receiving-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to process data from 'External Reviewers' and GitHub comments.
- Ingestion points: Feedback processing logic for human partners and external reviewers defined in
SKILL.md. - Boundary markers: Present. The skill establishes strong boundaries by requiring the agent to 'VERIFY: Check against codebase reality' and 'EVALUATE' technical soundness before taking action.
- Capability inventory: The skill assumes the agent has access to the local codebase (via
grep) and external repositories (viagh api). - Sanitization: The skill implements behavioral sanitization by requiring the agent to push back against suggestions that are technically incorrect, break functionality, or violate architectural decisions.
- [SAFE]: The core logic of the skill is defensive. It explicitly forbids 'blind implementation' and 'performative agreement', which are common failure modes that lead to successful prompt injection. By requiring technical verification for every suggestion, the skill reduces the risk of malicious instructions being executed.
Audit Metadata