All 5-hour Pro tokens consumed by /goal looping on "Error running remote compact task"

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.513.31313 (2867)

What subscription do you have?

Pro

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

<img width="797" height="1028" alt="Image" src="https://github.com/user-attachments/assets/48c5344e-7a75-4653-a47c-46adf83d8824" />

A chat with /goal got stuck and consumed my full 5-hour Pro token allowance in less than 45 minutes, with /fast on.

This bug has been lingering for a while, and the team seems to be refusing to fix it. Their rationale appears to be: "We've got to fix the underlying timeout issue." But until the underlying timeout issue is resolved, you still need to fix the impact it is having today, and has been having for the last few weeks.

A note for the team: whenever gpt-5.4 or gpt-5.5 fails to compact, gpt-5.4-mini succeeds. Perhaps add a fallback where, if compacting fails, it retries with gpt-5.4-mini? I do not know what the right answer is, but this needs to be fixed sooner rather than later: both the underlying issue and the surface-level issue.

What steps can reproduce the bug?

1. Start a new chat session with a /goal that involves a task requiring compaction (e.g. a long-running coding task).
2. Enable /fast mode.
3. Let the session run — observe it looping on "Error running remote compact task".
4. The session consumes the full 5-hour Pro token budget in under 45 minutes without completing the goal.

What is the expected behavior?

When compaction fails (e.g. gpt-5.4 or gpt-5.5 cannot compact), the session should either gracefully fall back to a smaller model (e.g. gpt-5.4-mini) for the compaction step, or surface a clear error to the user and stop consuming tokens — rather than looping indefinitely until the token budget is exhausted.

Additional information

_No response_