fix(gateway): gateway sessions lack activity watchdog - agent loop stalls silently with no detection and notification

agent/agent_init.py

@@ -33,6 +33,7 @@
 from agent.context_compressor import ContextCompressor
 from agent.iteration_budget import IterationBudget
 from agent.memory_manager import StreamingContextScrubber
+from agent.session_activity import ActivityProvenance
 from agent.model_metadata import (
     MINIMUM_CONTEXT_LENGTH,
     fetch_model_metadata,
@@ -882,6 +883,11 @@ def init_agent(
     # notifications to show progress.
     agent._last_activity_ts: float = time.time()
     agent._last_activity_desc: str = "initializing"
+    # Default / unmigrated paths and _touch_activity stamp unknown; named
+    # provenances are stamped by compression writers (heartbeat / timeout / cooldown).
+    agent._last_activity_provenance = ActivityProvenance.UNKNOWN
+    # Rate-limit durable SessionDB activity stamps from _touch_activity (#72016).
+    agent._session_activity_last_persist_mono: float = 0.0
     agent._current_tool: str | None = None
     agent._api_call_count: int = 0
     # Opt-out flag for the between-turns MCP tool refresh (build_turn_context).

agent/context_compressor.py

@@ -1916,6 +1916,24 @@ def _record_compression_failure_cooldown(
             self._cooldown_persist_failed = True
             logger.debug("compression failure cooldown persist failed (non-sqlite): %s", exc)
 
+    def record_timeout_failure(self, error: str) -> None:
+        """Record a consecutive timeout failure using the shared cooldown ladder.
+
+        Used by both the summary-LLM exception handler (inline at line ~3714)
+        and the host-level ``compress_context`` timeout wrapper in
+        ``run_compress_context_with_progress_timeout``. Avoids re-implementing
+        the ladder at each call site (#62452).
+        """
+        _TIMEOUT_COOLDOWN_LADDER = (60, 300, 900)
+        self._consecutive_timeout_failures = (
+            getattr(self, "_consecutive_timeout_failures", 0) + 1
+        )
+        cooldown = _TIMEOUT_COOLDOWN_LADDER[
+            min(self._consecutive_timeout_failures,
+                len(_TIMEOUT_COOLDOWN_LADDER)) - 1
+        ]
+        self._record_compression_failure_cooldown(float(cooldown), error)
+
     def _clear_compression_failure_cooldown(self) -> None:
         self._summary_failure_cooldown_until = 0.0
         self._last_summary_error = None