Automatic continuation loop consumed tokens without making progress

Summary

@tibo-openai, I encountered a runaway automatic-continuation loop in a long-running Codex task.

After making real progress on a documentation-localization goal, the task repeatedly emitted only 继续处理中 (“continuing”) without editing files or making meaningful progress. The empty continuation turns continued to consume tokens and time, and the goal was eventually marked blocked.

Expected behavior

• A long-running goal should resume actual work, or visibly pause/stop with a clear reason.
• Empty automatic continuation turns should not continue consuming tokens.
• Users need a reliable way to stop a runaway continuation loop.

Impact

This wasted a substantial number of tokens and significant time.

Please investigate whether this is a task-continuation scheduling/state bug, and whether usage consumed by these empty turns can be restored.