This guide teaches users how to build custom skills for Claude, which are folders containing instructions that teach the AI to handle specific tasks or workflows. It is written for developers, power users, and teams who want Claude to follow consistent processes without re-explaining preferences each time. The document opens with an introduction explaining what skills are and how they work,…
This guide teaches users how to build custom skills for Claude, which are folders containing instructions that teach the AI to handle specific tasks or workflows. It is written for developers, power users, and teams who want Claude to follow consistent processes without re-explaining preferences each time.
The document opens with an introduction explaining what skills are and how they work, including the three-level progressive disclosure system (YAML frontmatter, SKILL.md body, and linked files). It covers core design principles such as composability and portability across Claude.ai, Claude Code, and the API.
Chapter 1 covers fundamentals: the required SKILL.md file structure, optional scripts, references, and assets folders. It also explains how skills complement MCP servers using a kitchen analogy where MCP provides tools and skills provide recipes.
Chapter 2 focuses on planning and design, starting with use case definition and three common categories: document and asset creation, workflow automation, and MCP enhancement. It includes success criteria with both quantitative and qualitative metrics.
The guide details technical requirements including file naming conventions (kebab-case), YAML frontmatter fields (name, description, license, compatibility, metadata), and security restrictions. It provides templates for writing effective descriptions and main instructions with step-by-step workflows and troubleshooting sections.