nv-eval
Pass
Audited by Gen Agent Trust Hub on Apr 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is purely instructional and provides a methodology for agent evaluation without executing any dangerous code or accessing sensitive data.
- [SAFE]: Code examples provided (e.g., Python snippets for linting and testing) are conceptual and intended for user implementation, posing no direct execution risk.
- [SAFE]: No instances of prompt injection, data exfiltration, or obfuscation were found.
- [SAFE]: Indirect Prompt Injection Surface: The skill defines a judge prompt template in Phase 2. 1. Ingestion points: agent_output and acceptance_criteria in Phase 2. 2. Boundary markers: Uses text headers; lacks robust delimiters. 3. Capability inventory: None. 4. Sanitization: Absent.
Audit Metadata