product-quality-perspective
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFENO_CODEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data (project materials) as part of its core functionality, which constitutes an indirect prompt injection surface.
- Ingestion points: Untrusted project materials such as requirements, PR diffs, and test reports are processed through
SKILL.mdand the prompts defined in theprompts/directory. - Boundary markers (present): The skill incorporates several mitigation strategies, including a strict report contract (Summary, Facts, Evidence, etc.) and a pre-delivery checklist that requires the explicit separation of evidence from inferences.
- Capability inventory: The skill is passive and lacks access to dangerous capabilities such as shell execution, network request utilities, or persistent storage modification tools.
- Sanitization (absent): No formal sanitization or encoding of input data is performed; the skill relies on instructional boundaries for content isolation.
Audit Metadata