Context
repo-sentinel-lite is currently available only as a pre-push/manual full-repository hook in this repository. Promoting it directly to a required remote check would not create a trustworthy publication boundary yet.
Measured evidence
All output was kept redacted; this issue records counts only.
Decision
Do not add a blocking remote gate and do not regenerate the baseline automatically. A baseline refresh without review could suppress a real secret, while the current full scan would create excessive false-positive noise.
Acceptance criteria
- Consume a reviewed release that contains changed-files scanning, baseline audit, redacted output, and fingerprint deduplication, or explicitly pin an immutable reviewed commit.
- Manually classify ambiguous and unmatched findings without exposing token values in logs or issues.
- Confirm no real credential is present; rotate first if any real secret is found.
- Define the blocking policy: error-level changed-file findings fail; baseline drift remains a separate audit signal.
- Ensure the workflow runs for all relevant file changes with
contents: read only.
- Add a synthetic pass/fail/redaction integration test before making the check required.
- Record rollback as removal of the remote job while retaining the local pre-push hook.
Non-goals
- No bulk baseline refresh for metric reduction.
- No claim that repo-sentinel replaces human authorization, PII review, or the canonical placeholder validator.
Context
repo-sentinel-liteis currently available only as a pre-push/manual full-repository hook in this repository. Promoting it directly to a required remote check would not create a trustworthy publication boundary yet.Measured evidence
All output was kept redacted; this issue records counts only.
repo-sentinel-lite v0.7.1: 146 unsuppressed error-level findings against currentmain.Decision
Do not add a blocking remote gate and do not regenerate the baseline automatically. A baseline refresh without review could suppress a real secret, while the current full scan would create excessive false-positive noise.
Acceptance criteria
contents: readonly.Non-goals