refactor-claude-md
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches prompt engineering guidelines and model-specific documentation from official domains (
platform.claude.comandcode.claude.com). These downloads are used to calibrate the audit process and provide the agent with current best practices. - [COMMAND_EXECUTION]: The skill utilizes the
claudeCLI to probe system prompts for redundancy. It employs safety-oriented flags such as--safe-mode,--no-session-persistence, and--max-turns 1to ensure the probing process is isolated and does not affect the active session or execute external plugins. - [INDIRECT_PROMPT_INJECTION]: The skill processes content from local
CLAUDE.mdfiles which may contain untrusted data. To mitigate potential risks, the skill is designed to audit these instructions rather than blindly follow them, and it incorporates mandatory user review steps (AskUserQuestion) before any modifications (deletions, rewrites, or demotions) are committed to the file system.
Audit Metadata