What problem does it solve?
Skill-creator removes the friction of designing, validating, and iterating Claude-compatible skills by providing a repeatable, end-to-end workflow for drafting SKILL.md files, running trigger and functional evaluations, grading outputs, and packaging distributable skill bundles. It centralizes scaffolding, test-case generation, evaluation tooling, and a review UI so teams don't rely on ad-hoc prompts or manual benchmarking.
Core Features & Use Cases
- Draft & Scaffold: Create well-formed SKILL.md frontmatter and guidance for when a skill should trigger.
- Eval & Benchmark: Generate eval sets, run parallel trigger tests, grade outputs, aggregate pass rates, timing, and token metrics.
- Iterate & Improve: Automate description optimization, propose revisions, and run iterative train/test loops to avoid overfitting.
- Packaging & Review: Package skills into .skill archives, generate static review pages, and provide grader/comparator/analyzer agents for rigorous blind comparisons.
- Use Case: A developer or product manager drafts a new skill, seeds 2–3 test prompts, runs a 3-iteration optimization loop, reviews results in the eval viewer, and packages the final skill for distribution.
Quick Start
Tell the skill-creator to draft a new SKILL.md for your desired workflow, add two realistic test prompts to evals/evals.json, run the eval loop to generate a benchmark, and produce a packaged .skill.
