auditing-skills

You can scan AI agent skills for prompt injection and dangerous bash patterns by running an automated security audit on the skill directory to identify and report risks in SKILL.md files and associated scripts.

Validating SKILL.md frontmatter structure involves checking for misconfigurations and quality standard violations, while the security audit identifies supply chain risks and prompt injection vectors to ensure safe deployment.

You can audit both single-skill directories and entire skill-collection directories, allowing you to validate multiple AI skills at once and generate actionable findings for remediation across the collection.

The audit supports integration with available external security tools like shellcheck, gitleaks, and semgrep to supplement built-in scanners for bash patterns, frontmatter validation, and package installation checks.

Scoring and behavior are driven by a local configuration file that manages suppression rules and coordinates built-in scanners alongside external tools to produce actionable security findings.

AI agent skills are flagged for supply chain risks when the audit detects dangerous package installation patterns or misconfigurations in the skill directory that could lead to insecure dependencies.