 
     @pytest.mark.asyncio
@@ -190,7 +190,7 @@ class TestBasePlatformTopicSessions:
 
         assert adapter.processing_hooks == [
             ("start", "1"),
-            ("complete", "1", False),
+            ("complete", "1", ProcessingOutcome.FAILURE),
         ]
 
     @pytest.mark.asyncio
@@ -218,5 +218,31 @@ class TestBasePlatformTopicSessions:
 
         assert adapter.processing_hooks == [
             ("start", "1"),
-            ("complete", "1", False),
+            ("complete", "1", ProcessingOutcome.FAILURE),
+        ]
+
+    @pytest.mark.asyncio
+    async def test_cancel_background_tasks_marks_expected_cancellation_cancelled(self):
+        adapter = DummyTelegramAdapter()
+        release = asyncio.Event()
+
+        async def handler(_event):
+            await release.wait()
+            return "ack"
+
+        async def hold_typing(_chat_id, interval=2.0, metadata=None):
+            await asyncio.Event().wait()
+
+        adapter.set_message_handler(handler)
+        adapter._keep_typing = hold_typing
+
+        event = _make_event("-1001", "17585")
+        await adapter.handle_message(event)
+        await asyncio.sleep(0)
+
+        await adapter.cancel_background_tasks()
+
+        assert adapter.processing_hooks == [
+            ("start", "1"),
+            ("complete", "1", ProcessingOutcome.CANCELLED),
         ]

