refine-file
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from existing instruction files, agent prompts, and path rules for auditing purposes.
- Ingestion points: The skill reads the named file in Step 0.
- Capability inventory: The skill has the capability to delete (CUT), move (MOVE), and modify (SHAPE) file content across the repository.
- Boundary markers: The instructions lack explicit delimiters or instructions to the agent to disregard embedded directions within the audited source text.
- Sanitization: The skill employs a confidence-based gating system (c >= 0.75) and a verification step in Step 4, though these depend on LLM reasoning rather than fixed security boundaries.
Audit Metadata