CI Watch and Fix

You are running the "CI Watch and Fix" loop on the current branch.

Settings (edit before running):
- Branch or pull request: the current branch
- Max fix pushes: 5
- Poll interval: 60 seconds, give up on a single run after 30 minutes

Goal: every required CI check on the pull request reports success.

Each iteration:
1. Find the newest CI run for this branch. If it is still running, wait and poll at the interval above.
2. If any required job failed, open the first one and read its logs. Quote the step name and the 5 to 15 lines that show the error.
3. If the failing command can run locally, run it to confirm the cause.
4. Make the smallest fix, commit it with a message like "ci: fix <job name>: <cause>", and push to the same branch.
5. Go back to step 1 and wait for the new run.

Stop when:
- All required checks pass. Success.
- You have pushed the max number of fixes.
- A failure is caused by infrastructure, secrets, quotas, or a flaky external service rather than the code. Report it instead of changing code.

Rules:
- Only push to this branch. Never force-push, never push to the default branch, never merge.
- Do not edit CI workflow files to skip, disable, or mark jobs as allowed to fail.
- Do not print secret values from logs or environment variables.
- Re-running a job is allowed once per failure when the logs point to a flaky network step.

Final report: a table of iterations with job name, cause, fix commit, and resulting status.