skill-security-auditor

To scan AI agent skills for security risks, audit local skill directories or git repositories to detect command execution, prompt injection, and credential access. This generates a PASS, WARN, or FAIL verdict with remediation guidance to gate third-party plugins safely.

Static analysis for prompt injection is a scanning mechanism that reviews SKILL.md and reference markdown files for hidden instructions and safety bypasses. It identifies suspicious directives without executing code, preventing malicious boundary violations during skill installation.

Yes, you can audit a git repository for unsafe dependencies. The scanner evaluates skill packages for supply-chain issues, unsafe deserialization, and exfiltration risks, returning a clear PASS, WARN, or FAIL decision based on the repository's boundary-violating files.

The best way to detect exfiltration and privilege escalation in markdown instructions is to perform static document scanning on SKILL.md files. This process catches hidden directives and safety bypasses, producing actionable findings before the skill package is installed.

Yes, the security audit checks for supply-chain vulnerabilities and code execution. It statically scans skill directories and repositories to identify risky dependencies and command execution paths, yielding a FAIL or WARN status if boundary violations are found.