The public repository is JoshMcMillen/second-brain-models.
- GitHub Actions default permissions are read-only; workflows cannot approve pull requests.
- Only GitHub-owned Actions are allowed, and workflow references are pinned to full commit hashes.
- Secret scanning, push protection, dependency vulnerability alerts, Dependabot security updates, and private vulnerability reporting are enabled.
model-evaluationandmodel-publishrequire review byJoshMcMillenfor manual evaluation reruns and publication.model-evaluation-autohas no reviewer wait and accepts deployments only from protected branches. It is used solely by the read-only post-merge evaluator and has no secrets or publication authority.CODEOWNERSassigns the repository to@JoshMcMillen.
The intended main protection requires a pull request, one CODEOWNER approval, resolved conversations, linear history, and the untrusted-change-checks status. Force pushes and branch deletion are disabled. Repository administrators may bypass the pull-request requirement so the solo owner is not permanently deadlocked by the one-review rule; bypasses remain exceptional owner actions.
Machine-readable environment and branch settings are retained under .github/settings/ for audit and reapplication.
- No production catalog signing key or GitHub signing secret exists. The signing fixture is test-only.
- No R2 credentials exist because R2 is not enabled for the Cloudflare account.
- Publishing and revocation remain disabled until their artifact-first R2 receipts are implemented.
- The protected
evaluate.ymlworkflow automatically qualifies one exact candidate after its manifest merges tomain, without R2 or secrets. Manual reruns remain reviewer-gated. Both paths retain the deterministic result and security receipts but cannot commit, publish, or promote anything.