skills-toolkit

Skills Toolkit for Claude Code

A plugin for creating and managing Claude Code skills and plugins.

About skill-composer vs Claude's skill-creator

This toolkit includes skill-composer, which differs significantly from Claude's built-in skill-creator:

Aspect	skill-composer	Claude's skill-creator
Philosophy	Convention enforcement upfront	Evaluation-driven iteration
Approach	Opinionated, enforced best practices from start	Learn via test feedback loops
Interviews	Multi-stage requirements gathering (purpose, scope, complexity, tools)	Minimal upfront requirements
Best Practices	Enforces from the start (frontmatter, 80% rule, references, token efficiency)	Discovered through empirical testing
Workflow	Design → Structure → Build → [Optional: Refine → Test]	Build → Test → Improve based on results
Output	Fully structured skill (SKILL.md, references/, scripts/ as needed)	Working skill you iterate on
Use When	Building skills with conventions you want enforced consistently	Building skills you'll test and improve empirically
Ecosystem	Part of full toolkit (refiner, tester, plugin-creator)	Standalone, integrates with evaluation

TL;DR: skill-composer enforces conventions upfront. Claude's skill-creator improves via evaluation loops.

Use skill-composer if:

• You want conventions enforced from the ground up
• You're building consistent, production-ready skills
• You value structured SKILL.md, references, and tool scoping upfront
• You want to avoid common skill-building mistakes by design

Use Claude's skill-creator if:

• You want to build and improve skills through empirical testing
• You prefer iterative refinement based on evaluation results
• You're comfortable discovering best practices via feedback loops
• You want to measure skill effectiveness with benchmarks

Table of Contents

• Available Skills
  • skill-composer
  • skill-refiner
  • skill-tester
  • plugin-creator
  • hook-creator
  • subagent-creator
  • ask-user-question
• Quick Start: From Knowledge to Plugin
• Installation
• Usage Scenarios
• Design Notes: Architecture & DRY
• Inspiration
• License
• Author

skill-composer

Command: /skills-toolkit:skill-composer

Interactive guide for creating new skills from scratch:

• Name and describe your skill
• Structure SKILL.md (frontmatter + instructions)
• Set up reference documentation
• Configure tool permissions
• Apply best practices during creation

skill-refiner

Command: /skills-toolkit:skill-refiner

Interactive guide for improving and validating existing skills:

• Refine skill structure and clarity
• Validate against best practices
• Reduce token usage and improve efficiency
• Check production readiness
• Apply the 80% rule for optimization

skill-tester

Command: /skills-toolkit:skill-tester

Empirically test and benchmark Claude Code skills:

• Quick Workflow: Fast pass/fail validation of new skills
• Full Pipeline: Comprehensive benchmarking with baseline comparison
• Measure impact: Pass rates, token usage, timing metrics
• Iteration tracking: Compare skill performance across refinement cycles
• Data-driven improvements: Prove skills help with empirical evidence

plugin-creator

Command: /skills-toolkit:plugin-creator

Interactive guide for creating and validating plugins:

• Generate .claude-plugin/plugin.json manifest
• Organize skills, commands, hooks, MCP/LSP servers
• Convert existing projects to plugins
• Configure installation scope (user/project/managed)
• Validate plugin structure

hook-creator

Command: /skills-toolkit:hook-creator

Interactive guide for creating and validating hooks:

• Build plugin hooks from scratch (command, prompt, agent types)
• Validate existing hooks against best practices
• Refine hook quality (event matching, error handling, performance)
• Configure event matching, matchers, and error handling
• Test hooks with validation workflows

subagent-creator

Command: /skills-toolkit:subagent-creator

Interactive guide for creating and validating subagents:

• Build subagents with clear delegation signals
• Configure tool access with permission modes
• Set up hooks for agent coordination
• Validate against best practices for reliability

ask-user-question

Command: /skills-toolkit:ask-user-question