If you discover a security vulnerability in this skill — particularly in the Guideline #12 hard rule + bypass-protocol logic, the pre- push gate runbook, the /security-review-scoped wrapper, or any other security-critical surface — please do not open a public issue.

GitHub Security Advisories: Open a private advisory

This routes directly to the maintainers + keeps the disclosure private until coordinated.

• A description of the issue
• Steps to reproduce (skill invocation that triggers the issue + the expected vs actual behavior)
• The skill version (visible in the SKILL.md frontmatter or version: field — currently 2026.05.24-v5.1)
• Your assessment of impact (confidentiality / integrity / availability / mechanism-bypass severity)
• Any proposed fix or mitigation

• Acknowledgment within 7 days
• Initial triage + impact assessment within 14 days
• Coordinated fix + advisory publication when ready
• Credit in the advisory (unless you prefer to remain anonymous)

• The skill's mechanical hard rules (Guideline #12 freshness gate; bypass-phrase enforcement; pre-push gate row failures)
• The references/ content as it drives operator behavior
• The scripts/control_chart.py helper
• The bundled companion-skills/security-review-scoped/ wrapper
• The references/templates/ files (templates emit into operator repos; bugs here propagate)

• Operator misuse (e.g., typing the wrong bypass phrase intentionally)
• Issues in Claude Code itself, in the Anthropic API, or in any third-party tool the skill orchestrates (gh, pytest, osv-scanner, cosign, grype, etc.) — report those upstream
• Operator-side configuration drift (e.g., enforce_admins: false on a project that needs true) — this is by-design operator choice

The skill is designed to be self-improving: running it against a real project surfaces issues with the skill itself, captured as SF- findings* (skill iteration findings) in each ship's docs/security-review-vX.Y.Z.md. These are tracked in this repo's issues / discussions, not as security advisories, unless they have a security implication.

Example: examples/v0.10.4-walkthrough.md documents 9 skill iteration findings discovered during the v5.1 prototype run; these flow into v5.1.x candidate work.

This skill ships as Markdown + Python helpers + YAML templates. No binary distributions; no compiled artifacts. Tampering at install time is mitigated by:

• Cloning from https://github.com/Polycentric-Labs/pre-release-review over TLS
• Reviewing the diff before pulling updates (the skill is designed to be auditable; the maintenance rubric in references/maintenance.md enforces audit-grade discipline)
• The scripts/control_chart.py helper is stdlib-only — no transitive dependencies

For deeper supply-chain assurance (signed releases, SBOM, reproducibility), see the skill's own references/step-7-post-tag.md which captures the SLSA L3 + NIST SSDF PS.3.1 verification surface the skill encourages operators to ship.