Skill Development

To create a Claude Code skill, structure your project with a SKILL.md file, a scripts/ directory, and a references/ folder. Follow a development workflow that progresses from understanding use cases to writing imperative instructions and testing the modular capabilities.

Progressive disclosure is a best practice for AI agent skills that structures documentation to reveal complexity gradually. It ensures skills remain discoverable and efficient by guiding the agent through foundational concepts before introducing advanced scripting and reference materials.

Claude Code plugin documentation should use an imperative or infinitive writing style in the third person. This approach ensures instructions in the SKILL.md file are clear, actionable, and easily processed by AI agents executing modular tasks.

To structure API documentation references for an AI agent skill, place the reference documents inside the references/ directory. This separation from the main SKILL.md allows the agent to access detailed API schemas only when needed for specific tasks.

Yes, you can include existing scripts in the scripts/ directory when developing modular capabilities for Claude Code. This allows you to integrate pre-built automation logic directly into your AI agent skills alongside the main SKILL.md instructions.

The workflow for refining Claude Code skills involves understanding use cases, structuring the SKILL.md anatomy, applying metadata quality standards, and conducting iteration and testing. This ensures the final plugin capabilities are effective and efficient.