fix(cron): close codex app-server session in teardown to prevent subprocess leak (#62101)

agent/codex_runtime.py

@@ -922,9 +922,32 @@ def run_codex_create_stream_fallback(agent, api_kwargs: dict, client: Any = None
     return run_codex_stream(agent, api_kwargs, client=client)
 
 
+def close_codex_session(agent) -> None:
+    """Close the Codex app-server session if it exists.
+
+    Ephemeral agents (e.g., cron jobs) need explicit cleanup to avoid
+    leaking the ``codex app-server`` subprocess. The session is closed on
+    the turn-crash, ``should_retire``, and compression-drop paths, but
+    cron's ``finally`` block tears down the agent via ``agent.close()``,
+    which does not include this cleanup.
+
+    This helper is safe to call multiple times — ``CodexAppServerSession.close()``
+    is idempotent.
+    """
+    session = getattr(agent, "_codex_session", None)
+    if session is None:
+        return
+    try:
+        session.close()
+    except Exception:
+        logger.debug("codex app-server session close failed", exc_info=True)
+    agent._codex_session = None
+
+
 __all__ = [
     "run_codex_app_server_turn",
     "run_codex_stream",
     "run_codex_create_stream_fallback",
     "_consume_codex_event_stream",
+    "close_codex_session",
 ]

cron/scheduler.py

@@ -3336,6 +3336,14 @@ def _teardown_cron_agent(agent, job_id: str) -> None:
             agent.close()
     except (Exception, KeyboardInterrupt) as e:
         logger.debug("Job '%s': failed to close agent resources: %s", job_id, e)
+    # Close the Codex app-server session if it exists. Ephemeral agents
+    # (cron jobs) leak the subprocess without this cleanup (#62101).
+    try:
+        from agent.codex_runtime import close_codex_session
+        if agent is not None:
+            close_codex_session(agent)
+    except (Exception, KeyboardInterrupt) as e:
+        logger.debug("Job '%s': failed to close codex session: %s", job_id, e)
     # Each cron run spins up a short-lived worker thread whose event loop
     # dies as soon as the ``ThreadPoolExecutor`` shuts down. Any async
     # httpx clients cached under that loop are now unusable — reap them