What problem does it solve?
Building AI skills that work reliably in production requires more than good initial instructions—it needs rigorous testing, quantitative benchmarking, and iterative refinement to ensure consistent triggering and high-quality outputs. Without a standardized creation workflow, developers rely on ad-hoc trial and error that rarely measures what actually matters.
Core Features & Use Cases
- Structured Skill Creation: Guide users from intent capture to packaged skill file, including draft writing, test case design, and progressive disclosure architecture.
- Quantitative Evaluation: Spawn parallel with-skill and baseline runs, grade outputs against assertions, and aggregate results into statistical benchmarks with mean, stddev, and delta analysis.
- Description Optimization: Run automated train/test loops to iteratively improve skill descriptions, maximizing trigger accuracy while preventing overfitting to specific phrasings.
- Use Case: A developer building a custom code-review skill can use this to draft the SKILL.md file, run 10 eval prompts comparing with and without the skill, analyze pass rates and token costs, and optimize the description until it triggers reliably for relevant queries.
Quick Start
Use the skill-creator skill to build a new skill for summarizing meeting transcripts by drafting the SKILL.md file, creating test prompts, running evaluations with baselines, and iterating based on the benchmark results.
