correctness-review
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of code diffs generated via
git diff. These diffs are then passed to a secondary reviewer agent or sub-agent for analysis. Adversarial content within the code comments or logic of the analyzed files could potentially influence the reviewer agent's behavior or findings. - Ingestion points: The skill ingests diff data via the
git diff <base>...<head>command output (SKILL.md, Step 1 & 3). - Boundary markers: While a rubric (RUBRIC.md) is provided to scope the review, there are no explicit instructions to the sub-agent to treat comments or strings within the diff as non-instructional data.
- Capability inventory: The skill uses
delegate_taskto spawn a sub-agent and performs read operations on the local file system to judge findings (SKILL.md, Step 3 & 4). - Sanitization: No sanitization or filtering of the diff content is performed before it is passed to the reviewer model.
Audit Metadata