How to use Claude Skills in Claude Code?

TL;DR

Claude skills let you store reusable instructions as skill files with metadata and body. Claude then uses progressive disclosure to load only the necessary parts of a skill, plus optional files and folders, enabling targeted, efficient context. This video compares skills with MCP and shows a concrete example with Claude Code.

Transcript

Hey everyone, I'm Shaw. In this video, I'll explain Claude's new skills feature. I'll start by discussing what this is and how it works, highlight how it fits together with existing concepts like MCP and sub agents and then finally walk through a concrete example of using skills with Claude code. So, what are Claude skills? These are just reusable ... Read More

Key Insights

• Skills are reusable, specialized instructions that Claude can access when needed.
• A skill consists of metadata and a body; metadata is short, body contains the detailed instructions.
• Claude looks up and uses the most relevant skill to a given use case.
• Skills can be organized in folders and extended with additional files and even external tools.
• Progressive disclosure limits context window usage by injecting content only when needed.
• Skills are designed to reduce token waste and focus Claude’s attention on relevant tasks.
• MCP and skills overlap but function differently, with skills remaining Claude specific.
• Claude Code can run scripts and access tools via its skills ecosystem.

Questions & Answers

Q: What are Claude skills in simple terms?

Claude skills are bundles of reusable instructions that Claude can access when it needs to perform a task. Each skill has a minimal metadata description and a longer body of instructions. This setup lets you write instructions once and reuse them across conversations, making Claude more efficient and responsive to context.

Q: How is a skill structured and stored?

A skill is stored as a folder named after the skill containing a skill.md file. The skill.md has two parts: metadata with a short name and description, and the body with the actual instructions. The metadata is compact so Claude can load it quickly, while the body provides the full guidance when the skill is invoked.

Q: What is progressive disclosure in Claude skills?

Progressive disclosure is the process where Claude initially loads only the skill metadata to keep the initial context small. When the skill is needed, Claude loads the full skill body, and it can also access additional files or folders referenced within the skill. This approach minimizes token usage while keeping capabilities accessible.

Q: How do skills differ from MCP?

Skills are Claude specific and work with Claude Code, while MCP is a universal standard that enables any LLM to talk to any application. Skills load content gradually to manage context, whereas MCP can dump broader tool metadata into the context window at the outset, which can be far larger in token usage.

Q: Can skills access tools or run code?

Yes, skills can reference tools and even run code. Claude has access to a bash shell, Python, and NodeJS, so a skill can invoke scripts or commands like python script/fu.py to perform functions. This allows Claude to extend its capabilities with task-specific utilities.

Q: What is a concrete example of a skill in this video?

The video demonstrates an AI Tutor skill as a concrete example. It shows how you can define the tutor’s instructions, organize them, and then have Claude apply them in real conversations. The example illustrates how metadata and body content guide Claude to explain technical concepts and validate ideas.

Q: How can you organize skills for large workflows?

You can organize skills into folders to manage many instructions and tools. This lets Claude read a subset of content as needed and drill into not just one skill but multiple related ones. Additional folders can hold more how-tos or scripts, enabling scalable, modular workflows without overwhelming the context window.

Q: Why is Claude skills useful for repeated tasks?

Claude skills are useful because they let you write complex instructions once and reuse them across sessions. With progressive disclosure, Claude only loads what is needed for the current task, reducing token usage and keeping the conversation focused. This makes repetitive workflows faster and more predictable.

Summary & Key Takeaways

• The video introduces Claude skills as reusable instruction bundles stored in skill.md files with metadata and a body. It explains how Claude reads only the metadata at startup and loads the body when the skill is invoked, reducing context load.

• It explains progressive disclosure, where Claude gradually reveals content from skills, folders, and tools as needed during a session.

• A concrete example demonstrates an AI Tutor skill, showing how skills are organized, accessed, and extended with additional files and folders for more complex workflows.