/ultrareview: free run consumed when cloud session crashes from upstream rate limiting
What happened
Ran /ultrareview on a branch (Free ultrareview 1 of 3). The cloud session was dispatched and "Remote review completed" — but with the error:
API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
No findings were produced. Despite the crash, the free-run counter decremented: the follow-up prompt confirmed "you have 2 free runs left." So a server-side rate limit (explicitly not my usage quota, per the error message itself) burned one of my three free monthly runs.
Expected
When the cloud job fails because of upstream/infra rate limiting (not user quota, not user error), the free run should not be consumed — or it should be auto-credited back. The error message itself makes the distinction clear ("not your usage limit"), so the system already has enough signal to know this is on Anthropic's side.
Repro context
- Command:
/ultrareview(no args, local branch bundle) - Branch scope: 235 files changed, 41895 insertions(+), 2339 deletions(-)
- Session ID:
011CvAxiNaV9VmHuywJWYbrK - Track URL: https://claude.ai/code/session_011CvAxiNaV9VmHuywJWYbrK?from=cli
- "Cooked for 2s" before failing — failure was near-immediate, suggesting upstream throttling rather than mid-review work.
Suggestion
Either:
- Don't decrement the free-run counter when the cloud session exits with an upstream rate-limit / server error (any non-user-caused failure), or
- Surface a "report run as failed → credits back" affordance in the post-crash prompt that already says "you have 2 free runs left."
Right now there's no way to recover the run, and re-running risks hitting the same backpressure window and burning another.
Environment
- Claude Code CLI
- macOS Darwin 25.3.0
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