skill-security-auditor

Scanning for prompt injection risks involves analyzing the SKILL.md file and code patterns to detect unsafe instructions and risky imports. The auditor evaluates local directories or remote repositories to surface potential security risks and remediation guidance.

A security audit for AI agent skills is an analysis of code risks, prompt injection patterns, dependency supply chains, and filesystem access. It detects dangerous code patterns like eval or subprocess with shell to prevent sensitive data exposure or system compromise.

Checking external skill dependencies for typosquatting and unpinned versions requires inspecting dependency files within the skill directory. The audit process flags unpinned versions, runtime installs, and supply chain risks to generate a PASS, WARN, or FAIL verdict.

You can audit a remote repository URL for unsafe filesystem access. The auditor inspects the target to flag boundary violations and filesystem access outside the skill directory, outputting a JSON report of findings and recommended mitigations.

Detecting dangerous code patterns in external skills requires scanning for risky imports and functions like eval, os.system, and subprocess with shell. This process identifies potential security risks before integration into AI agents.