tests/gateway/test_bluebubbles.py

@@ -359,3 +359,257 @@ class TestBlueBubblesAttachmentDownload:
             adapter._download_attachment("att-guid", {"mimeType": "image/png"})
         )
         assert result is None
+
+
+# ---------------------------------------------------------------------------
+# Webhook registration
+# ---------------------------------------------------------------------------
+
+
+class TestBlueBubblesWebhookUrl:
+    """_webhook_url property normalises local hosts to 'localhost'."""
+
+    def test_default_host(self, monkeypatch):
+        adapter = _make_adapter(monkeypatch)
+        # Default webhook_host is 0.0.0.0 → normalized to localhost
+        assert "localhost" in adapter._webhook_url
+        assert str(adapter.webhook_port) in adapter._webhook_url
+        assert adapter.webhook_path in adapter._webhook_url
+
+    @pytest.mark.parametrize("host", ["0.0.0.0", "127.0.0.1", "localhost", "::"])
+    def test_local_hosts_normalized(self, monkeypatch, host):
+        adapter = _make_adapter(monkeypatch, webhook_host=host)
+        assert adapter._webhook_url.startswith("http://localhost:")
+
+    def test_custom_host_preserved(self, monkeypatch):
+        adapter = _make_adapter(monkeypatch, webhook_host="192.168.1.50")
+        assert "192.168.1.50" in adapter._webhook_url
+
+
+class TestBlueBubblesWebhookRegistration:
+    """Tests for _register_webhook, _unregister_webhook, _find_registered_webhooks."""
+
+    @staticmethod
+    def _mock_client(get_response=None, post_response=None, delete_ok=True):
+        """Build a tiny mock httpx.AsyncClient."""
+
+        async def mock_get(*args, **kwargs):
+            class R:
+                status_code = 200
+                def raise_for_status(self):
+                    pass
+                def json(self):
+                    return get_response or {"status": 200, "data": []}
+            return R()
+
+        async def mock_post(*args, **kwargs):
+            class R:
+                status_code = 200
+                def raise_for_status(self):
+                    pass
+                def json(self):
+                    return post_response or {"status": 200, "data": {}}
+            return R()
+
+        async def mock_delete(*args, **kwargs):
+            class R:
+                status_code = 200 if delete_ok else 500
+                def raise_for_status(self_inner):
+                    if not delete_ok:
+                        raise Exception("delete failed")
+            return R()
+
+        return type(
+            "MockClient", (),
+            {"get": mock_get, "post": mock_post, "delete": mock_delete},
+        )()
+
+    # -- _find_registered_webhooks --
+
+    def test_find_registered_webhooks_returns_matches(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        url = adapter._webhook_url
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": [
+                {"id": 1, "url": url, "events": ["new-message"]},
+                {"id": 2, "url": "http://other:9999/hook", "events": ["message"]},
+            ]}
+        )
+        result = asyncio.get_event_loop().run_until_complete(
+            adapter._find_registered_webhooks(url)
+        )
+        assert len(result) == 1
+        assert result[0]["id"] == 1
+
+    def test_find_registered_webhooks_empty_when_none(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": []}
+        )
+        result = asyncio.get_event_loop().run_until_complete(
+            adapter._find_registered_webhooks(adapter._webhook_url)
+        )
+        assert result == []
+
+    def test_find_registered_webhooks_handles_api_error(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client()
+
+        # Override _api_get to raise
+        async def bad_get(path):
+            raise ConnectionError("server down")
+        adapter._api_get = bad_get
+
+        result = asyncio.get_event_loop().run_until_complete(
+            adapter._find_registered_webhooks(adapter._webhook_url)
+        )
+        assert result == []
+
+    # -- _register_webhook --
+
+    def test_register_fresh(self, monkeypatch):
+        """No existing webhook → POST creates one."""
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": []},
+            post_response={"status": 200, "data": {"id": 42}},
+        )
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._register_webhook()
+        )
+        assert ok is True
+
+    def test_register_accepts_201(self, monkeypatch):
+        """BB might return 201 Created — must still succeed."""
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": []},
+            post_response={"status": 201, "data": {"id": 43}},
+        )
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._register_webhook()
+        )
+        assert ok is True
+
+    def test_register_reuses_existing(self, monkeypatch):
+        """Crash resilience — existing registration is reused, no POST needed."""
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        url = adapter._webhook_url
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": [
+                {"id": 7, "url": url, "events": ["new-message"]},
+            ]},
+        )
+
+        # Track whether POST was called
+        post_called = False
+        orig_api_post = adapter._api_post
+        async def tracking_post(path, payload):
+            nonlocal post_called
+            post_called = True
+            return await orig_api_post(path, payload)
+        adapter._api_post = tracking_post
+
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._register_webhook()
+        )
+        assert ok is True
+        assert not post_called, "Should reuse existing, not POST again"
+
+    def test_register_returns_false_without_client(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = None
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._register_webhook()
+        )
+        assert ok is False
+
+    def test_register_returns_false_on_server_error(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": []},
+            post_response={"status": 500, "message": "internal error"},
+        )
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._register_webhook()
+        )
+        assert ok is False
+
+    # -- _unregister_webhook --
+
+    def test_unregister_removes_matching(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        url = adapter._webhook_url
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": [
+                {"id": 10, "url": url},
+            ]},
+        )
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._unregister_webhook()
+        )
+        assert ok is True
+
+    def test_unregister_removes_all_duplicates(self, monkeypatch):
+        """Multiple orphaned registrations for same URL — all get removed."""
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        url = adapter._webhook_url
+        deleted_ids = []
+
+        async def mock_delete(*args, **kwargs):
+            # Extract ID from URL
+            url_str = args[0] if args else ""
+            deleted_ids.append(url_str)
+            class R:
+                status_code = 200
+                def raise_for_status(self):
+                    pass
+            return R()
+
+        adapter.client = self._mock_client(
+            get_response={"status": 200, "data": [
+                {"id": 1, "url": url},
+                {"id": 2, "url": url},
+                {"id": 3, "url": "http://other/hook"},
+            ]},
+        )
+        adapter.client.delete = mock_delete
+
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._unregister_webhook()
+        )
+        assert ok is True
+        assert len(deleted_ids) == 2
+
+    def test_unregister_returns_false_without_client(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = None
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._unregister_webhook()
+        )
+        assert ok is False
+
+    def test_unregister_handles_api_failure_gracefully(self, monkeypatch):
+        import asyncio
+        adapter = _make_adapter(monkeypatch)
+        adapter.client = self._mock_client()
+
+        async def bad_get(path):
+            raise ConnectionError("server down")
+        adapter._api_get = bad_get
+
+        ok = asyncio.get_event_loop().run_until_complete(
+            adapter._unregister_webhook()
+        )
+        assert ok is False

tests/gateway/test_command_bypass_active_session.py

@@ -160,6 +160,22 @@ class TestCommandBypassActiveSession:
         assert sk not in adapter._pending_messages
         assert any("handled:status" in r for r in adapter.sent_responses)
 
+    @pytest.mark.asyncio
+    async def test_background_bypasses_guard(self):
+        """/background must bypass so it spawns a parallel task, not an interrupt."""
+        adapter = _make_adapter()
+        sk = _session_key()
+        adapter._active_sessions[sk] = asyncio.Event()
+
+        await adapter.handle_message(_make_event("/background summarize HN"))
+
+        assert sk not in adapter._pending_messages, (
+            "/background was queued as a pending message instead of being dispatched"
+        )
+        assert any("handled:background" in r for r in adapter.sent_responses), (
+            "/background response was not sent back to the user"
+        )
+
 
 # ---------------------------------------------------------------------------
 # Tests: non-bypass messages still get queued

tests/gateway/test_delivery.py

@@ -1,7 +1,7 @@
 """Tests for the delivery routing module."""
 
