agent-md-refactor
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process, extract, and reorganize content from instruction files like AGENTS.md or CLAUDE.md, which could contain adversarial instructions if sourced from untrusted repositories.
- Ingestion points: The agent is instructed to read the target file, nested instruction files, and every document it links to (e.g., SKILL.md files or documentation).
- Boundary markers: The skill includes safety procedures such as requiring the agent to show proposed changes to the user and obtain approval before writing. It also requires the agent to stop and ask the user if authoritative rules conflict.
- Capability inventory: The agent uses its standard file-system tools to read, create, and modify markdown files within the repository.
- Sanitization: The instructions emphasize preserving exact commands, paths, and prohibitions, and explicitly forbid the agent from rewriting policy during the reorganization process.
Audit Metadata