code-review
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The instructions include the directive to "Ground findings in the actual code and stated requirements, not earlier conversation history." This mimics "ignore previous instructions" patterns by commanding the agent to disregard previous context in favor of the provided data.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon code changes, which are untrusted external inputs.
- Ingestion points: The agent processes source code, build scripts, and requirement files (SKILL.md).
- Boundary markers: The instructions lack explicit technical delimiters to separate untrusted code from instructions, although they provide conceptual grounding rules.
- Capability inventory: The skill description implies the ability to execute "project runners," manage "build directories," and control system "processes," "writers," and "handles" (SKILL.md).
- Sanitization: There are no mentioned mechanisms for sanitizing or escaping the content of the files before processing.
Audit Metadata