code-review-loop
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted external data sources, creating an attack surface where instructions embedded in pull request content could influence agent behavior.
- Ingestion points: The skill ingests external data from pull request diffs in Step 2 and human reviewer comments in Step 3.
- Boundary markers: The instructions do not define boundary markers or explicit instructions for the agent to ignore potentially malicious commands embedded within the diffs or comments.
- Capability inventory: The skill involves significant capabilities including committing code changes (Step 2, Step 4) and merging pull requests (Step 6), which could be exploited if an injection is successful.
- Sanitization: The process does not include a verification or sanitization step to ensure that external comments or code changes do not contain adversarial instructions before the agent acts on them.
Audit Metadata