The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled script 'scripts/audit-skills.mjs' using the local Node.js runtime. This script performs static analysis on other skill files within the user's project. The script uses only built-in Node.js modules (fs, path) and does not require external package installation or network access.\n- [SAFE]: The skill requires read access to the project's skill directories and optional write access for its 'patch mode.' This access is necessary for its stated purpose of auditing and improving skill files and does not involve unauthorized data exposure or exfiltration.\n- [PROMPT_INJECTION]: As an auditing tool, this skill processes content from other SKILL.md files, creating an indirect prompt injection surface. Findings: 1. Ingestion point: reads file content from skills/*/SKILL.md using the fs module; 2. Boundary markers: none explicitly defined in the agent instructions; 3. Capability inventory: local script execution via node, file system read access, and optional write access; 4. Sanitization: the skill relies on deterministic script validation for structure, but manual instructions require the agent to process raw text which could contain malicious instructions designed to influence the audit outcome. This surface is inherent to the skill's purpose and is mitigated by the deterministic validation script.

llm-wiki-skill-doctor