| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
Mutable tags are movable pointers: whoever controls an action, or anyone who compromises it, can re-point the tag and the next run executes unreviewed code with the job's token, including in the release workflow that publishes to PyPI. Same pattern as the tj-actions/changed-files incident (CVE-2025-30066). Pinning to the full commit sha freezes what runs; the version each sha corresponds to stays as a comment. Every sha was resolved from the upstream repository and cross-checked against its release tag. No workflow logic changes.
Without a permissions block, every job's GITHUB_TOKEN inherits the repository default scope, which is broader than these workflows need. All six jobs only check out, test, or publish through dedicated secrets (codecov token, PyPI token), the GitHub token itself is never used. Everything gets contents read.
Scans the workflows on each push to master and on pull requests, and fails when something regresses: an unpinned action, a job without a permissions block, an archived dependency, a known CVE. It runs on the CLI's built-in default configuration, no config file needed. Each run publishes the score to score.getplumber.io, shown as a badge in the README.
| Back | FazBrowse Home | New Git URL |
Hi, drive-by CI hardening in three commits, one logical change each.
Commit 1 pins every action to its commit sha, versions kept as comments. A tag like @v1 is a movable pointer: whoever controls the action, or anyone who compromises it, can re-point it and your next run executes their code with the job's token, including in release.yaml which publishes to PyPI. Same pattern as the tj-actions/changed-files incident (CVE-2025-30066). All shas were resolved from the upstream repos and cross checked against their release tags. The pins stay maintainable: if you want them bumped automatically, a dependabot config with the github-actions ecosystem does it, one small block. Worth knowing: codecov/codecov-action is still on v1, bumping it to a current major would be a good separate change, this PR only freezes what you already run.
Commit 2 adds permissions: contents: read to both workflows. The scope is exact, not guessed: all six jobs only check out, test, or publish through dedicated secrets (codecov token, PyPI token), the GitHub token itself is never used.
Commit 3 adds Plumber to CI, the tool I used to find the issues in the first place, so none of this quietly drifts back:
One heads up: if the org uses an Actions allowlist in settings, patterns written against tags (like owner/action@v1) stop matching once refs are shas and workflows refuse to start. Entries need to be owner/action@* in that case.
To be fully transparent: I work on Plumber. If you do not want the tool in your CI, no hard feelings, say so and I remove that commit from the PR, the two hardening commits are the part that matters and they stand entirely on their own.