What problem does it solve? Writing an effective Agent Skill is hard to get right on the first try: descriptions may fail to trigger on relevant queries or over-trigger on unrelated ones, and there is no built-in way to measure whether a skill actually improves agent output. This Skill provides a complete draft-test-review-improve loop with quantitative benchmarks. ## Core Features & Use Cases - Skill Authoring Guidance: Step-by-step workflow for capturing intent, writing SKILL.md frontmatter and instructions, and structuring scripts, references, and assets. - A/B Evaluation Harness: Runs test prompts with and without the skill via subagents, grades outputs against expectations, and aggregates pass rates, timing, and token usage into benchmark.json and an HTML review report. - Description Optimization Loop: Iteratively rewrites the skill description using train/test trigger evals to maximize triggering accuracy, with blinded history to avoid overfitting. - Packaging: Bundles a validated skill folder into a distributable .skill zip file. - Use Case: You wrote a skill for generating release notes but the agent rarely uses it. Run the trigger eval loop to test 20 realistic queries, let the optimizer rewrite the description, and verify the improved triggering rate in the HTML report. ## Quick Start Ask the agent to create a new skill for your task, then run the evals and review the benchmark report to iteratively improve it.
