What problem does it solve? Defining quality criteria beyond correctness for LLM evaluation is error-prone: trait types must be matched to the quality being assessed, judge prompts must be explicit, and construction rules (unique names, no dots, required fields) are easy to violate. This Skill guides rubric authoring for the karenina framework through a validated five-step procedure. ## Core Features & Use Cases - Trait Type Selection: Maps each quality (safety, citations, readability, coverage, fact-checking) to the right trait type among LLMRubricTrait, RegexRubricTrait, CallableRubricTrait, MetricRubricTrait, and AgenticRubricTrait. - DynamicRubric Support: Builds conditional rubrics where traits are evaluated only when their concept is detected in the response. - Validation Workflow: Runs a validation script against generated rubric files and iterates on errors until the rubric passes. - Use Case: A researcher evaluating biomedical AI answers wants to score safety, citation presence, and key-point coverage. The Skill interviews them, generates the trait definitions from a skeleton, validates the file, and confirms the final rubric. ## Quick Start Create a karenina rubric that evaluates safety, citation quality, and coverage of key points for my question-answering benchmark.
