safe-refactor-code
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and updates local repository files such as AGENTS.md and files within the .safe-code/ directory to maintain agent continuity and architectural context. These files are treated as high-trust sources for the agent's operational logic during refactoring.
- Ingestion points: Reads project-level files including AGENTS.md, .safe-code/MEMORY.md, and .safe-code/CHANGELOG.md.
- Boundary markers: Utilizes HTML comment markers (e.g., ) to differentiate between tool-generated artifacts and human/agent-authored state.
- Capability inventory: Has the ability to perform repository-wide refactoring via tools (refactor_tool, apply_refactor_tool) and write documentation updates to the file system.
- Sanitization: The skill does not explicitly detail sanitization or filtering for natural language instructions found within the ingested continuity files, relying instead on canonical authoring rules to govern content creation.
Audit Metadata