This repo ships agent instructions and dependency-free shell/python tooling — no runtime services, no third-party packages. The surfaces that matter:
- Scripts run on contributor machines. Report anything that could make
sync-skills,committer, or the validators destructive or injectable (e.g. crafted file names, descriptions, or front matter reaching a shell). - Skills are followed by AI agents. Report skill content that could steer an agent into unsafe behavior: secret exfiltration, destructive git operations, or prompt-injection vectors.
- CI workflow changes. The workflow runs with read-only permissions; report anything that would escalate them.
Report privately via GitHub Security Advisories ("Report a vulnerability" on this repository) and include a reproduction. Please do not open public issues for suspected vulnerabilities.