What Are Claude Code Skills?
Claude Code skills are one of its most powerful features — and most beginners don't know they exist. Here's exactly what they are and how to use them.
Sam Okafor is a fictional AI persona, not a real person. This article was written by AI and reviewed by a human editor before publishing. How we work →

If you've been using Claude Code, you're probably using it the basic way — open a project, start chatting, ask it to write code. That works fine. But there's a layer on top of that most people miss entirely: Skills.
What Is a Skill?
A Skill is a markdown file that gives Claude specialized knowledge for a specific task. When a skill is installed in your project, Claude automatically knows when to use it based on what you're asking.
Think of it like hiring a specialist. Regular Claude is a generalist. A skill turns Claude into an expert in one specific domain — UI design, marketing copy, code reviews, whatever the skill covers.
Where Do Skills Live?
Skills go in one of two places:
# Project-specific skill (only for this project)
your-project/.claude/skills/<skill-name>/SKILL.md
# Personal skill (available across all your projects)
~/.claude/skills/<skill-name>/SKILL.md
The difference matters. A skill for your company's design system belongs in the project folder. A skill for writing better commit messages belongs in your personal folder so every project can use it.
How Does Claude Know When to Use a Skill?
This is the clever part. Every skill has a description field at the top — and that description is what Claude reads to decide whether to activate it.
---
name: ui design system
description: Use when building UI components, choosing colors, picking fonts, or reviewing design code
---
When you say "help me pick a color palette for my dashboard," Claude sees that description and knows to pull in the design skill. You don't have to manually invoke it.
A Real Example: UI UX Pro Max
UI UX Pro Max is a skill that gives Claude a full design intelligence system — 67 UI styles, 96 color palettes, 57 font pairings, and guidelines for 13 tech stacks.
Without the skill, if you ask Claude "what colors should I use for my SaaS dashboard?" you get a generic answer. With the skill installed, Claude runs a reasoning engine against industry-specific rules and returns a complete, opinionated design system tailored to your product.
That's the difference a skill makes.
How to Install a Skill
- Download or clone the skill files
- Copy the skill folder into
.claude/skills/in your project (or~/.claude/skills/for personal use) - That's it — Claude picks it up automatically on the next session
Skills vs CLAUDE.md
You might be wondering: can't I just put instructions in my CLAUDE.md file instead?
You can, but skills are better for complex, reusable knowledge. CLAUDE.md is for project-specific rules and context — your tech stack, your build commands, gotchas Claude can't figure out from the code. Skills are for deep domain expertise you want to apply repeatedly, potentially across multiple projects.
Use both. They work together.
What Skills Are Available?
The ecosystem is still early, but growing fast. A few worth knowing about:
- — design system generation for any stack
- — 33 skills covering SEO, copywriting, email, ads, and CRO
- Community-built skills on GitHub for code review, testing patterns, Docker deployment, and more
You can also write your own. If you find yourself giving Claude the same context over and over, that's a sign you should turn it into a skill.
The Bottom Line
Skills are the fastest way to level up what Claude Code can do without writing a single line of code. If you're a beginner using Claude to build apps, installing one or two domain-specific skills is probably the highest-leverage thing you can do today.
From the comments
AI personas · answered by the authorQuick one before I go installing a bunch of these: does loading a skill cost me extra tokens every session, or only when Claude actually decides to use it?
Good instinct to ask. The article's key mechanic is that every skill has a `description` field, and that description is what Claude reads to decide whether to activate it. So the lightweight part (the description) is what's checked against your request — the full skill body only matters once Claude pulls it in for a relevant task. The piece I can't promise you is exact token accounting, because the article doesn't quantify that. If surprise usage is your worry, keep your installed set small and domain-specific rather than hoarding skills you rarely trigger.
Makes sense. So the move is install one or two I'll actually use, not a whole pile.
That matches the article's own advice — it says installing one or two domain-specific skills is probably the highest-leverage thing a beginner can do. It frames a couple, not a pile.
The example skill here is a hosted thing on someone's domain, but the article also says skills are just markdown files you copy into a folder. Can I get everything from open community sources instead of relying on a vendor?
Yes, that's consistent with what the article describes. A skill is defined as a markdown file (`SKILL.md`) that lives in `.claude/skills/` in your project or `~/.claude/skills/` for personal use — nothing about that requires a vendor. The article explicitly points to community-built skills on GitHub for things like code review, testing patterns, and Docker deployment, and it says you can write your own. UI UX Pro Max is just the worked example, not a requirement.
And if I write my own, it's the same install path? No special registry?
Same path. The install steps in the article are just: get the skill files, copy the folder into `.claude/skills/` (or `~/.claude/skills/`), and Claude picks it up on the next session. The article also notes that if you keep giving Claude the same context repeatedly, that's the signal to turn it into your own skill.
If a skill and my CLAUDE.md both try to tell Claude how to do the same thing, which one wins?
The article doesn't lay out a conflict-resolution or precedence rule, so I won't invent one. What it does say is that the two are meant for different jobs: CLAUDE.md is for project-specific rules and context — your stack, build commands, gotchas — while skills are for deep, reusable domain expertise you apply repeatedly, potentially across projects. The guidance is to use both together rather than have them compete. If you scope each to what it's actually for, you mostly avoid the overlap in the first place.
Separate the concerns and there's nothing to fight over. Fine.
The StackBrief weekly
New reviews and the AI-coding-tool news worth knowing — with our take. One email a week, unsubscribe anytime.
Keep reading

What Is a CLAUDE.md File? (And How to Set One Up)
Learn what a CLAUDE.md file is, why Claude Code reads it automatically, and how to write and set one up for your project in under 10 minutes.
May 8, 2026
What Are Claude Code Agents? A Plain-English Guide
Claude Code agents act on your behalf — not just chat. Learn what Claude Code agents are, how subagents work, and when you'd actually use them.
March 15, 2026
Claude Code vs Claude.ai: Skills, Agents, and Workflows
Confused about Claude Code vs Claude.ai? This plain-English guide explains skills, agents, and workflows — and which ones actually apply to your subscription.
May 12, 2026