claude-md-audit
Pass
Audited by Gen Agent Trust Hub on Mar 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security risks were identified. The skill's behavior is restricted to reading local configuration files and proposing optimizations to the user.
- [DATA_EXFILTRATION]: The skill reads the global configuration file at ~/.claude/CLAUDE.md and local project files. This access is transparently disclosed and necessary for the primary function of auditing configurations. No data is transmitted externally.
- [PROMPT_INJECTION]: The skill ingests content from external files like package.json and CLAUDE.md, creating a surface for indirect prompt injection. Ingestion points: File reading instructions 1 and 2. Boundary markers: Not explicitly defined. Capability inventory: Local file read and conditional write. Sanitization: Relies on LLM reasoning.
- [COMMAND_EXECUTION]: The skill includes a 'confirm before write' safety mechanism, ensuring that no file system changes occur without explicit user approval.
Audit Metadata