Introduction

Agent Skills are reusable, filesystem-based capabilities that extend Claude's functionality. They package domain-specific expertise, workflows, and best practices into discoverable components that Claude automatically uses when relevant.

A skill is simply a Markdown file (typically named SKILL.md) stored in the .claude/skills/ directory of your project. Each file contains instructions, rules, and examples that teach Claude how to handle a specific type of task — like writing database migrations, generating API docs, or following your team's code style. When you ask Claude to do something that matches a skill's domain, it automatically loads those instructions and follows them.

Skills matter because they let you encode expertise once and reuse it forever. Instead of repeating "always use our logging format" or "follow this deployment checklist" in every conversation, you write a skill and Claude remembers it across all sessions. They're especially useful for tasks that require consistency — brand voice, testing patterns, security rules, or framework-specific conventions.

Here's what a minimal skill file looks like at .claude/skills/sql-migrations/SKILL.md:

# SQL Migration Skill

When generating database migrations:
- Always use reversible migrations (include both up and down)
- Prefix filenames with a UTC timestamp
- Never drop columns in production — mark deprecated instead
- Use `ALTER TABLE ... ADD COLUMN` with a default value

You can also invoke skills explicitly using slash commands. For example, if you have a commit skill installed, typing /commit in Claude Code triggers that skill's workflow directly.

Key takeaway: Skills turn your one-off instructions into permanent, shareable knowledge that Claude applies automatically — think of them as "muscle memory" for your AI assistant.