llm-as-judge
Pass
Audited by Gen Agent Trust Hub on Apr 2, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill content was analyzed and no malicious patterns, obfuscation, or unauthorized access attempts were detected. The instructions follow standard patterns for prompt engineering and qualitative assessment.
- [PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface, as the core functionality involves an agent processing untrusted artifacts for evaluation. This is inherent to the 'LLM-as-Judge' pattern and does not represent a malicious implementation in this context.
- Ingestion points: The
artifactparameter in the Phase 3 evaluation template (SKILL.md). - Boundary markers: The use of structured JSON for the review request provides logical separation between the artifact and the evaluation instructions.
- Capability inventory: The skill is instructional and does not utilize tools for command execution, file system access, or network communication.
- Sanitization: The skill does not define specific sanitization routines for the artifacts being judged.
Audit Metadata