skill-creator

Create and iteratively refine Claude Code skills from intent capture to evaluation.

Updated Jan 16, 2026
One-click install
npx skills add https://github.com/kumewata/dotfiles --skill skill-creator-kumewata
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: skill-creator
Source: https://github.com/kumewata/dotfiles/tree/main/config/agents/skills/skill-creator
Command: npx skills add https://github.com/kumewata/dotfiles --skill skill-creator-kumewata

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires anthropic, pyyaml, and includes scripts (resource) components.

What problem does it solve?

The Skill Creator provides a structured, repeatable workflow for building Claude Code skills—from capturing user intent and drafting SKILL.md, to designing eval prompts, running iterative experiments, and refining the skill based on feedback.

Core Features & Use Cases

  • Capture Intent: Define what the skill enables and when it should trigger.
  • Interview and Research: Gather edge cases, dependencies, and success criteria to inform test prompts.
  • Write the SKILL.md: Create a frontmatter with name and description, followed by a concise, actionable instruction body.
  • Skill Writing Guide: Guidance on progressive disclosure, descriptions, and testing patterns.
  • Progressive Disclosure: Four-level loading pattern: Metadata, SKILL.md body, and optional bundled resources.
  • Reference and Tooling: Optional scripts/references/assets to support deterministic tasks and deeper knowledge.

Quick Start

Draft a new SKILL.md, prepare a small eval set, and run the loop to start iterating toward a polished skill.

Frequently Asked Questions about skill-creator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create and refine Claude Code skills with a structured workflow?

To create and refine Claude Code skills, this Skill captures user intent, drafts a SKILL.md with frontmatter, designs evaluation prompts, and runs iterative experiments to optimize the skill based on feedback.

What is progressive disclosure in Claude skill development?

Progressive disclosure in Claude skill development is a four-level loading pattern that exposes metadata, the SKILL.md body, and optional bundled resources incrementally to optimize context usage.

How do I write a SKILL.md file with frontmatter for Claude automation?

Writing a SKILL.md file requires creating a frontmatter with name and description fields, followed by a concise, actionable instruction body that defines when the skill should trigger and what it enables.

Do I need pyyaml and anthropic dependencies to build Claude skills?

Yes, you need the anthropic and pyyaml dependencies to run the evaluation loops and parse frontmatter when building and iteratively refining Claude Code skills.

What's the best way to evaluate and iterate on a Claude Code skill?

The best way to evaluate a Claude Code skill is to prepare a small evaluation set, run iterative experiments using designed prompts, and refine the skill based on the resulting feedback loop.

Why does my Claude skill workflow need an intent capture and interview phase?

An intent capture and interview phase is needed to gather edge cases, dependencies, and success criteria that inform the test prompts and ensure the skill handles real-world scenarios correctly.