What to do#
Read the task’s latest state and let the underlying Codex workflow handle the capacity retry when applicable. Do not send the same instruction again merely because a retry state appears. Repetition can create duplicate work once capacity returns.
If you need to change the request, inspect the original conversation before adding a new instruction. Distinguish a model-capacity delay from a failed message submission; the former may already have your request.
Timing limits#
The reviewed materials do not establish an exact reliable retry countdown in CueSpec. A visible waiting state is useful, but a fabricated number of seconds would be misleading. Similarly, a timeout in the companion should not be interpreted as the task having completed elsewhere.
A confirmed delivery and later capacity wait can coexist. The receipt answers “did the message arrive?” The task state answers “what is happening with the work?” If the state stays unclear, retain the latest visible message and version details for support.