Environment

• hermes-agent: 0.12.0
• Python: 3.13.5
• aiohttp: 3.13.5
• Platform: WeChat (iLink)

Problem Description

When a cron job tries to deliver a message to WeChat, the send fails with the following error in gateway.log:

Timeout context manager should be used inside a task
[Weixin] send chunk failed ... attempt=1/5

The same WeChat adapter works perfectly when the user interacts in the active gateway session, but fails when triggered by a cron job or a delegate_task subagent.

Root Cause Analysis

The issue is a thread/event-loop mismatch between the gateway's main async context and the cron scheduler's async invocation:

1. WeixinAdapter.__init__ (gateway/platforms/weixin.py:1219) creates self._send_session = aiohttp.ClientSession(...) in the gateway's main async event loop (when the user first connects via WeChat). This session is stored in the global _LIVE_ADAPTERS dict keyed by token.

2. send_weixin_direct (gateway/platforms/weixin.py:2031-2038) has a "live adapter" fast path that preferentially uses this existing session:

   live_adapter = _LIVE_ADAPTERS.get(resolved_token)
   send_session = getattr(live_adapter, '_send_session', None)
   if live_adapter is not None and send_session is not None and not send_session.closed:
       last_result = await live_adapter.send(chat_id, cleaned)  # <- called from wrong loop

3. cron/scheduler.py:481 runs the delivery coroutine in a fresh event loop via asyncio.run(coro) - inside a ThreadPoolExecutor on the fallback path (line 488-490). This creates a new thread with a new event loop that has no awareness of the gateway's loop.

4. When live_adapter.send() executes HTTP requests, aiohttp 3.13.5 detects that the session's internal state (SSL context, socket handles) is tied to the original gateway loop and throws:

   Timeout context manager should be used inside a task

Expected Behavior

Cron jobs and subagents should be able to send WeChat messages without errors.

Suggested Fix

In send_weixin_direct, the live adapter fast path should detect whether it is being called from the same event loop as the one the session was created in. If not (i.e., running in a cron/delegate context), it should skip the live adapter path and fall through to the standalone aiohttp.ClientSession() path, which creates a fresh, self-contained session per call.

A possible implementation - check whether we are in an asyncio.run() thread vs. the gateway's main loop, and fall through to the standalone path if not:

# Pseudocode for gateway/platforms/weixin.py around line 2031
import asyncio
import threading

live_adapter = _LIVE_ADAPTERS.get(resolved_token)
send_session = getattr(live_adapter, '_send_session', None)

# Only use live adapter if we are in the same event loop context.
# Skip the live path if we are in asyncio.run() from a different thread
# (cron scheduler uses ThreadPoolExecutor + asyncio.run, creating a new loop).
use_live = (
    live_adapter is not None
    and send_session is not None
    and not send_session.closed
)
if use_live:
    try:
        current_loop = asyncio.get_running_loop()
        # If we got here from a ThreadPoolExecutor thread with its own
        # asyncio.run() loop, current_loop differs from the gateway loop -
        # fall through to standalone path in that case.
    except RuntimeError:
        use_live = False  # No running loop, fall through

if use_live:
    # ... live adapter path (existing code)
else:
    # ... standalone path (existing async with ClientSession() block)

Alternatively, the fix could be in cron/scheduler.py to ensure cron delivery always uses the standalone session path for WeChat.

Additional Context

• last_run_at: null confirms cron jobs never successfully delivered a WeChat message
• Manual trigger of cron job updates next_run_at but does not execute the agent
• The error is 100% reproducible on every WeChat cron delivery attempt