repo-protections

name	repo-protections
description	Configures GitHub repository protections - rulesets (require PR, required checks, no force pushes), secret scanning and push protection, Dependabot, private vulnerability reporting, and Actions hardening. Use when the user asks to protect a branch, set up branch protection or rulesets, enable security features, or harden a repo. Not for writing SECURITY.md or fixing code vulnerabilities.
license	MIT
argument-hint	["ruleset|scanning|dependabot|pvr|all"]

repo-protections

Applies the server-side gates that client-side hooks and good intentions can't replace. The failures this skill fixes: models suggest the legacy branch-protection UI instead of rulesets, don't know the gh api payloads, forget that CODEOWNERS does nothing until a ruleset requires code-owner review, and let bot PRs auto-merge.

When NOT to use

• SECURITY.md and other prose files → community-health.
• Fixing a vulnerability in code, rotating leaked credentials → incident work, not settings.
• Release pipelines / OIDC trusted publishing → out of scope here.
• CI workflow authoring → out of scope (naming required checks is in scope).

Workflow

1. Audit first (read-only; degrades gracefully on 403/404):

   python3 "${CLAUDE_SKILL_DIR}/scripts/check_protections.py" [owner/repo]
   

   Reports: ruleset/protection on the default branch, secret scanning, push protection, Dependabot config, PVR — with pass/warn/fail per control.
2. Read references/protections-playbook.md, then apply what's missing, in this order (each step has the exact command in the playbook):
   • Ruleset on the default branch (prefer rulesets over classic protection: auditable bypasses, apply to admins, org-inheritable): require PR; required status checks named exactly as they report (matrix CI → one aggregator job as the only required check; merge queues need the merge_group trigger); block force pushes and deletions; require code-owner review only if CODEOWNERS exists — and if it doesn't, say that CODEOWNERS + this toggle is the pair that makes path ownership real.
   • Secret scanning + push protection — default on public; private repos need GHAS/Secret Protection. Enable via the security_and_analysis PATCH.
   • Dependabot — .github/dependabot.yml for the ecosystems actually in the repo: weekly, grouped, 7-day cooldown. State the two limits: security updates bypass cooldown by design; cooldown does not cover transitive npm deps. Never recommend blanket auto-merge of bot PRs — bots carry unearned trust; they get the same CI + review gate as humans.
   • Private Vulnerability Reporting — public repos only, off by default.
   • Actions hardening (baseline only): pin third-party actions to commit SHAs with a version comment, top-level permissions: contents: read, per-job elevation, caution with pull_request_target.
3. Verify: re-run the check script; every applied control must show as active. For rulesets also confirm with gh api repos/{owner}/{repo}/rulesets.
4. Report what was enabled, what was skipped and why (plan limits, missing CODEOWNERS, private-repo constraints).

Output spec

Requested protections active and verified by re-audit; anything unapplicable (private repo on free plan, missing GHAS) reported explicitly with the unlock condition — never silently skipped.

Gotchas

• Rulesets and branch protection on private repos require a paid plan; the API returns 403/upgrade errors — surface that, don't retry blindly.
• A required status check that has never run on the repo can't be selected/matched; run the workflow once first.
• Required checks match on the reported check name, not the workflow filename.
• Secret-scanning push protection blocks only known patterns — present it as one layer, not a guarantee.
• Enabling everything on a fork or throwaway is noise — confirm the repo is the long-lived one before applying org-grade gates.

Files

• references/protections-playbook.md — ruleset recipe with gh api JSON payload, per-control commands + verification, minimum-viable posture table, Scorecard note.
• scripts/check_protections.py — read-only audit; non-zero exit when the default branch has no ruleset/protection.