skill-security-auditor

You can audit AI agent skills by scanning local directories or Git repositories to evaluate SKILL.md content, scripts, and dependencies for prompt injection risks. This generates a structured JSON report with a PASS/WARN/FAIL verdict and remediation guidance.

Checking for prompt injection vulnerabilities involves static and dynamic scanning of SKILL.md files and analyzing scripts/ with references/ for risky patterns. The auditor identifies safety issues and returns actionable findings to help you make trust decisions before installation.

Scanning local skill directories detects malicious code and data leakage exposure by analyzing code patterns and filesystem interactions. The auditor evaluates dependencies and scripts to proactively identify security risks, returning a structured report with a clear PASS, WARN, or FAIL verdict.

The best way to evaluate supply-chain risks is running a dedicated security audit that analyzes dependencies, scripts, and SKILL.md content. This detects risky patterns and filesystem interactions, providing a structured JSON report with remediation guidance for secure trust decisions.

The security audit operates with no external dependencies, directly analyzing local skill directories and Git repositories. It evaluates SKILL.md content, scripts, and filesystem interactions natively to produce a structured JSON report with security findings.

Limitations involve relying on static and dynamic scanning of scripts/ and references/ to detect risky patterns and prompt injection risks. The auditor provides a PASS/WARN/FAIL verdict based on filesystem interactions and code analysis, but cannot execute the skill to observe runtime behavior.