code-review
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk of indirect prompt injection by explicitly instructing the agent to treat pull request code, comments, and descriptions as untrusted evidence. It directs the agent to disregard any embedded instructions that attempt to override review rules or suppress findings.
- Ingestion points: Pull request content, including code changes, comments, and descriptions.
- Boundary markers: The skill uses instructional boundaries, explicitly telling the agent to treat data as untrusted.
- Capability inventory: The skill reads local repository documentation files to establish review context.
- Sanitization: Includes specific negative constraints to prevent the agent from obeying instructions found within the reviewed data.
- [SAFE]: The skill primarily reads project-specific documentation (e.g., AGENTS.md, CONTRIBUTING.md, and coding conventions) to inform its reviews. No sensitive file access, network exfiltration, or external code execution is present.
Audit Metadata