What problem does it solve? Writing an effective skill for an AI agent is hard to get right on the first try: descriptions undertrigger, instructions overfit to examples, and there is no systematic way to know whether a change actually helped. This Skill provides a complete authoring and evaluation workflow for creating new skills, improving existing ones, and measuring their performance with quantitative benchmarks and trigger-accuracy tests. ## Core Features & Use Cases - Guided skill authoring: Captures intent through interview questions, drafts SKILL.md files following progressive-disclosure and naming conventions, and detects the correct target directory (UrhoX repo vs. user workspace). - Eval-driven iteration loop: Spawns parallel with-skill and baseline subagent runs, drafts assertions, captures timing/token data, grades outputs, and aggregates results into benchmark.json with pass-rate deltas. - Review viewer and feedback loop: Launches a local HTML viewer (or static file in headless environments) showing outputs and benchmark tabs, collects user feedback into feedback.json, and drives the next iteration. - Description optimization: Generates realistic should-trigger/should-not-trigger eval queries, runs an automated train/test optimization loop, and reports the best-performing description. - Use Case: A user says "I want to make a skill that converts meeting notes into Jira tickets." The Skill interviews them, drafts the skill, creates test prompts, runs with-skill vs. baseline comparisons, shows results in a browser viewer, and iterates until the user is satisfied. ## Quick Start Help me create a new skill that converts my weekly meeting notes into structured Jira tickets, then set up test cases to verify it works.