-from gateway.config import Platform, GatewayConfig, PlatformConfig, HomeChannel
-from gateway.delivery import DeliveryRouter, DeliveryTarget
+from gateway.config import Platform
+from gateway.delivery import DeliveryTarget
 from gateway.session import SessionSource
 
 
@@ -65,10 +65,4 @@ class TestTargetToStringRoundtrip:
         assert reparsed.chat_id == "999"
 
 
-class TestDeliveryRouter:
-    def test_resolve_targets_does_not_duplicate_local_when_explicit(self):
-        router = DeliveryRouter(GatewayConfig(always_log_local=True))
 
-        targets = router.resolve_targets(["local"])
-
-        assert [target.platform for target in targets] == [Platform.LOCAL]

tests/gateway/test_discord_channel_controls.py

@@ -81,6 +81,7 @@ def adapter(monkeypatch):
     config = PlatformConfig(enabled=True, token="fake-token")
     adapter = DiscordAdapter(config)
     adapter._client = SimpleNamespace(user=SimpleNamespace(id=999))
+    adapter._text_batch_delay_seconds = 0  # disable batching for tests
     adapter.handle_message = AsyncMock()
     return adapter
 

tests/gateway/test_discord_channel_skills.py

@@ -0,0 +1,64 @@
+"""Tests for Discord channel_skill_bindings auto-skill resolution."""
+from unittest.mock import MagicMock
+import pytest
+
+
+def _make_adapter():
+    """Create a minimal DiscordAdapter with mocked config."""
+    from gateway.platforms.discord import DiscordAdapter
+    adapter = object.__new__(DiscordAdapter)
+    adapter.config = MagicMock()
+    adapter.config.extra = {}
+    return adapter
+
+
+class TestResolveChannelSkills:
+    def test_no_bindings_returns_none(self):
+        adapter = _make_adapter()
+        assert adapter._resolve_channel_skills("123") is None
+
+    def test_match_by_channel_id(self):
+        adapter = _make_adapter()
+        adapter.config.extra = {
+            "channel_skill_bindings": [
+                {"id": "100", "skills": ["skill-a", "skill-b"]},
+            ]
+        }
+        assert adapter._resolve_channel_skills("100") == ["skill-a", "skill-b"]
+
+    def test_match_by_parent_id(self):
+        adapter = _make_adapter()
+        adapter.config.extra = {
+            "channel_skill_bindings": [
+                {"id": "200", "skills": ["forum-skill"]},
+            ]
+        }
+        # channel_id doesn't match, but parent_id does (forum thread)
+        assert adapter._resolve_channel_skills("999", parent_id="200") == ["forum-skill"]
+
+    def test_no_match_returns_none(self):
+        adapter = _make_adapter()
+        adapter.config.extra = {
+            "channel_skill_bindings": [
+                {"id": "100", "skills": ["skill-a"]},
+            ]
+        }
+        assert adapter._resolve_channel_skills("999") is None
+
+    def test_single_skill_string(self):
+        adapter = _make_adapter()
+        adapter.config.extra = {
+            "channel_skill_bindings": [
+                {"id": "100", "skill": "solo-skill"},
+            ]
+        }
+        assert adapter._resolve_channel_skills("100") == ["solo-skill"]
+
+    def test_dedup_preserves_order(self):
+        adapter = _make_adapter()
+        adapter.config.extra = {
+            "channel_skill_bindings": [
+                {"id": "100", "skills": ["a", "b", "a", "c", "b"]},
+            ]
+        }
+        assert adapter._resolve_channel_skills("100") == ["a", "b", "c"]

tests/gateway/test_discord_free_response.py

@@ -91,6 +91,7 @@ def adapter(monkeypatch):
     config = PlatformConfig(enabled=True, token="fake-token")
     adapter = DiscordAdapter(config)
     adapter._client = SimpleNamespace(user=SimpleNamespace(id=999))
+    adapter._text_batch_delay_seconds = 0  # disable batching for tests
     adapter.handle_message = AsyncMock()
     return adapter
 

tests/gateway/test_discord_reactions.py

@@ -8,7 +8,7 @@ from unittest.mock import AsyncMock, MagicMock
 import pytest
 
 from gateway.config import Platform, PlatformConfig
-from gateway.platforms.base import MessageEvent, MessageType, SendResult
+from gateway.platforms.base import MessageEvent, MessageType, ProcessingOutcome, SendResult
 from gateway.session import SessionSource, build_session_key
 
