Summary

A scheduled agent turn using the Codex app-server runtime can fail at the internal terminal-idle watchdog even when the scheduled job already has a longer run-specific timeout configured. The surfaced error currently tells users to increase agents.defaults.timeoutSeconds, but that setting is not the actual limiting timeout in this failure mode.

Observed

• OpenClaw version: 2026.5.18.
• A scheduled agentTurn using codex/gpt-5.5 had timeoutSeconds set above 30 minutes.
• The run failed after roughly 30 minutes with the internal log message:
  • codex app-server turn idle timed out waiting for terminal event
  • timeoutMs=1800000
• The user-facing error was:
  • Request timed out before a response was generated. Please try again, or increase agents.defaults.timeoutSeconds in your config.

Expected

• The Codex app-server terminal-idle watchdog should be configurable or derived from the effective run timeout ceiling.
• If this watchdog fires, diagnostics should name the actual limiting timeout instead of pointing only to agents.defaults.timeoutSeconds.
• A run-specific timeout should not be silently superseded by a lower hardcoded internal Codex terminal-idle timeout.

Local mitigation used

A local hotfix made the Codex terminal-idle timeout configurable via an environment variable and increased the affected scheduled job timeout so the outer job budget is higher than the internal idle watchdog.

This issue is for upstreaming the behavior cleanly rather than relying on local patched bundled files.