improve-agents-md
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill provides guidelines and modes for refactoring local instruction files and does not use any dangerous commands, external network connections, or sensitive data access.
- [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it is designed to ingest and process content from user-provided files like CLAUDE.md and AGENTS.md. Instructions within those files could theoretically attempt to influence the agent's behavior during the analysis or rewrite process. This risk is inherent to the skill's primary purpose of document processing.
- Ingestion points: Reads existing root instruction files and repository layout (reference/analyze-mode.md).
- Boundary markers: The methodology encourages the use of conditional blocks (e.g., <important if="...">), which provide some structure, but no global boundary markers are enforced for all ingested data.
- Capability inventory: The skill is restricted to reading files and writing Markdown files within the user's workspace (SKILL.md, reference/rewrite-mode.md).
- Sanitization: No explicit sanitization of processed instructions is described in the provided files.
Audit Metadata