ai-evals

Pass

Audited by Gen Agent Trust Hub on Oct 17, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is a legitimate methodology and toolkit for AI evaluation. The instructions and reference materials follow industry best practices for data handling and security testing.
  • [COMMAND_EXECUTION]: The scripts scripts/test_analyze_paired_results.py and scripts/test_prompt_eval_runner.py utilize subprocess.run to execute the skill's own internal analysis scripts for verification purposes. These calls use controlled arguments and targets, representing a standard and safe automated testing pattern.
  • [DATA_EXPOSURE]: The skill documentation (references/dataset-construction.md, references/eval-dataset-design.md) explicitly mandates the de-identification of PII and the removal of secrets from production logs before they are incorporated into evaluation datasets, demonstrating proactive security discipline.
  • [INDIRECT_PROMPT_INJECTION]: The skill identifies indirect prompt injection as a major threat to agentic systems and provides specific, actionable red-teaming strategies to detect and mitigate such vulnerabilities in the systems being evaluated. It does not introduce any surface for such attacks itself.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 17, 2026, 08:26 AM