What does this PR do?

Fix the long-standing RuntimeError: Timeout context manager should be used inside a task raised by the Weixin adapter whenever the agent uses the send_message tool to send media (files, images, voice, video).

Root cause (verified with runtime debug logging on macOS, Python 3.11.15, aiohttp 3.13.5):

When the agent invokes the send_message tool, model_tools._run_async() detects an active event loop in the gateway's main thread and dispatches the call to a worker thread that calls asyncio.run(). This creates a fresh event loop in the worker thread. Inside that loop, _send_to_platform → WeixinAdapter.send_document → _send_file → _api_post reuses self._send_session, which is an aiohttp.ClientSession bound to the gateway's main loop. aiohttp's TimerContext.__enter__ then calls asyncio.current_task() against the session's original loop, gets None, and raises.

The bug only fires for media sends, because the gateway delivers normal text replies inline on its own loop — that path never crosses loops.

Debug capture from a failing call:

[WEIXIN-DBG] task=<Task-33 coro=<_send_to_platform() at send_message_tool.py:507>
             cb=[_run_until_complete_cb()]>
             loop=<_UnixSelectorEventLoop ...>
             session_loop=<_UnixSelectorEventLoop ...>
             thread='asyncio_1'                       ← worker thread, fresh loop
ERROR: Timeout context manager should be used inside a task

loop and session_loop reprs look identical but are different instances (one in MainThread, one in asyncio_1).

Fix: add WeixinAdapter._loop_safe_session(), an async context manager that swaps in a per-call aiohttp.ClientSession bound to the running loop when it detects the cached session belongs to a different loop. Each send_* method wraps its body with async with self._loop_safe_session():. When the call is on the original loop the helper is a no-op (no new session, no extra connection).

Related Issue

Refs #13099, #12154, #13281, #16293

(Partial fix — addresses the immediate Weixin symptom in all four issues. The deeper model_tools._run_async cross-loop architecture concern raised in #13281 is left for a separate change.)

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• gateway/platforms/weixin.py:
  • Add import contextlib
  • Add WeixinAdapter._loop_safe_session() async context manager
  • Wrap method bodies with async with self._loop_safe_session(): in send, send_image, send_image_file, send_document, send_video, send_voice
  • send_typing is left untouched (gateway-internal, never crosses loops)
• tests/gateway/test_weixin.py: add TestWeixinLoopSafeSession regression class with three cases (foreign-loop session swap, same-loop pass-through, no-session no-op)

How to Test

1. pytest tests/gateway/test_weixin.py -q — 45 passed (42 existing + 3 new)
2. End-to-end repro on macOS:
   • Configure Weixin gateway, start hermes gateway, chat with the bot from WeChat
   • Ask the agent to send a file (e.g. "请把 README 发给我")
   • Before: [Weixin] send_document failed to=...: Timeout context manager should be used inside a task
   • After: file is delivered successfully

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(weixin): ...)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix (no unrelated commits)
• I've run pytest tests/gateway/test_weixin.py -q and all tests pass
• I've added tests for my changes
• I've tested on my platform: macOS 15.x, Python 3.11.15

Documentation & Housekeeping

• I've updated relevant documentation — N/A (internal behavior fix, no user-facing surface change)
• I've updated cli-config.yaml.example if I added/changed config keys — N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
• I've considered cross-platform impact — fix is OS-independent (the bug shows up on macOS but the cross-loop issue is platform-agnostic; Linux just happens to schedule timing-sensitive paths differently)
• I've updated tool descriptions/schemas if I changed tool behavior — N/A

Screenshots / Logs