Cerberus AI-Agent Runtime Check

Version updated for https://github.com/Odingard/cerberus-action to version v1.0.1.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

This GitHub Action checks Cerberus’s allow/block behavior by running two bundled, maintained fixtures through the Cerberus runtime security layer. It provides a PASS/FLAG/BLOCK verdict and an uploaded evidence artifact, ensuring CI runs without disruptions. The action is report-only by default but can be configured to fail if protection does not behave as expected.

What’s Changed

v1.0.1 — positioning + hardening

A correctness and honesty pass before broad promotion. No behavior change to the check itself.

  • Reframed positioning. This Action runs a reproducible Cerberus allow/block self-check against two bundled, maintained fixtures (a benign workflow that should run, a Lethal-Trifecta workflow that should be blocked). It does not analyze the caller’s own application workflows — the README and summary now say so plainly.
  • Pinned core-version default to 3.1.0 (was latest) for reproducible runs.
  • Default runtime is now Node 24, and the bundled official actions were bumped to current majors (checkout@v5, setup-node@v5, upload-artifact@v7) — clears the Node 20 deprecation warnings. Self-test now runs warning-free.
  • README now links the public Odingard/cerberus-core instead of the private core repo.
  • Refreshed Marketplace banner + listing screenshots to match the new wording.

Usage is unchanged:

- uses: Odingard/cerberus-action@v1