openclaw - 💡(How to fix) Fix [Bug]: Discord channel message marked completed while embedded run aborted, then session stalls

Summary

On OpenClaw 2026.5.12 (f066dd2), a Discord channel turn can be marked message.processed outcome=completed even though the embedded run ended aborted=true. After that, the same channel session can remain in stalled_agent_run with activeWorkKind=embedded_run, eventually surfacing the generic user-facing timeout text:

Request timed out before a response was generated. Please try again, or increase `agents.defaults.timeoutSeconds` in your config.

This does not appear to be a simple agents.defaults.timeoutSeconds issue. The observed run aborted after ~8s, the message was marked completed, then the session was later reported as stalled.

Environment

• OpenClaw CLI/Gateway: 2026.5.12 (f066dd2)
• Runtime: macOS LaunchAgent, Node 22 via Homebrew
• Channel: Discord channel session
• Model: openai/gpt-5.5
• Relevant defaults at time of mitigation:
  • agents.defaults.timeoutSeconds=1800
  • agents.defaults.contextTokens=120000
  • active-memory.allowedChatTypes=["direct"] (channel active-memory disabled after mitigation)

Evidence

Incident:

• Discord message id: 1504785435535081615
• Timestamp: 2026-05-15T18:00:05+08:00
• Session key: agent:tino:discord:channel:1480145267398545594
• Session id: 0b401b0a-47dc-4091-84b3-62eae77f7c63

Gateway log sequence:

2026-05-15T18:00:05.741+08:00 DEBUG message queued: sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 sessionKey=agent:tino:discord:channel:1480145267398545594 source=dispatch queueDepth=1 sessionState=idle
2026-05-15T18:00:07.730+08:00 DEBUG run registered: sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 totalActive=1
2026-05-15T18:00:15.059+08:00 DEBUG session state: sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 sessionKey=agent:tino:discord:channel:1480145267398545594 prev=processing new=idle reason="run_completed" queueDepth=0
2026-05-15T18:00:15.060+08:00 DEBUG run cleared: sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 totalActive=0
2026-05-15T18:00:15.066+08:00 DEBUG embedded run done: runId=d2a9e294-1826-4610-8add-fb7d52b57653 sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 durationMs=8376 aborted=true
2026-05-15T18:00:16.035+08:00 DEBUG message processed: channel=discord chatId=channel:1480145267398545594 messageId=1504785435535081615 sessionId=unknown sessionKey=agent:tino:discord:channel:1480145267398545594 outcome=completed duration=10302ms
2026-05-15T18:03:13.813+08:00 WARN stalled session: sessionId=0b401b0a-47dc-4091-84b3-62eae77f7c63 sessionKey=agent:tino:discord:channel:1480145267398545594 state=processing age=145s queueDepth=1 reason=active_work_without_progress classification=stalled_agent_run activeWorkKind=embedded_run lastProgress=codex_app_server:notification:rawResponseItem/completed lastProgressAge=144s terminalProgressStale=true recovery=none

Key inconsistency:

• Embedded run ended with aborted=true
• Channel message was still marked outcome=completed
• Later diagnostics classified the same session as stalled_agent_run
• Recovery remained none

Expected behavior

One of these should happen:

1. If the embedded run aborts, the Discord message should not be recorded as successfully completed unless a final user-visible response was actually sent.
2. If a session is detected as stalled_agent_run activeWorkKind=embedded_run, the runtime should either abort/recover the run or clear the session state safely.
3. The user-facing error should distinguish "wall-clock timeout" from "embedded run aborted / stale session state" so operators are not misled into only increasing agents.defaults.timeoutSeconds.

Actual behavior

The channel turn was treated as completed, then the session remained/returned to a stalled embedded-run state and eventually produced the generic timeout text.

Local mitigation used

We mitigated locally by:

• Rotating only the affected Discord channel session-key mapping after backing up the session store.
• Preserving the transcript/session files.
• Restarting Gateway to reload in-memory state.
• Adding a watchdog that detects stalled_agent_run + activeWorkKind=embedded_run for Discord channel sessions and performs the same targeted rotation/restart.

This restored the channel, but it is a workaround rather than a native fix.

Diagnostics bundle

A redacted trajectory export was generated locally with:

openclaw sessions export-trajectory \
  --agent tino \
  --store /Users/betalpha/.openclaw/agents/tino/sessions/sessions.json.bak-timeoutfix-1778842350 \
  --session-key agent:tino:discord:channel:1480145267398545594 \
  --workspace /Users/betalpha/clawd/tino \
  --output discord-stall-2026-05-15-1504785435535081615 \
  --json

The export contains 465 total events: 199 runtime events and 266 transcript events. I can provide the redacted export if useful.

Related issues that look adjacent but not identical:

• #71127
• #73496
• #74586