beepometer/gutcheck
Per-diff verification-coverage report for AI-generated changes: gutcheck breaks each changed function and reruns only its own tests, classifying every changed function as proven, hollow, one-sided, unverifiable, or untested — with a replayable receipt per verdict. Fail-closed: what it cannot verify is reported, never guessed.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| path | Project directory to probe. | — | . |
| since | Git ref that scopes the probe to the diff. Default: the PR base sha (falls back to HEAD~1). Use fetch-depth: 0 on checkout, or any ref reachable in the clone. | — | "" |
| max-probes | Cap on functions probed (bounds CI latency on a big diff). | — | 40 |
| comment | Post/update a sticky PR comment with the verification table (needs permissions: pull-requests: write). | — | true |
| sarif-file | When set, write SARIF 2.1.0 here (upload with github/codeql-action/upload-sarif). | — | "" |
| fail-on-hollow | Fail the job when a hollow test is proven (a test that passes with its function gutted). | — | true |
| node-version | Node version for the probe (needs >= 20; 22 recommended). | — | 22 |
Outputs
no outputs