AI/ML
///Advanced
LLM Evaluation Harness
Builds a repeatable evaluation harness for LLM prompts, models and pipelines: curates golden datasets, defines graded rubrics and pass thresholds, scores outputs on correctness and style, tracks regressions across prompt or model changes, and reports results with worked examples of failures. Use when shipping or changing prompts, comparing models, guarding quality in production, or when the user asks to evaluate, score or regression-test LLM output.
Preview
# LLM Evaluation Harness A prompt that works on Tuesday can silently regress on Thursday. This skill turns "the output looks fine" into a scored, repeatable benchmark: a golden dataset, explicit rubrics, automated scoring, and a regression gate that fails the
Related skills