Thanks for helping keep this suite safe.

Please report security issues privately — do not open a public issue.

• Preferred: Create a security advisory via the repository's Security → Report a vulnerability tab (GitHub Security Advisories). This is private by default and lets us coordinate a fix before disclosure.
• Fallback: Email the maintainer with details.

Please include:

• The affected skill, script, or file and its version.
• A description of the issue and why it is a security problem.
• Steps to reproduce, or a minimal proof of concept.

In scope:

• Security issues in the deterministic tooling (scripts/**, python-*/scripts/**): arbitrary code execution, unsafe path or AST handling, secret leakage, or tooling that can be tricked into misreporting a repository.
• Skill instructions that could cause an agent to leak secrets, approve destructive commands, or skip identity/authorization checks.
• The LLM-as-judge harness (scripts/judge_eval.py, scripts/run_evals.py): prompt-injection from skill outputs; API-key handling.

Not in scope (they are the subject of the suite, not its code):

• Vulnerabilities in a user's own application; the python-security skill exists to review those. Report them through the affected project's own process.

We aim to acknowledge reports within 48 hours and to triage vulnerabilities with a severity assessment. Fixes land as a release and the advisory is published after a fix is available.

• Install skills only from sources you trust and review what you copy. This suite ships a python3 scripts/validate_suite.py consistency check — run it after copying the suite into another repository.
• The eval harness reads skills' outputs and sends them to the configured model endpoint; export LLM_API_KEY only in environments that should hold it.
• Never commit secrets. Treat any fixture or example that contains a credential as an error and report it under this policy.