Merge branch 'main' of github.com:NousResearch/hermes-agent into bb/gui

# Conflicts:
#	cli.py
#	hermes_cli/main.py
#	run_agent.py
#	tests/hermes_cli/test_cmd_update.py
#	tools/mcp_tool.py
#	web/src/lib/gatewayClient.ts

gateway/platforms/api_server.py

@@ -71,6 +71,35 @@ def _coerce_port(value: Any, default: int = DEFAULT_PORT) -> int:
         return default
 
 
+_TRUE_REQUEST_BOOL_STRINGS = frozenset({"1", "true", "yes", "on"})
+_FALSE_REQUEST_BOOL_STRINGS = frozenset({"0", "false", "no", "off"})
+
+
+def _coerce_request_bool(value: Any, default: bool = False) -> bool:
+    """Normalize boolean-like API payload values.
+
+    External clients should send real JSON booleans, but some OpenAI-compatible
+    frontends and middleware serialize flags like ``stream`` as strings.  Using
+    Python truthiness on those values misroutes requests because ``"false"`` is
+    still truthy.  Treat only explicit bool-ish scalars as booleans; everything
+    else falls back to the caller's default.
+    """
+    if isinstance(value, bool):
+        return value
+    if value is None:
+        return default
+    if isinstance(value, str):
+        normalized = value.strip().lower()
+        if normalized in _TRUE_REQUEST_BOOL_STRINGS:
+            return True
+        if normalized in _FALSE_REQUEST_BOOL_STRINGS:
+            return False
+        return default
+    if isinstance(value, (int, float)):
+        return bool(value)
+    return default
+
+
 def _normalize_chat_content(
     content: Any, *, _max_depth: int = 10, _depth: int = 0,
 ) -> str:
@@ -481,7 +510,12 @@ else:
     body_limit_middleware = None  # type: ignore[assignment]
 
 _SECURITY_HEADERS = {
+    "Content-Security-Policy": "default-src 'none'; frame-ancestors 'none'",
+    "Permissions-Policy": "camera=(), microphone=(), geolocation=()",
+    "Strict-Transport-Security": "max-age=31536000; includeSubDomains",
     "X-Content-Type-Options": "nosniff",
+    "X-Frame-Options": "DENY",
+    "X-XSS-Protection": "0",
     "Referrer-Policy": "no-referrer",
 }
 
@@ -1005,7 +1039,7 @@ class APIServerAdapter(BasePlatformAdapter):
                 status=400,
             )
 
-        stream = body.get("stream", False)
+        stream = _coerce_request_bool(body.get("stream"), default=False)
 
         # Extract system message (becomes ephemeral system prompt layered ON TOP of core)
         system_prompt = None
@@ -2082,7 +2116,7 @@ class APIServerAdapter(BasePlatformAdapter):
         instructions = body.get("instructions")
         previous_response_id = body.get("previous_response_id")
         conversation = body.get("conversation")
-        store = body.get("store", True)
+        store = _coerce_request_bool(body.get("store"), default=True)
 
         # conversation and previous_response_id are mutually exclusive
         if conversation and previous_response_id:
@@ -2165,7 +2199,7 @@ class APIServerAdapter(BasePlatformAdapter):
         # groups the entire conversation under one session entry.
         session_id = stored_session_id or str(uuid.uuid4())
 
-        stream = bool(body.get("stream", False))
+        stream = _coerce_request_bool(body.get("stream"), default=False)
         if stream:
             # Streaming branch — emit OpenAI Responses SSE events as the
             # agent runs so frontends can render text deltas and tool
@@ -3228,7 +3262,10 @@ class APIServerAdapter(BasePlatformAdapter):
                 status=409,
             )
 
-        resolve_all = bool(body.get("all") or body.get("resolve_all"))
+        resolve_all = (
+            _coerce_request_bool(body.get("all"), default=False)
+            or _coerce_request_bool(body.get("resolve_all"), default=False)
+        )
         try:
             from tools.approval import resolve_gateway_approval
 

gateway/platforms/base.py

@@ -2014,6 +2014,13 @@ class BasePlatformAdapter(ABC):
             text = f"{caption}\n{text}"
         return await self.send(chat_id=chat_id, content=text, reply_to=reply_to, metadata=metadata)
 
