← Back to all free guides

Learn Claude
Skills
in 10 Minutes

Everyone overcomplicates Skills. You can build one, install it and be using it in 10 minutes. Here’s exactly how.

A Skill is a saved set of instructions Claude pulls in on its own when your request matches. Instead of re-typing "write this in my voice, under 4 sentences, casual, always end with a next step" every time, you save it once and Claude applies it whenever you ask for an email.

Under the hood it’s a folder with one file called SKILL.md. The top holds a name and a short description that tells Claude when to use it. Below that are the instructions. No code needed.

You Need

Any Claude plan, including Free, on the web, desktop or mobile app. Turn on Code execution and file creation in Settings, then Capabilities (it may read Create and edit files in the newer layout). Skills need it.

(Anthropic: Use skills in Claude)

Build Your First Skill

01

Start a chat and ask for a skill

Open a new chat. Pick the strongest model your plan offers. Claude has a built-in skill-creator that it reads when you ask it to make a skill, so you don’t need to set anything up. Paste this and fill in the bracket.

Build A Skill
Use your skill-creator skill to help me build a Claude Skill for [THE TASK YOU KEEP RE-EXPLAINING, e.g. "writing client follow-up emails in my voice"]. How I want this to go: 1. Interview me first, one question at a time. Ask what the skill should do, when it should and should NOT run, the tone, what every output must include, what it must always avoid, and ask me for one example of a great result. Wait for each answer. 2. If I attach examples, templates or past work, study them and tell me the patterns you see before you write anything. 3. Write the SKILL.md. Use a short lowercase-hyphen name. Keep the description under 200 characters and make it specific about when to use it, plus one "Do NOT use for" line. 4. Show me the full file and three test requests: one that should trigger it, one borderline, and one that should NOT trigger it. Tell me how you expect each to go. 5. Revise until I say it's done, then package it as a ZIP with the skill folder at the root so I can upload it. Rules: Don't invent details about my process. If something is unclear, ask. Keep the instructions plain and specific, with no filler.

Need an idea? Try follow-up emails in your voice, or turning meeting notes into action items.

02

Answer the interview out loud

Your answers decide how good the skill is forever. Claude asks what it should do, the tone, what to always include and avoid, and what a great example looks like.

Talk your answers instead of typing them. Use the microphone in the Claude app or a dictation tool like Wispr Flow. You’ll naturally give way more detail, and Claude will organize the rambling.

When Claude shows the SKILL.md, push back on anything that feels off: "make the tone more casual" or "always include a subject line."

03

Install it

Save the skill when Claude packages it. If you get a ZIP file instead, install it yourself:

  1. Go to Customize (being renamed Plugins), then Skills.
  2. Click +, then Create skill, then Upload a skill, and pick the ZIP. Don’t unzip it first.
  3. Make sure the toggle next to it is on.

Building it yourself? Zip the folder so the skill folder sits at the root of the ZIP, and make the folder name match the skill’s name.

04

Test it in a fresh chat

Open a new chat and ask for the task in a few different ways. When the skill runs, you’ll see Claude mention it in its thinking. If it doesn’t fire, or fires on the wrong things, run this.

Fix A Skill That Won’t Fire
I have a skill called [SKILL NAME] that isn't firing when I expect it to (or fires when it shouldn't). 1. Read its description back to me word for word. 2. Here are 3 requests where it SHOULD have run: [PASTE]. Here are 2 where it should NOT run: [PASTE]. For each, tell me honestly whether the current description would make you use the skill, and why. 3. Point to the exact words that are too vague or too narrow. 4. Write 2 improved descriptions, each under 200 characters, that would fire on the first list and stay quiet on the second. Don't touch the instructions yet. The description is what you read to decide, so fix that first.

Common Fix

Most skill problems live in the description, not the instructions. Too vague ("helps with emails") and it fires on everything. Too narrow ("only for follow-ups to enterprise SaaS prospects") and it never fires. Aim for the middle.

What A SKILL.md Looks Like

You don’t have to write this by hand, but it helps to know what you’re looking at.

Example SKILL.md
--- name: email-drafter description: Drafts professional emails in the user's voice. Use when the user asks to write, draft, or compose any email, message, or reply. Do NOT use for casual brainstorming, summarizing, or analysis tasks. --- # Email Drafter ## Instructions When the user asks you to draft an email: 1. Ask who it's to and the key point (if not provided) 2. Write in a warm but professional tone 3. Keep it under 150 words unless specified 4. Always include a clear subject line 5. End with a specific next step or call to action 6. Never use "I hope this email finds you well" ## Examples [Example input and output here]
  1. The description is the trigger. Claude reads it to decide when to use the skill. The claude.ai limit is 200 characters, and the name can be up to 64. Anthropic’s own skill-creator says Claude tends to undertrigger skills, so make the description a little pushy about when to run.
  2. The instructions are the how. "Write a good email" is useless. "Under 150 words, warm but professional, always end with a next step" works.

(Anthropic: How to create custom skills and the skill-creator skill)

Find More Skills
  1. The directory. In Customize (or Plugins), open Skills, click +, then Browse skills and click Add (some layouts say Install). Plugins and connectors live in the same directory under their own tabs. Plugins bundle several skills together.
  2. Connect your real tools. Connectors (Gmail, Google Calendar, Drive, Slack and more) give skills real data to work with. A morning-briefing skill gets a lot better when it can actually read your calendar.
  3. Anthropic’s open-source library. Browse github.com/anthropics/skills, download any folder, zip it and upload it the same way.
  4. My ready-made ones. I built 10 professional skills you can copy right now, from status reports to meeting recaps.

(Anthropic: Browse skills, connectors and plugins)

Pro Tips

01

Always add a Do NOT use line

List what the skill should stay away from. That one line kills most false triggers.

02

Test one before you build five

Build it, test it, fix it, then move on. Bugs stack up fast when you batch.

03

Pair skills with Projects

A skill running inside a Project also gets that Project’s files and instructions. Same skill, much better output.

04

Don’t fear a big library

Claude only keeps each skill’s name and description in view and loads the full instructions when a task matches. Turn off any skill that keeps firing when it shouldn’t.

05

Review them monthly

Your workflows change. Once a month, reread your top five skills and update anything stale.

Want more? Read 5 Skill Hacks Nobody Talks About or Why Your Skill Isn’t Working.