review-change
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as pull request descriptions, code diffs, and repository documentation (e.g., CONTRIBUTING.md, REVIEW.md). This creates a surface for indirect prompt injection where malicious content could attempt to influence the agent's logic.
- Ingestion points: AGENTS.md, repository guidelines, and external tracker data (issues/PRs).
- Boundary markers: The skill explicitly instructs the coordinator to use a "fresh context" and "separate fresh context" for sub-reviewers to isolate the analysis of untrusted data.
- Capability inventory: The skill is primarily analytical; it defines a logic flow for reviews and references external shared methods for investigation but contains no direct command or network execution logic within the analyzed files.
- Sanitization: The skill relies on architectural isolation (fresh contexts) to manage the risks of untrusted content.
Audit Metadata