+    def prepare_tts_text(self, text: str) -> str:
+        """Prepare text for TTS. Override to filter tool output, code, etc.
+
+        Default strips markdown formatting and truncates to 4000 chars.
+        """
+        return re.sub(r'[*_`#\[\]()]', '', text)[:4000].strip()
+
     async def play_tts(
         self,
         chat_id: str,
@@ -3144,7 +3151,7 @@ class BasePlatformAdapter(ABC):
                         from tools.tts_tool import text_to_speech_tool, check_tts_requirements
                         if check_tts_requirements():
                             import json as _json
-                            speech_text = re.sub(r'[*_`#\[\]()]', '', text_content)[:4000].strip()
+                            speech_text = self.prepare_tts_text(text_content)
                             if not speech_text:
                                 raise ValueError("Empty text after markdown cleanup")
                             tts_result_str = await asyncio.to_thread(

gateway/platforms/discord.py

@@ -3639,18 +3639,18 @@ class DiscordAdapter(BasePlatformAdapter):
         configured = self.config.extra.get("thread_require_mention")
         if configured is not None:
             if isinstance(configured, str):
-                return configured.lower() not in ("false", "0", "no", "off")
+                return configured.lower() not in {"false", "0", "no", "off"}
             return bool(configured)
-        return os.getenv("DISCORD_THREAD_REQUIRE_MENTION", "false").lower() in ("true", "1", "yes", "on")
+        return os.getenv("DISCORD_THREAD_REQUIRE_MENTION", "false").lower() in {"true", "1", "yes", "on"}
 
     def _discord_history_backfill(self) -> bool:
         """Return whether history backfill is enabled for shared sessions."""
         configured = self.config.extra.get("history_backfill")
         if configured is not None:
             if isinstance(configured, str):
-                return configured.lower() not in ("false", "0", "no", "off")
+                return configured.lower() not in {"false", "0", "no", "off"}
             return bool(configured)
-        return os.getenv("DISCORD_HISTORY_BACKFILL", "true").lower() in ("true", "1", "yes")
+        return os.getenv("DISCORD_HISTORY_BACKFILL", "true").lower() in {"true", "1", "yes"}
 
     def _discord_history_backfill_limit(self) -> int:
         """Return the max number of messages to scan backwards for context.
@@ -3737,7 +3737,7 @@ class DiscordAdapter(BasePlatformAdapter):
                     break
 
                 # Skip system messages (pins, joins, thread renames, etc.)
-                if msg.type not in (discord.MessageType.default, discord.MessageType.reply):
+                if msg.type not in {discord.MessageType.default, discord.MessageType.reply}:
                     continue
 
                 # Respect DISCORD_ALLOW_BOTS for other bots.

gateway/platforms/helpers.py

@@ -168,8 +168,8 @@ class TextBatchAggregator:
 # Pre-compiled regexes for performance
 _RE_BOLD = re.compile(r"\*\*(.+?)\*\*", re.DOTALL)
 _RE_ITALIC_STAR = re.compile(r"\*(.+?)\*", re.DOTALL)
-_RE_BOLD_UNDER = re.compile(r"__(.+?)__", re.DOTALL)
-_RE_ITALIC_UNDER = re.compile(r"_(.+?)_", re.DOTALL)
+_RE_BOLD_UNDER = re.compile(r"\b__(?![\s_])(.+?)(?<![\s_])__\b", re.DOTALL)
+_RE_ITALIC_UNDER = re.compile(r"\b_(?![\s_])(.+?)(?<![\s_])_\b", re.DOTALL)
 _RE_CODE_BLOCK = re.compile(r"```[a-zA-Z0-9_+-]*\n?")
 _RE_INLINE_CODE = re.compile(r"`(.+?)`")
 _RE_HEADING = re.compile(r"^#{1,6}\s+", re.MULTILINE)

gateway/platforms/matrix.py

@@ -348,6 +348,17 @@ class MatrixAdapter(BasePlatformAdapter):
         self._sync_task: Optional[asyncio.Task] = None
         self._closing = False
         self._startup_ts: float = 0.0
+        # Clock-skew detection: count grace-check drops that happen well
+        # after startup (i.e. not initial-sync backfill).  If the host's
+        # system clock is set ahead of real time, the startup grace check
+        # `event_ts < startup_ts - 5` silently drops every live message.
+        # See #12614 — the symptom is "bot joins rooms but never replies".
+        # Drops only count when their skew matches the first sampled drop
+        # (within 60s), so varied-age backfill from freshly-invited rooms
+        # doesn't trip the heuristic.
+        self._late_grace_drops: int = 0
+        self._late_grace_skew: float = 0.0
+        self._clock_skew_warned: bool = False
 
