evaluate-code-review
Pass
Audited by Gen Agent Trust Hub on May 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill's primary logic focuses on reconstructing the ground truth of code changes and verifying feedback against the codebase using a multi-step verification lens. It explicitly forbids performative agreement, which reduces the risk of social engineering attacks.
- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external PR comments and markdown documents. This is a known risk for the intended use case. The skill mitigates this by using an independent subagent and structured extraction points. Evidence: Ingestion points include gh pr view --comments and local audit files; Capability inventory includes gh api and git operations.
Audit Metadata