duck-refactor
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes source code files to track references, creating a standard surface for indirect prompt injection.\n- Ingestion points: Codebase files are read during the reference tracing stage (SKILL.md, Step 2).\n- Boundary markers: Mandatory execution approval (Step 5) and bounded phase limits (Step 3) are strictly enforced.\n- Capability inventory: The skill is capable of file modification (Step 6) and test execution (Step 7), both gated by approval.\n- Sanitization: The skill focuses on structural integrity but does not explicitly filter content for embedded instructions.
Audit Metadata