Evaluate Skill

Orchestrate a cross-tier evaluation of an AI skill to determine its clarity and robustness.

Procedure

1. Load inputs

   • Read the skill file at {{ skill-path }}
   • Read the test cases file at {{ test-cases-path }}
   • Validate that test cases is a JSON array of objects with input and expectedOutcome fields

2. Set up evaluation matrix

   • Model tiers to test: opus, sonnet, haiku
   • For each tier, for each test case: plan one blind test run