openclaw - 💡(How to fix) Fix Codex cron agentTurn regression: app-server attempt/turn idle timeout after 2026.5.27 [1 pull requests]

Summary

After updating to OpenClaw 2026.5.27 (27ae826), a previously reliable scheduled Codex agentTurn cron began failing consistently after tool-use continuation with Codex app-server timeout errors.

This appears to be a regression in Codex app-server turn/attempt completion handling for longer cron agent turns. The same scheduled job completed successfully before the update with similar or higher token usage. After the update, it fails even when the workload is reduced to a single selected item.

Related but not identical: #85526. That issue focuses on side effects being completed before final artifact delivery. In this case the run can fail before producing or persisting the intended output, even after reducing scope.

Environment

• OpenClaw: 2026.5.27 (27ae826)
• Codex plugin/package: @openclaw/[email protected]
• Runtime/provider: openai-codex
• Model: gpt-5.5
• Job type: scheduled cron agentTurn
• Session target: isolated
• Payload: long-running tool-using agent task with lightweight context enabled

Observed Behavior

The cron begins normally, runs several shell/tool calls, then terminates with errors like:

Codex stopped before confirming the turn was complete. Some work may already have been performed; verify the current state before retrying.

Trajectory diagnostics observed on failed runs include:

codex app-server attempt timed out
codex app-server turn idle timed out waiting for turn/completed

The cron state is marked error, delivery is not-delivered, and the intended output is not written.

Regression Signal

Before updating to 2026.5.27, the same cron shape completed successfully with similar or higher total token usage. After updating, it fails consistently.

Recent run pattern:

• Pre-update successful run: completed and delivered, roughly 103k total tokens.
• Post-update scheduled run: failed after about 202s, roughly 98k total tokens.
• Post-update manual retry: failed after about 189s, roughly 86k total tokens.
• Post-update reduced-scope retry: failed after about 135s, roughly 53k total tokens.

The reduced-scope retry changed the task prompt to process only one selected item, avoid dumping large arrays, and keep final output concise. It still failed with the same app-server completion/attempt timeout class, so this does not appear to be only a batch-size or prompt-noise issue.

Expected Behavior

A scheduled Codex agentTurn that is actively making tool progress should either:

• complete and emit/deliver the final assistant response, or
• fail with a more specific actionable diagnostic that identifies which timeout was exceeded and how to configure or avoid it.

If a new app-server attempt timeout or terminal-idle behavior was introduced or tightened in 2026.5.27, it should be documented and ideally tunable for cron/isolated agent turns.

Notes

I intentionally omitted local paths, business data, prompt contents, and session IDs from this public report. I can provide sanitized excerpts or private details if maintainers need them.