agent-md-refactor
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is designed for documentation management and does not exhibit any malicious patterns. It lacks network access, privilege escalation, or persistence mechanisms.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided agent instruction files (e.g., CLAUDE.md). While these files are untrusted data sources, the skill only performs text refactoring and file-writing operations within the project directory. It does not execute the content of the files or use them in a way that could lead to command execution or data exfiltration. The risk is negligible as it is an organizational tool.
- Ingestion points: Reads existing agent instruction files (CLAUDE.md, AGENTS.md, COPILOT.md) from the project root.
- Boundary markers: None explicitly defined in the refactoring templates, though Phase 1 includes a contradiction analysis step which acts as a manual review phase.
- Capability inventory: File-write (creates a
.claude/directory and multiple.mdfiles containing the refactored instructions). - Sanitization: Content is refactored and reformatted according to provided templates, but no explicit escaping of malicious prompt content is performed.
Audit Metadata