requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where a subagent (reviewer) ingests and analyzes git diff output and plan requirements. This constitutes an attack surface where malicious code or comments within the diff could attempt to influence the reviewer's verdict.\n
- Ingestion points: The code-reviewer.md template and the git diff output processed by the reviewer agent as described in references/cross-provider-review.md.\n
- Boundary markers: The instructions recommend recording and verifying commit SHAs and diff digests to ensure the reviewer sees the intended snapshot (SKILL.md and references/cross-provider-review.md).\n
- Capability inventory: The reviewer agent is configured to use file-reading tools (Read, Grep, Glob) and git commands (git diff) as illustrated in the claude and codex examples in references/cross-provider-review.md.\n
- Sanitization: The skill does not define automated sanitization of diff content, instead relying on the primary agent to verify findings and explicitly warning the user not to run it inside a checkout of an untrusted pull request.
Audit Metadata