Skill UI

Formal evaluation framework for Claude Code sessions that applies evaluation-driven development to define capability and regression checks, run deterministic and model graders, and track pass@k/pass^k reliability metrics before shipping.

Master the art of evaluating AI, LLM, and agent systems. Learn to decide when an evaluation is needed, design robust experiments, choose appropriate graders, and interpret results under hostile audit. Covers everything from spot checks to CI gates and benchmarking.

Builds the Phase 0 evaluation harness that gates every fine-tuning run: golden sets from trace error analysis, per-failure-mode graders, calibrated LLM judges, drift suite, and base-model baseline.