init-or-refactor-agents-md
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The workflow incorporates a human-in-the-loop requirement, mandating that the agent present conflicts and the proposed new instruction file for confirmation before writing to disk or deleting existing files.
- [SAFE]: The 'Target policies' section explicitly instructs the agent to preserve security-critical rules, invariants, and trust-boundary validations during the refactoring process.
- [SAFE]: The skill mandates a security review of all new dependencies, including fit, license, and vulnerability assessments, prior to seeking user approval for their addition.
- [INDIRECT_PROMPT_INJECTION]: The skill reads instructions from repo-local files and uses them to generate new instructions and determine file deletions, creating a surface where malicious content in those files could influence the agent's output.
- Ingestion points: AGENTS.md, CLAUDE.md, CODEX.md, and .cursorrules (Workflow Step 1).
- Boundary markers: Human confirmation step (Workflow Step 6).
- Capability inventory: File reading, file writing, and file deletion.
- Sanitization: No content-based sanitization is specified for the instructions being refactored.
Audit Metadata