zj-implement
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided specifications and tickets, which serves as an ingestion point for potentially untrusted data. \n
- Ingestion points: The skill reads 'work described by the user in the spec or tickets' as seen in SKILL.md. \n
- Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potential instructions embedded within the specifications or tickets. \n
- Capability inventory: The skill instructions involve running typechecking, executing test suites, and committing code to a branch, which implies the agent has access to the shell and file system (SKILL.md). \n
- Sanitization: No sanitization, validation, or escaping of the input data is mentioned or implemented in the instructions.
Audit Metadata