hanselman-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code content and pull request metadata, which could be crafted to contain instructions that hijack the reviewer agents' behavior.
- Ingestion points: The workflow in
SKILL.md(Step 1) utilizesgit diff,gh pr view, andgh pr diffto fetch external content for review. - Boundary markers: The prompt template provided in
SKILL.mddoes not employ secure boundary markers (such as unique delimiters or escape sequences) to isolate the[FULL CODE OR DIFF HERE]content from the instructions. - Capability inventory: The skill uses the
tasktool to spawn sub-agents (rubber-ducktype) and theread_agenttool to retrieve results, along with standard developer tools likegitandghCLI. - Sanitization: The instructions do not specify any sanitization, filtering, or validation of the code diffs or PR metadata before they are interpolated into the review prompts.
Audit Metadata