 
@@ -212,7 +212,7 @@ async def test_reactions_disabled_via_env_zero(adapter, monkeypatch):
 
     event = _make_event("5", raw_message)
     await adapter.on_processing_start(event)
-    await adapter.on_processing_complete(event, success=True)
+    await adapter.on_processing_complete(event, ProcessingOutcome.SUCCESS)
 
     raw_message.add_reaction.assert_not_awaited()
     raw_message.remove_reaction.assert_not_awaited()
@@ -232,3 +232,17 @@ async def test_reactions_enabled_by_default(adapter, monkeypatch):
     await adapter.on_processing_start(event)
 
     raw_message.add_reaction.assert_awaited_once_with("👀")
+
+
+@pytest.mark.asyncio
+async def test_on_processing_complete_cancelled_removes_eyes_without_terminal_reaction(adapter):
+    raw_message = SimpleNamespace(
+        add_reaction=AsyncMock(),
+        remove_reaction=AsyncMock(),
+    )
+
+    event = _make_event("7", raw_message)
+    await adapter.on_processing_complete(event, ProcessingOutcome.CANCELLED)
+
+    raw_message.remove_reaction.assert_awaited_once_with("👀", adapter._client.user)
+    raw_message.add_reaction.assert_not_awaited()

tests/gateway/test_discord_slash_commands.py

@@ -62,6 +62,7 @@ def adapter():
         fetch_channel=AsyncMock(),
         user=SimpleNamespace(id=99999, name="HermesBot"),
     )
+    adapter._text_batch_delay_seconds = 0  # disable batching for tests
     return adapter
 
 

