pywf-oss-fix-ci

Find from the logs why a hosted service is red, fix the cause in the repo, and confirm it is green again. Also add or drop a Python version.

Read ../pywf-oss-base/SKILL.md, then:

• ../pywf-oss-base/references/04-python-versions.md
• ../pywf-oss-base/references/19-github-actions.md
• ../pywf-oss-base/references/20-codecov.md
• ../pywf-oss-base/references/21-readthedocs.md
• ../pywf-oss-base/references/25-python-release-cycle.md (only for adding or dropping a Python version)

Steps

1. Find the red service and its log with the commands in 19-github-actions.md, 20-codecov.md, or 21-readthedocs.md. Quote the decisive log lines to the user.
2. Reproduce locally where possible: mise run cov for tests, mise run build-doc for docs. If it passes locally, compare OS and Python version, the installed extras (uv sync --extra test in CI, --all-extras locally), and the working directory.
3. Fix the cause, not the symptom.
   • Never skip, xfail, or delete a failing test to turn CI green without the user's yes.
   • Never narrow the matrix permanently to hide a failure.
4. Verify. Rerun locally. Then, after the user's yes, push the fix, or for an infrastructure flake run gh run rerun <run_id> --failed. Check the new run with gh run view.

Boundaries

• Reading logs and runs needs no approval. Pushing, rerunning, deleting caches, and rerunning setup tasks each need the user's yes.

Next step: pywf-oss-upgrade-deps when the cause is a dependency, or when a Python version change lifts or adds dependency caps; pywf-oss-changelog to record a Python version change.