skill-creator

What problem does it solve? Most agent skills are written once and never measured, so they fail unpredictably in production. This Skill enforces a test-driven workflow for building skills: write failing baseline tests before drafting, run evals, benchmark results, and iterate until the skill measurably works. ## Core Features & Use Cases - Test-first skill creation: Requires 3+ pressure scenarios run without the skill (RED phase) before writing any SKILL.md content, proving the skill is actually needed. - Eval and benchmark loop: Spawns with-skill and baseline runs, grades outputs against assertions, aggregates pass rates and token usage with scripts, and renders an HTML eval viewer for human review. - Description optimization: Generates trigger eval queries and runs an optimization loop to tune the frontmatter description for reliable skill triggering. - Use Case: You want to build a skill that formats sales spreadsheets. This Skill walks you through capturing intent, writing failing baseline tests, drafting the SKILL.md, running benchmark evals, and refining until pass rates improve. ## Quick Start Ask the agent to help you create a new skill for your workflow and it will guide you through baseline testing, drafting, and evaluation.