Problem

When the cron scheduler's live adapter delivery fails (e.g. ret=-2 network error), it falls back to a standalone path that runs _send_to_platform inside asyncio.run() in a thread-pool executor — a fresh event loop.

send_weixin_direct then finds the live adapter in _LIVE_ADAPTERS and unconditionally reuses its aiohttp.ClientSession, which was created on the gateway event loop. Using that session from a different loop causes aiohttp's internal asyncio.timeout() to raise:

Timeout context manager should be used inside a task

because asyncio.timeout() (Python 3.11+) checks asyncio.current_task(), and while a Task does exist on the new loop, the connector is bound to the old one — the mismatch triggers the assertion.

The result is three retried failures followed by a delivery error log, and the cron message never reaches the user.

Fix

Before reusing the live adapter session in send_weixin_direct, compare connector._loop against asyncio.get_running_loop(). If they differ, skip the live adapter branch and fall through to the existing fresh-session path (async with aiohttp.ClientSession(...)).

The live adapter fast-path is preserved for all callers that are already on the correct loop (e.g. the /send_message tool running inside the gateway).

Test Plan

• pytest tests/gateway/test_weixin.py — 42 passed, 0 failures
• Reproduces via: cron job delivery where first live-adapter attempt returns ret=-2, triggering the standalone fallback path