hermes - 💡(How to fix) Fix QQ Bot: WebSocket reconnect causes approval deadlock (session mismatch)

Environment

• Platform: QQ Bot
• Hermes version: latest
• OS: WSL2 Ubuntu 24.04

Bug Description

When QQ Bot WebSocket disconnects (code=4009 Session timed out, which QQ servers enforce ~every 30 minutes), the gateway reconnects automatically — but old session's pending approval requests are rejected in the new session due to session mismatch, creating an infinite deadlock.

Reproduction

1. QQ Bot connects and is processing normally
2. QQ server drops WebSocket with code=4009 (normal server-side timeout)
3. Gateway reconnects with new WebSocket session
4. Any pending approval request from the old session is rejected:

   Rejected unauthorized approval click for session agent:main:qqbot:dm:XXX (operator=XXX)

5. The agent enters a loop: command waits 295s for user approval → timeout → retry → rejected again → infinite loop

Logs

WARNING gateway.platforms.qqbot.adapter: [QQBot:1904121396] WebSocket closed: code=4009 reason=Session timed out
...
WARNING gateway.platforms.qqbot.adapter: [QQBot:1904121396] Rejected unauthorized approval click for session agent:main:qqbot:dm:A37B46B0A04E3326D727DFD81FCFEBA9
WARNING agent.tool_executor: Tool terminal returned error (295.58s): BLOCKED: Command timed out without user response

Expected Behavior

On WebSocket reconnect, pending approvals from the old session should either be:

• Auto-cleared/cancelled, OR
• Transferred to the new session, OR
• At minimum, not cause an infinite retry loop

Workaround

Currently using a watchdog script that SIGKILLs the gateway when the deadlock pattern is detected, relying on systemd auto-restart:

journalctl --user -u hermes-gateway.service -n 50 | grep -q "Rejected unauthorized approval click" && systemctl --user kill -s SIGKILL hermes-gateway.service