clean-code-review
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting untrusted data from external PR diffs and file contents (SKILL.md, Scope section) and has the capability to write back to those PRs using the
gh apitool. It lacks explicit boundary markers or instructions to sanitize or ignore potential commands embedded within code comments, rendering it vulnerable to attackers who might include malicious instructions in their PR code to manipulate the agent's review output. - [METADATA_POISONING]: The instructions contain a concealment directive that mandates the removal of any AI-related attribution, signatures, or trailers from its output (SKILL.md, Delivery section). This practice is intended to make the agent's output indistinguishable from human-authored comments, facilitating deceptive interactions by hiding the use of AI from other contributors to the repository.
Audit Metadata