fix(gateway): prevent concurrent agent runs for the same session

Place a sentinel in _running_agents immediately after the "already
running" guard check passes — before any await.  Without this, the
numerous await points between the guard (line 1324) and agent
registration (track_agent at line 4790) create a window where a
second message for the same session can bypass the guard and start
a duplicate agent, corrupting the transcript.

The await gap includes: hook emissions, vision enrichment (external
API call), audio transcription (external API call), session hygiene
compression, and the run_in_executor call itself.  For messages with
media attachments the window can be several seconds wide.

The sentinel is wrapped in try/finally so it is always cleaned up —
even if the handler raises or takes an early-return path.  When the
real AIAgent is created, track_agent() overwrites the sentinel with
the actual instance (preserving interrupt support).

Also handles the edge case where a message arrives while the sentinel
is set but no real agent exists yet: the message is queued via the
adapter's pending-message mechanism instead of attempting to call
interrupt() on the sentinel object.

gateway/run.py

@@ -222,6 +222,12 @@ from gateway.platforms.base import BasePlatformAdapter, MessageEvent, MessageTyp
 
 logger = logging.getLogger(__name__)
 
+# Sentinel placed into _running_agents immediately when a session starts
+# processing, *before* any await.  Prevents a second message for the same
+# session from bypassing the "already running" guard during the async gap
+# between the guard check and actual agent creation.
+_AGENT_PENDING_SENTINEL = object()
+
 
 def _resolve_runtime_agent_kwargs() -> dict:
     """Resolve provider credentials for gateway-created AIAgent instances."""
@@ -1346,7 +1352,14 @@ class GatewayRunner:
                         adapter._pending_messages[_quick_key] = event
                 return None
 
-            running_agent = self._running_agents[_quick_key]
+            running_agent = self._running_agents.get(_quick_key)
+            if running_agent is _AGENT_PENDING_SENTINEL:
+                # Agent is being set up but not ready yet — queue the message
+                # so it will be picked up after the agent starts.
+                adapter = self.adapters.get(source.platform)
+                if adapter:
+                    adapter._pending_messages[_quick_key] = event
+                return None
             logger.debug("PRIORITY interrupt for session %s", _quick_key[:20])
             running_agent.interrupt(event.text)
             if _quick_key in self._pending_messages:
@@ -1354,7 +1367,7 @@ class GatewayRunner:
             else:
                 self._pending_messages[_quick_key] = event.text
             return None
-        
+
         # Check for commands
         command = event.get_command()
         
@@ -1527,7 +1540,29 @@ class GatewayRunner:
         # Pending exec approvals are handled by /approve and /deny commands above.
         # No bare text matching — "yes" in normal conversation must not trigger
         # execution of a dangerous command.
-        
+
+        # ── Claim this session before any await ───────────────────────
+        # Between here and _run_agent registering the real AIAgent, there
+        # are numerous await points (hooks, vision enrichment, STT,
+        # session hygiene compression).  Without this sentinel a second
+        # message arriving during any of those yields would pass the
+        # "already running" guard and spin up a duplicate agent for the
+        # same session — corrupting the transcript.
+        self._running_agents[_quick_key] = _AGENT_PENDING_SENTINEL
+
+        try:
+            return await self._handle_message_with_agent(event, source, _quick_key)
+        finally:
+            # If _run_agent replaced the sentinel with a real agent and
+            # then cleaned it up, this is a no-op.  If we exited early
+            # (exception, command fallthrough, etc.) the sentinel must
+            # not linger or the session would be permanently locked out.
+            if self._running_agents.get(_quick_key) is _AGENT_PENDING_SENTINEL:
+                del self._running_agents[_quick_key]
+
+    async def _handle_message_with_agent(self, event, source, _quick_key: str):
+        """Inner handler that runs under the _running_agents sentinel guard."""
+
         # Get or create session
         session_entry = self.session_store.get_or_create_session(source)
         session_key = session_entry.session_key