adapt
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts on untrusted external data (user requests and repository code), which serves as an entry point for indirect prompt injection attacks.\n- Ingestion points: Processes user-provided requests, code history, and codebase contents (SKILL.md, Step 1).\n- Boundary markers: Instructions do not specify the use of delimiters or 'ignore' instructions when ingesting this data.\n- Capability inventory: The skill allows the agent to execute arbitrary work, hand off tasks to other models, and run verification proofs (SKILL.md, Steps 3 and 4).\n- Sanitization: There are no explicit steps for sanitizing or escaping the content from requests or code before it is used to formulate the 'contract' or 'route'.
Audit Metadata