openclaw - 💡(How to fix) Fix Stable 2026.5.7 still hits Codex app-server rawResponseItem/completed idle timeout after status output

Summary

A production Telegram forum-topic turn on stable 2026.5.7 hit the same Codex app-server completion-watchdog failure described in #77984: the assistant emitted a status/progress sentence, then the turn went internally quiet and was aborted after the 60s completion-idle watchdog with lastActivityReason="notification:rawResponseItem/completed".

I first tried to add this as a comment to #77984, but that issue is locked. This may simply be a stable-release/backport gap if the main-branch fix from c22f414 is intentionally unreleased in 2026.5.7.

Environment

• OpenClaw 2026.5.7 (eeef486)
• @openclaw/codex 2026.5.7
• @openai/codex 0.128.0
• Node v22.22.2
• Linux x64 (Ubuntu, kernel 6.8.0-111-generic)
• Runtime: Codex app-server / agentRuntime.id=codex
• Model route: openai/gpt-5.5 via OpenAI Codex OAuth profile
• Channel: Telegram forum topic

Steps to reproduce from the observed flow

1. Use OpenClaw stable 2026.5.7 with @openclaw/codex 2026.5.7 and Codex runtime enabled.
2. In a Telegram forum topic, send a normal request that requires a short lookup/tool phase and then generation of a visual HTML/file report.
3. Let the assistant emit a status/progress sentence after the lookup, before the file-generation step.
4. Observe the turn after roughly 60 seconds of no further current-turn app-server events.

Observed behavior

User-visible sequence:

1. User requested a visual HTML report from previous analysis.
2. The turn started and performed a memory_search tool call successfully.
3. The assistant produced a status/progress sentence saying it had the report data and was assembling the HTML.
4. No file/report was delivered.
5. The session ended as an aborted assistant message with errorMessage="codex app-server attempt timed out".

Trajectory evidence from the affected run:

{
  "type": "tool.call",
  "ts": "2026-05-11T11:41:25.318Z",
  "data": {
    "name": "memory_search"
  }
}

{
  "type": "tool.result",
  "ts": "2026-05-11T11:41:26.266Z",
  "data": {
    "name": "memory_search",
    "success": true
  }
}

{
  "type": "turn.completion_idle_timeout",
  "ts": "2026-05-11T11:44:01.240Z",
  "data": {
    "idleMs": 60008,
    "timeoutMs": 60000,
    "lastActivityReason": "notification:rawResponseItem/completed"
  }
}

{
  "type": "model.completed",
  "ts": "2026-05-11T11:44:01.253Z",
  "data": {
    "timedOut": true,
    "yieldDetected": false,
    "aborted": true,
    "promptError": "codex app-server attempt timed out",
    "assistantTexts": [
      "[redacted prefix] I have the report data now: monthly trend, property workload, keyword/problem signals, time-of-day pattern, and specific evidence links. I’m assembling the HTML with inline SVG/CSS charts so it opens cleanly in Telegram."
    ]
  }
}

{
  "type": "session.ended",
  "ts": "2026-05-11T11:44:01.253Z",
  "data": {
    "status": "error",
    "timedOut": true,
    "yieldDetected": false,
    "promptError": "codex app-server turn idle timed out waiting for turn/completed"
  }
}

Gateway journal at the same time:

2026-05-11T11:44:01.245Z [agent/embedded] codex app-server turn idle timed out waiting for completion
2026-05-11T11:44:01.367Z [agent/embedded] Profile openai-codex:<redacted> timed out. Trying next account...
2026-05-11T11:44:01.374Z [agent/embedded] embedded run failover decision: stage=assistant decision=surface_error reason=timeout from=openai/gpt-5.5 rawError=codex app-server attempt timed out

Expected behavior

On stable releases that include the #77984 fix, this should match the behavior described in c22f414: same-turn non-terminal activity, including rawResponseItem/completed, should not leave the short completion watchdog armed in a way that kills ordinary report/file-generation turns after a status sentence.

If 2026.5.7 predates that fix, the actionable request is to make the release/backport status explicit, because the public closed issue reads as fixed while the current stable release still has the user-visible failure.

Related issues / commits

• #77984 — exact rawResponseItem/completed watchdog issue, closed/locked as fixed on main
• #78756 — broader Codex app-server timeout behavior
• c22f414c6976f66652aa7174838b7121673e810f — main-branch fix noted in #77984