skill-security-auditor

To scan AI agent skills for security risks, you can run a security auditor on a local skill directory or git repository URL. It identifies static patterns for command execution, prompt injection, and credential access, producing a structured PASS, WARN, or FAIL report with remediation guidance.

Prompt injection detection in AI skill repositories identifies malicious instructions in SKILL.md and reference docs that attempt to override agent behavior, bypass safety protocols, or exfiltrate files. It flags these dangerous textual patterns before you install the skill.

Yes, you can integrate this security scanner into pre-install CI pipelines to automatically reject skill PRs with critical findings. It supports manual audits and automated CI pre-install gates for Claude Code plugins, OpenClaw skills, and Codex skills.

To detect dependency typosquatting and unpinned packages in AI skills, the scanner performs dependency and filesystem checks. It flags typosquatted package names, unpinned dependencies, hidden files, binaries, and reads to credential locations.

During an AI skill security audit, the static code scanning feature flags dangerous constructs such as eval and exec usage, shell injections, subprocess calls with shell=True, and obfuscated payloads that pose security risks.

Static analysis for AI skill vulnerability scanning detects predefined dangerous patterns and filesystem anomalies but may miss dynamic runtime behaviors or novel zero-day exploits not present in its static pattern definitions.