         # Cache: room_id → bool (is DM)
         self._dm_rooms: Dict[str, bool] = {}
@@ -842,6 +853,11 @@ class MatrixAdapter(BasePlatformAdapter):
 
         # Initial sync to catch up, then start background sync.
         self._startup_ts = time.time()
+        # Reset clock-skew detector for each connect cycle so a reconnect
+        # after the user fixes NTP doesn't inherit stale counters.
+        self._late_grace_drops = 0
+        self._late_grace_skew = 0.0
+        self._clock_skew_warned = False
         self._closing = False
 
         try:
@@ -1542,6 +1558,49 @@ class MatrixAdapter(BasePlatformAdapter):
         )
         event_ts = raw_ts / 1000.0 if raw_ts else 0.0
         if event_ts and event_ts < self._startup_ts - _STARTUP_GRACE_SECONDS:
+            # If we are well past startup but events are still being dropped
+            # by the grace check, the host clock is probably set ahead of
+            # real time — every live event then looks "older than startup".
+            # Warn once so users can fix NTP instead of chasing a ghost.
+            # See #12614 (Schnurzel700, April 2026).
+            #
+            # Filter out backfill (events legitimately old) by requiring:
+            #  - we are >30s past startup (initial-sync replay window closed)
+            #  - the skew is *consistent* across consecutive drops, which is
+            #    the signature of a constant clock offset rather than a
+            #    variable-age room history.  Backfill from a freshly invited
+            #    room can deliver events spanning hours/days — those skews
+            #    will be all over the place and reset the counter.
+            if not self._clock_skew_warned and (
+                time.time() - self._startup_ts > 30
+            ):
+                skew = self._startup_ts - event_ts
+                # Sanity bound: malformed events with negative or absurd
+                # timestamps shouldn't count.
+                if 5 < skew < 86400:
+                    if self._late_grace_drops == 0:
+                        self._late_grace_skew = skew
+                        self._late_grace_drops = 1
+                    elif abs(skew - self._late_grace_skew) < 60:
+                        # Consistent offset → likely real clock skew.
+                        self._late_grace_drops += 1
+                    else:
+                        # Varied skew → likely backfill, restart sampling.
+                        self._late_grace_skew = skew
+                        self._late_grace_drops = 1
+                    if self._late_grace_drops >= 3:
+                        logger.warning(
+                            "Matrix: dropped %d consecutive live events as "
+                            "'too old' more than 30s after startup (skew "
+                            "≈ %.0fs). The host system clock is likely set "
+                            "ahead of real time, which causes the startup "
+                            "grace filter to silently discard every incoming "
+                            "message. Run `timedatectl set-ntp true` (or "
+                            "sync NTP) and restart the bot.",
+                            self._late_grace_drops,
+                            skew,
+                        )
+                        self._clock_skew_warned = True
             return
 
         # Extract content from the event.

gateway/platforms/slack.py

@@ -482,7 +482,7 @@ class SlackAdapter(BasePlatformAdapter):
             "text": text,
         }
         try:
-            async with aiohttp.ClientSession() as session:
+            async with aiohttp.ClientSession(trust_env=True) as session:
                 async with session.post(
                     ctx["response_url"],
                     json=payload,

gateway/platforms/sms.py

@@ -128,6 +128,7 @@ class SmsAdapter(BasePlatformAdapter):
         await site.start()
         self._http_session = aiohttp.ClientSession(
             timeout=aiohttp.ClientTimeout(total=30),
+            trust_env=True,
         )
         self._running = True
 
@@ -169,6 +170,7 @@ class SmsAdapter(BasePlatformAdapter):
 
         session = self._http_session or aiohttp.ClientSession(
             timeout=aiohttp.ClientTimeout(total=30),
+            trust_env=True,
         )
         try:
             for chunk in chunks:

gateway/platforms/telegram.py

@@ -1663,7 +1663,17 @@ class TelegramAdapter(BasePlatformAdapter):
                                 continue
                         raise
                 message_ids.append(str(msg.message_id))
-            
+
+            # Re-trigger typing indicator after sending a message.
+            # Telegram clears the typing state when a new message is delivered,
+            # so without this the "...typing" bubble disappears mid-response
+            # (especially noticeable when the agent sends intermediate progress
+            # messages like "Checking:" before running tools).
+            try:
+                await self.send_typing(chat_id, metadata=metadata)
+            except Exception:
+                pass  # Typing failures are non-fatal
+
             return SendResult(
                 success=True,
                 message_id=message_ids[0] if message_ids else None,