{
  "skill_name": "sync-skill-taxonomy",
  "evals": [
    {
      "id": 1,
      "name": "basic-invocation",
      "prompt": "Apply the sync-skill-taxonomy skill to skills/: classify every skill as user-invoked or model-invoked and fix the frontmatter flags.",
      "expected_output": "Every skill classified, with disable-model-invocation set on user-invoked skills and metadata.category injected on all.",
      "assertions": [
        "The output classifies each skill explicitly",
        "User-invoked skills get disable-model-invocation: true",
        "All skills receive metadata.category"
      ]
    },
    {
      "id": 2,
      "name": "ambiguous-input",
      "prompt": "Sync the taxonomy, but I'm not sure which skills should be auto-loaded.",
      "expected_output": "The skill applies its classification criteria (routing context vs explicit invocation) and reports borderline cases for decision instead of guessing silently.",
      "assertions": [
        "The output states the classification criteria it used",
        "Borderline cases are surfaced explicitly",
        "The output does not silently reclassify model-invoked skills"
      ]
    },
    {
      "id": 3,
      "name": "complex-scenario",
      "prompt": "Run sync-skill-taxonomy after a batch import added 30 new skills with no frontmatter conventions, some duplicates of existing skills.",
      "expected_output": "Classification of all new skills plus duplicate detection, with a report of changes made.",
      "assertions": [
        "Duplicates are detected and flagged",
        "New skills receive consistent frontmatter",
        "The output includes an auditable change report"
      ]
    }
  ]
}