Eval Agent MD is a specialized quality assurance tool designed to ensure your AI agent's rules and guidelines are consistently followed. By analyzing CLAUDE.md or other agent definition files, the skill automatically generates realistic test scenarios and executes them to provide a quantitative compliance score. Beyond reporting, it features an intelligent mutation loop that iteratively suggests and tests improved wording for rules that fail behavioral checks, helping you bridge the gap between intent and execution.
主要功能
01Holistic integration testing for cross-rule interactions
0216 GitHub stars
03LLM-as-judge scoring for objective behavioral evaluation
04Automated mutation loop to iteratively refine failing rules
05Automated scenario generation from rule definitions
06Detailed compliance reports with per-rule pass/fail evidence
使用场景
01Verifying that project-specific instructions in CLAUDE.md are being obeyed
02Stress-testing new agent definitions for role-boundary compliance
03Improving the reliability of developer guidelines through automated refinement