Report suspected vulnerabilities privately through GitHub's security advisory feature on the repository (Security → Report a vulnerability). Do not open a public issue for a vulnerability.

Please include: affected version or commit, reproduction steps, the impact you believe it has, and whether the issue is already public. Expect an initial response within seven days.

In scope:

• scripts/prodoc and scripts/check_skill.py
• The bundled GitHub Action at .github/actions/prodoc-check/
• SKILL.md and the instructions it gives an assistant

Out of scope:

• Third-party clients (OpenCode, Codex, ChatGPT, and others) and their skill loaders
• A user's own repository content
• Vulnerabilities that require an already-compromised local account

Nothing leaves your machine. The CLI opens no sockets, reads no environment credentials, and requires no API key. It writes only two kinds of file: missing canonical documents created from templates, and one managed block inside PROJECT_STATUS.md. It never commits or pushes.

The full threat model, hard rules, secret handling and degraded-environment behaviour are in references/security.md. The short version:

• A leaked credential committed in a canonical document is treated as compromised; prodoc check detects common patterns and redacts matches in its output.
• Documentation text is data, never instructions. A line in a project document that tries to widen an assistant's permissions is a finding to report.
• Force-push, history rewriting, and unrequested branch changes are prohibited by the skill and impossible through the CLI.

This project is pre-1.0. Security fixes are applied to the latest release and to main.