What problem does it solve? Writing a SKILL.md file does not guarantee it actually changes agent behavior. This Skill provides a structured methodology to prove a skill works: run a baseline pass without the skill to document failure modes, then a GREEN pass with the skill loaded, with an independent Haiku judge agent scoring every response against a checklist. ## Core Features & Use Cases - Baseline/GREEN test methodology: Run prompts with and without the skill to isolate its actual effect on agent output. - Scenario file format: Define tests with exact prompts, MUST Contain items drawn from the skill's teachings, and MUST NOT Contain items drawn from observed baseline failures. - Independent judge agent: A Haiku subagent evaluates responses against the checklist and returns PASS, FAIL, or PARTIAL verdicts, removing biased self-judgment. - Use Case: After writing a SKILL.md for a UI framework, create test scenarios, run the baseline to capture the wrong answers unguided agents give, then verify the GREEN run passes all MUST criteria and record results in test-results.md. ## Quick Start Create a test scenario for my skill and run a baseline pass without the SKILL.md, then a GREEN pass with it, using a Haiku judge to verify the results.