tests/gateway/test_fast_command.py

@@ -0,0 +1,191 @@
+"""Tests for gateway /fast support and Priority Processing routing."""
+
+import sys
+import threading
+import types
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, patch
+
+import pytest
+import yaml
+
+import gateway.run as gateway_run
+from gateway.config import Platform
+from gateway.platforms.base import MessageEvent
+from gateway.session import SessionSource
+
+
+class _CapturingAgent:
+    last_init = None
+    last_run = None
+
+    def __init__(self, *args, **kwargs):
+        type(self).last_init = dict(kwargs)
+        self.tools = []
+
+    def run_conversation(self, user_message, conversation_history=None, task_id=None, persist_user_message=None):
+        type(self).last_run = {
+            "user_message": user_message,
+            "conversation_history": conversation_history,
+            "task_id": task_id,
+            "persist_user_message": persist_user_message,
+        }
+        return {
+            "final_response": "ok",
+            "messages": [],
+            "api_calls": 1,
+            "completed": True,
+        }
+
+
+def _install_fake_agent(monkeypatch):
+    fake_run_agent = types.ModuleType("run_agent")
+    fake_run_agent.AIAgent = _CapturingAgent
+    monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent)
+
+
+def _make_runner():
+    runner = object.__new__(gateway_run.GatewayRunner)
+    runner.adapters = {}
+    runner._ephemeral_system_prompt = ""
+    runner._prefill_messages = []
+    runner._reasoning_config = None
+    runner._service_tier = None
+    runner._provider_routing = {}
+    runner._fallback_model = None
+    runner._smart_model_routing = {}
+    runner._running_agents = {}
+    runner._pending_model_notes = {}
+    runner._session_db = None
+    runner._agent_cache = {}
+    runner._agent_cache_lock = threading.Lock()
+    runner._session_model_overrides = {}
+    runner.hooks = SimpleNamespace(loaded_hooks=False)
+    runner.config = SimpleNamespace(streaming=None)
+    runner.session_store = SimpleNamespace(
+        get_or_create_session=lambda source: SimpleNamespace(session_id="session-1"),
+        load_transcript=lambda session_id: [],
+    )
+    runner._get_or_create_gateway_honcho = lambda session_key: (None, None)
+    runner._enrich_message_with_vision = AsyncMock(return_value="ENRICHED")
+    return runner
+
+
+def _make_source() -> SessionSource:
+    return SessionSource(
+        platform=Platform.TELEGRAM,
+        chat_id="12345",
+        chat_type="dm",
+        user_id="user-1",
+    )
+
+
+def _make_event(text: str) -> MessageEvent:
+    return MessageEvent(text=text, source=_make_source(), message_id="m1")
+
+
+def test_turn_route_injects_priority_processing_without_changing_runtime():
+    runner = _make_runner()
+    runner._service_tier = "priority"
+    runtime_kwargs = {
+        "api_key": "***",
+        "base_url": "https://openrouter.ai/api/v1",
+        "provider": "openrouter",
+        "api_mode": "chat_completions",
+        "command": None,
+        "args": [],
+        "credential_pool": None,
+    }
+
+    with patch("agent.smart_model_routing.resolve_turn_route", return_value={
+        "model": "gpt-5.4",
+        "runtime": dict(runtime_kwargs),
+        "label": None,
+        "signature": ("gpt-5.4", "openrouter", "https://openrouter.ai/api/v1", "chat_completions", None, ()),
+    }):
+        route = gateway_run.GatewayRunner._resolve_turn_agent_config(runner, "hi", "gpt-5.4", runtime_kwargs)
+
+    assert route["runtime"]["provider"] == "openrouter"
+    assert route["runtime"]["api_mode"] == "chat_completions"
+    assert route["request_overrides"] == {"service_tier": "priority"}
+
+
+def test_turn_route_skips_priority_processing_for_unsupported_models():
+    runner = _make_runner()
+    runner._service_tier = "priority"
+    runtime_kwargs = {
+        "api_key": "***",
+        "base_url": "https://openrouter.ai/api/v1",
+        "provider": "openrouter",
+        "api_mode": "chat_completions",
+        "command": None,
+        "args": [],
+        "credential_pool": None,
+    }
+
+    with patch("agent.smart_model_routing.resolve_turn_route", return_value={
+        "model": "gpt-5.3-codex",
+        "runtime": dict(runtime_kwargs),
+        "label": None,
+        "signature": ("gpt-5.3-codex", "openrouter", "https://openrouter.ai/api/v1", "chat_completions", None, ()),
+    }):
+        route = gateway_run.GatewayRunner._resolve_turn_agent_config(runner, "hi", "gpt-5.3-codex", runtime_kwargs)
+
+    assert route["request_overrides"] is None
+
+
+@pytest.mark.asyncio
+async def test_handle_fast_command_persists_config(monkeypatch, tmp_path):
+    runner = _make_runner()
+
+    monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path)
+    monkeypatch.setattr(gateway_run, "_load_gateway_config", lambda: {})
+    monkeypatch.setattr(gateway_run, "_resolve_gateway_model", lambda config=None: "gpt-5.4")
+
+    response = await runner._handle_fast_command(_make_event("/fast fast"))
+
+    assert "FAST" in response
+    assert runner._service_tier == "priority"
+
+    saved = yaml.safe_load((tmp_path / "config.yaml").read_text(encoding="utf-8"))
+    assert saved["agent"]["service_tier"] == "fast"
+
+
+@pytest.mark.asyncio
+async def test_run_agent_passes_priority_processing_to_gateway_agent(monkeypatch, tmp_path):
+    _install_fake_agent(monkeypatch)
+    runner = _make_runner()
+
+    (tmp_path / "config.yaml").write_text("agent:\n  service_tier: fast\n", encoding="utf-8")
+    monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path)
+    monkeypatch.setattr(gateway_run, "_env_path", tmp_path / ".env")
+    monkeypatch.setattr(gateway_run, "load_dotenv", lambda *args, **kwargs: None)
+    monkeypatch.setattr(gateway_run, "_load_gateway_config", lambda: {})
+    monkeypatch.setattr(gateway_run, "_resolve_gateway_model", lambda config=None: "gpt-5.4")
+    monkeypatch.setattr(
+        gateway_run,
+        "_resolve_runtime_agent_kwargs",
+        lambda: {
+            "provider": "openrouter",
+            "api_mode": "chat_completions",
+            "base_url": "https://openrouter.ai/api/v1",
+            "api_key": "***",
+        },
+    )
+
+    import hermes_cli.tools_config as tools_config
+    monkeypatch.setattr(tools_config, "_get_platform_tools", lambda user_config, platform_key: {"core"})
+
+    _CapturingAgent.last_init = None
+    result = await runner._run_agent(
+        message="hi",
+        context_prompt="",
+        history=[],
+        source=_make_source(),
+        session_id="session-1",
+        session_key="agent:main:telegram:dm:12345",
+    )
+
+    assert result["final_response"] == "ok"
+    assert _CapturingAgent.last_init["service_tier"] == "priority"
+    assert _CapturingAgent.last_init["request_overrides"] == {"service_tier": "priority"}