What problem does it solve?
Building AI skills that reliably trigger and perform well requires more than good instructions — it needs systematic testing, measurement, and iterative refinement. Without structured evaluation, skills undertrigger, behave inconsistently, or fail on edge cases, leaving developers unsure whether their changes actually help.
Core Features & Use Cases
- Structured Skill Creation: Capture user intent, draft SKILL.md files with proper frontmatter and progressive disclosure, and organize resources into scripts, references, and assets.
- Quantitative Evaluation: Run parallel test cases with and without the skill, grade outputs against assertions, and aggregate benchmark statistics to measure real impact.
- Description Optimization: Automatically optimize triggering descriptions using train/test splits to improve when the skill activates, addressing common undertriggering issues.
- Use Case: You want to build a skill that automates invoice data extraction. Use this skill to draft the initial SKILL.md, create test prompts with sample invoices, run benchmarks comparing with/without the skill, and iteratively improve the description and instructions until it reliably extracts line items and totals.
Quick Start
Use the skill-creator skill to build a new skill for extracting data from invoices by describing what you want it to do, then test it with sample prompts and refine based on the benchmark results.
