Skip to content

AI-driven qualitative assessment for skills.

skill judge will send a skill to an LLM for qualitative review — assessing clarity, completeness, and effectiveness beyond what deterministic checks can catch.

Terminal window
doraval skill judge <path> [options]

validate and drift are deterministic and run locally with zero network calls. judge is different:

  • Requires an API key or local model
  • Non-deterministic — same input may produce different output
  • Slower — network round-trip vs. instant local checks
  • Optional — fast CI uses validate + drift; deep feedback uses judge
FlagShortDescription
--format <type>-ftable (default) or json
--agent <name>-aForce a specific agent adapter
--verbose-vShow detailed diagnostics