What problem does it solve? Evaluating whether a skill actually improves on the bare model, whether a section earns its place, or whether a description triggers correctly is usually guesswork. This Skill runs structured evals against a skill's cases and reports evidence-backed results instead of impressions. ## Core Features & Use Cases - Four eval modes: baseline (skill vs bare model), variant (full vs stripped version), quality (output vs rubric via a read-only grader), and trigger (description firing measured through a platform adapter). - Isolated, reproducible runs: each case executes in a clean working directory with a from-scratch environment, staged fixtures, and immediate timing/token capture to timing.json. - Description optimization loop: near-miss query generation, stratified train/test split, and blinded iterative rewriting to improve trigger accuracy. - Use Case: After editing a skill's description, run trigger mode to measure whether it fires on the right queries and stays quiet on near misses, then pick the winning description by held-out test score. ## Quick Start Run the evals for the skill at path/to/my-skill in baseline mode and tell me whether it beats the bare model.
