Create Custom Skills in Claude AI (Step-by-Step Guide)

The single most valuable thing you can do to create custom skills in Claude AI takes under 20 minutes the first time — and pays you back in hours saved every week for as long as you use the tool. It is a markdown file, it needs no code, and once it is in the right place, Claude stops being a generic assistant and starts behaving like a specialist trained on your exact workflow.

A custom skill in Claude AI is a folder with a file named skill.md inside it. That file has two sections: YAML frontmatter that controls when Claude activates the skill, and a markdown body that tells Claude exactly what to do when it runs. There is no compilation, no deployment, and no registration — placing the folder in one of four recognized locations is the entire install step.

What a Claude AI Custom Skill Actually Is

A skill is a folder. Inside that folder lives skill.md, and that single file does all the work. The folder can also hold template files, reference documents, even scripts Claude can execute — and skill.md can point to every one of them by file name.

Skills follow the open agent skills standard, so the same skill.md is not locked to Claude. In Claude Code today, custom commands and skills are unified: a folder at .claude/skills/deploy/skill.md and a flat file at .claude/commands/deploy both create the same /deploy invocation. Skills are the recommended path because they support additional files, frontmatter controls, and automatic invocation — none of which a flat command file supports.

The skill.md Structure: Frontmatter and Body

Every skill.md has exactly two parts. The first is YAML frontmatter — the metadata block at the top of the file. The second is a plain markdown body — the instructions Claude follows when the skill runs.

Two frontmatter fields carry the most weight:

• name — a kebab-case identifier using lowercase letters, numbers, and hyphens. Skip it and the directory name becomes the skill name automatically.
• description — one sentence explaining what the skill does and when Claude should use it. This is the single most important line in the entire file.

Other useful frontmatter fields include disable_model_invocation (set to true for manual-only skills, such as a deploy command where you never want Claude deciding for you) and user_invocable (set to false for background reference skills you only want Claude to activate). The full reference lives in the official Claude skills documentation.

The body is plain markdown. Write it the way you would write an onboarding document for a new hire. Include these five sections: when to use the skill, the expected output structure, the tone and voice, the rules and constraints, and at least one template or worked example. Specific beats vague every time — a vague body produces vague output.

Why the Description Field Makes or Breaks Your Skill

The description is how Claude decides whether to activate your skill automatically. A vague description means Claude cannot identify the right moment to load it, and the skill only works when you type the slash command by hand.

A bad description looks like: help with reports. A good description looks like: generates a weekly status report in our company standard format, used when the user asks for a weekly update, sprint summary, team status, or Friday report. The second version lists four specific trigger phrases. When you say any one of them, Claude recognises the match and loads the skill.

The test that matters: could a brand-new Claude session with zero prior context correctly identify when to use this skill based only on this one sentence? If the answer is no, rewrite the description until it is yes. Every word in that sentence is earning its place — treat it that way.

Two invocation modes exist for every skill. Automatic invocation: Claude reads the description, detects relevance, and loads the skill without any command from you. Manual invocation: you type /skill-name and it runs immediately. Frontmatter flags let you restrict to either mode, depending on whether you want Claude making that call or keeping it firmly in your hands.

Where Skills Live and How Claude Finds Them

Skills can be stored in four locations, loaded in priority order:

• Personal — ~/.claude/skills/skill-name/skill.md. Available to you across every project on your machine.
• Project — .claude/skills/skill-name/skill.md inside a project folder. Available to anyone working in that directory.
• Plugin — distributed to anyone with the specific plugin enabled.
• Enterprise — distributed org-wide through managed settings.

Dropping the folder in the personal or project location is the install. There is no registration step, no restart required. Claude Code watches these directories for live changes and applies any edits on the very next conversation turn.

The Fast Way to Build Your First Skill

After training more than 79,000 students across 74 courses in AI and automation, the mistake I see most often is over-engineering the first skill. There is a faster approach: describe what you need to Claude and let it ask you the questions before writing a single line.

Open Claude, tell it you want to create a custom skill — a client proposal generator, a weekly report formatter, whatever fits your workflow — and ask it to interview you first. Claude will ask: What type of documents are you creating? What is your industry? Who are your clients? What sections do you want in the output? What tone should it use? What is the output format — plain text, markdown, something else? Do you have a standard pricing structure to reference?

Answer even rough versions of those questions and Claude generates a complete skill.md: well-formed frontmatter, a description with real trigger phrases, body sections covering output structure, tone, and constraints, and placeholder templates you can swap out immediately. Anthropic even maintains a skill creator skill in their own catalog — a meta-skill purpose-built for building other skills. Use it on your first one.

The skill folder can include supporting files alongside skill.md: template documents Claude pulls when the skill runs, brand style guides, pricing references, and sample inputs and outputs. Reference them in the body by file name and Claude reads them automatically whenever the skill activates.

Three Tests Every Skill Must Pass Before You Trust It

Before relying on any skill in a live workflow, run three checks:

• Trigger test — use the exact phrases from your description. Does Claude activate the skill automatically? Failure here means the description needs sharper, more specific trigger language.
• Output test — does the result match the structure, tone, and format you specified in the body? Drift here means the body needs tighter examples and explicit constraints.
• Edge case test — what happens when input is incomplete? Claude should ask for the missing pieces rather than silently guess. If it guesses, add explicit handling instructions to the body.

Trigger failures trace back to the description. Output failures trace back to the body. Fix the right section and most issues resolve in a single pass.

A skill under 100 lines, written once, eliminates the same repetitive prompt you type ten times a day. Write the skill.md, run the three tests, drop it in your personal skills folder, and Claude picks it up on the next turn with no further action required. That is the complete workflow to create custom skills in Claude AI that hold up in production.

---

Keep Learning

If this was useful, these are worth reading next:

• My 11-Year-Old Got Certified by Sheikh Hamdan's AI Initiative. Here's What He Built With It.
• Fix Broken AI Automations (Claude AI Troubleshooting Guide)
• Or go further with the AI Mastery Course — used by 79,000+ students across 150+ countries.

Method	Setup Time	Auto-Invokes	Supports Files	Best For	Cost
Claude Skill (skill.md)	15-25 min	Yes (via description)	Yes (folder + scripts)	Repeated workflows, team standards	Free with Claude Pro $20/mo
Claude Slash Command (flat file)	5 min	No (manual /trigger)	No	Quick prompts you invoke yourself	Free with Claude Pro
Claude Project (Pro/Team)	10 min	Within project only	Yes (knowledge files)	Single workspace, non-technical users	$20-25/user/mo
Custom GPT (OpenAI)	20-30 min	No (must select GPT)	Yes (up to 20 files)	Public-facing assistants	ChatGPT Plus $20/mo
Anthropic API + System Prompt	2-4 hours (code)	Programmatic only	Yes (full control)	Production apps, customer products	$3-15 per million tokens

Source: Anthropic Skills Documentation, Anthropic pricing, OpenAI pricing — verified May 2026.