What are the recommended token limits for skills?

For optimal performance, keep SKILL.md metadata around 100 tokens and the main body under 2,000 tokens, moving all other content to reference files.

How do I validate my new skill structure?

The skill-builder includes a 'validate-skill.js' script that checks for YAML syntax, token budgets, mandatory folder hierarchies, and link integrity.

Why is the /reference/ folder mandatory in this framework?

Placing all detailed markdown files in the /reference/ folder ensures Claude only loads them when specifically referenced, preventing context window saturation.

What is the 3-level pattern for Claude skills?

It is a structural hierarchy that loads metadata always, core instructions upon trigger, and detailed reference files only as needed to minimize token usage.

Claude Skill Builder

Name: Claude Skill Builder
Author: resolve-io

byresolve-io

•

개발자 도구

Guides the creation of high-performance Claude Code skills through token-optimized patterns and progressive disclosure architectures.

소개

The skill-builder provides a comprehensive framework for developing efficient, scalable Claude Code skills based on the .prism engineering methodology. It enforces a strict three-level loading pattern—separating metadata, core instructions, and detailed reference materials—to ensure optimal token usage and predictable AI behavior. By utilizing provided templates, checklists, and an automated validation script, developers can transform complex engineering workflows into actionable, maintainable AI skills that follow industry best practices like TDD, Clean Code, and Domain-Driven Design.

주요 기능

Three-level progressive disclosure architecture for context efficiency
Mandatory /reference/ folder structure for deferred resource loading
Automated validation script to verify skill structure and compliance
1 GitHub stars
Pre-built SKILL.md templates and step-by-step creation processes
Strict token optimization budgets (Metadata ~100, Body <2k tokens)

사용 사례

Implementing predictable AI behavior through structured documentation patterns
Building standardized Claude Code skills for custom engineering workflows
Refactoring existing AI skills to reduce token consumption and latency

소개

주요 기능

Three-level progressive disclosure architecture for context efficiency
Mandatory /reference/ folder structure for deferred resource loading
Automated validation script to verify skill structure and compliance
1 GitHub stars
Pre-built SKILL.md templates and step-by-step creation processes
Strict token optimization budgets (Metadata ~100, Body <2k tokens)

사용 사례

Implementing predictable AI behavior through structured documentation patterns
Building standardized Claude Code skills for custom engineering workflows
Refactoring existing AI skills to reduce token consumption and latency