What problem does it solve? Installing AI agent skills from untrusted sources can introduce malicious code, prompt injection, credential theft, or data exfiltration into your environment. This Skill statically audits a skill directory or git repository before installation and produces a clear PASS, WARN, or FAIL verdict with remediation guidance. ## Core Features & Use Cases - Code Pattern Scanning: Detects dangerous patterns in Python, Bash, and JavaScript/TypeScript files, including os.system, eval, subprocess with shell=True, base64 obfuscation, network exfiltration, and credential harvesting. - Prompt Injection Detection: Scans SKILL.md and markdown reference files for system prompt overrides, role hijacking, safety bypass directives, hidden zero-width characters, and data extraction instructions. - Supply Chain & Structure Checks: Flags typosquatted dependencies, unpinned versions, binary payloads, symlinks escaping the skill directory, and writes to shell config files. - Use Case: Before installing a community skill from GitHub, run the auditor against the repo URL to receive a severity-grouped report with exit codes suitable for CI/CD gating. ## Quick Start Ask the agent to audit a local skill directory or git repository URL with the security auditor script and report whether it passes, warns, or fails.
