Troubleshooting¶
"Missing required configuration"¶
The CLI lists every missing setting. In CI, make sure the job runs in a merge request pipeline (rules: - if: $CI_PIPELINE_SOURCE == "merge_request_event"), otherwise CI_MERGE_REQUEST_IID is not set.
"Failed to load merge request from GitLab: 401/403/404"¶
- The token needs the
apiscope and at least Developer access on the project. - Masked protected variables are not available on unprotected branches.
- On self-managed GitLab, check
--gitlab-api-urlends in/api/v4.
No comments appear¶
- The model may have answered
204for every hunk. Look for "No Feedback Summary" on the MR. - Renamed, deleted and binary files are skipped.
- Check the job log for
Failed to review <file>lines.
AWS Bedrock¶
- "The provided model identifier is invalid" / "on-demand throughput isn't supported": use an inference profile ID (
us.anthropic.…). See Bedrock. - AccessDeniedException: enable model access in the Bedrock console and check the IAM policy.
- Throttling: the client retries 429/5xx with exponential backoff. For large MRs, request a quota increase or use a smaller model.
Gemini¶
- 404 model not found: older models (for example
gemini-1.5-*) have been retired. SetGEMINI_MODELto a current stable model. - 400 API key not valid: create a new key in Google AI Studio.
Rate limits (any provider)¶
On HTTP 429 the tool waits 60 seconds and re-queues the hunk, so very large MRs can take a while. Consider when: manual for huge MRs.