Security

The canonical policy is SECURITY.md.

Reporting

Use GitHub's private vulnerability reporting. Please do not open a public issue for a suspected vulnerability.

What counts as a security issue here

This project ships instructions that agents act on, which widens the definition beyond ordinary code:

• Guidance in a SKILL.md that would lead an agent to write insecure code, weaken a control, or exfiltrate data.
• A command, snippet, or configuration that is destructive or unsafe as written.
• Anything that looks like a real credential, private hostname, or internal identifier.
• Prompt-injection content aimed at a model reading these files.

A wrong technical claim is a normal bug — open an issue. A claim that would cause harm if followed is a security report.