| name | judge-rubric |
| description | Generate a formalized rubric for scoring, grading, or evaluation in the current domain. Use when a judging task needs locked dimensions, pass-partial-fail boundaries, evidence requirements, tie-breakers, or confidence guidance before candidate comparison. |
| argument-hint | Domain, task contract, artifacts, scoring mode, and any existing evaluation criteria. |
Rubric Authoring
Use this skill to build a formal judging rubric before scoring responses, candidates, trajectories, benchmark outputs, or other evaluation artifacts.
Read references/rubric-techniques.md when you need current judging guidance on locked rubrics, task-adaptive rubric design, verifier-backed evidence, robustness checks, calibration, and chain-of-thought skepticism.
When the judging inputs are already available as a structured contract, use scripts/render_rubric.py to render the rubric package deterministically instead of drafting it manually.
When to use this skill
- The judging task is missing a stable rubric and needs one before scoring starts.
- The user asks for a grading rubric, scoring rubric, evaluation rubric, or formal scorecard.
- The domain has specific failure modes, policy constraints, or artifact requirements that must become explicit judging dimensions.
- The evaluation needs pass, partial, and fail boundaries instead of holistic free-form grading.
- The comparison is high-risk enough that tie-break rules, robustness checks, and confidence guidance must be defined in advance.
Required inputs
- The task contract or judging objective.
- The domain context, expected outputs, and any non-negotiable constraints.
- The artifacts that will later be judged, such as final outputs, process traces, validation logs, or benchmark summaries.
