evaluation-design
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: The skill contains only documentation and guidelines for evaluation design. No executable code, scripts, or automated tool invocations are present.
- [INDIRECT_PROMPT_INJECTION]: Analysis of the evaluation design surface: 1. Ingestion points: production traces and pilot trajectories (SKILL.md); 2. Boundary markers: explicit instructions to treat inputs as untrusted data rather than instructions (references/red-team-design.md, references/judge-validation.md); 3. Capability inventory: no subprocess calls, file writes, or network operations in skill files; 4. Sanitization: instructions to use explicit pass/fail definitions and outcome classes to prevent interpretation bias.
- [EXTERNAL_DOWNLOADS]: Referenced external materials from trusted organizations and well-known services (e.g., Anthropic, Microsoft, NVIDIA) are used for methodology only and are not executed or downloaded by the skill.
- [CREDENTIALS_SAFE]: No sensitive information, hardcoded credentials, or unauthorized file paths are exposed.
Audit Metadata