implementation-review
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository data (code diffs, README.md, AGENTS.md, .github/copilot-instructions.md) to perform its review. This content is interpolated into reviewer prompts, creating a surface where malicious instructions embedded in a repository could influence the sub-agents.
- Ingestion points: In
SKILL.md, the workflow requires gathering the target repo's diff and reading governing documents likeAGENTS.mdand.github/copilot-instructions.mdto define the review criteria. - Capability inventory: The 'Blast-radius check' in
references/reviewer-prompt.mdincludes a step to 'ran a script or test that exercises the real code', which could lead to the execution of untrusted local scripts found in the workspace. - Boundary markers: The templates in
references/reviewer-prompt.mdandreferences/plan-review-reviewer-prompt.mddo not include explicit delimiters or instructions to the LLM to ignore potentially adversarial content or instructions within the ingested implementation data. - Sanitization: The skill lacks mechanisms to sanitize or validate the content of the repository files before they are processed by the reviewer personas or used to construct prompts.
Audit Metadata