How to create your own Claude skill (step by step)

A Claude skill is a folder of instructions, and optionally templates and scripts, that Claude loads when a task calls for it. If you can write down how you do a task, you can make a skill. No coding is required for most of them.

This guide builds a real example step by step: a skill that writes a client follow-up email after a discovery call, in a consistent house style. New to skills? Start with how to use Claude skills.

Step 1: Pick a repeated task

Good skill candidates share three traits:

• You do it often, at least weekly.
• There is a right way to do it: a format, a checklist, a tone.
• You keep re-explaining it to Claude or to people.

Examples: weekly reports, proposal drafts, meeting follow-ups, CRM data cleanup, social post formatting, onboarding checklists.

Step 2: Write down how you do it today

Before touching any files, write the process in plain language as if training a new hire. For the follow-up email:

1. Read the call notes.
2. Open with thanks and one specific detail from the conversation.
3. Summarize what we heard as their top two or three problems.
4. Propose the next step, with a date.
5. Keep it under 200 words, warm but direct, no jargon.
6. Never mention pricing in the first follow-up.

That list is 80 percent of your skill.

Step 3: Create the folder and SKILL.md

A skill is a folder. The only required file is SKILL.md.

discovery-follow-up/
├── SKILL.md            required: name, description, instructions
├── example-email.md    optional: a great finished example
└── templates/
    └── follow-up.md    optional: a fill-in template

SKILL.md has two parts: a short header between --- lines (called frontmatter), and the instructions below it.

Here is the complete file for the example:

---
name: discovery-follow-up
description: Writes the follow-up email after a client discovery call, in
  Marketproof's house style. Use when the user shares call notes or asks
  for a follow-up, recap, or next-steps email after a first meeting.
---

# Discovery call follow-up

## Steps
1. Read the call notes the user provides.
2. Open with thanks and one specific detail from the conversation.
3. Summarize the client's top two or three problems in their words.
4. Propose one clear next step with a suggested date.
5. Close warmly, signed with the user's first name.

## Rules
- Under 200 words.
- Warm, direct, no jargon or buzzwords.
- Never mention pricing or timelines we have not agreed.
- If the notes are missing the client's name or next step, ask first.

## Example
See example-email.md for a finished email that meets these rules.

Two rules for the header, from Anthropic's documentation: the name can only use lowercase letters, numbers, and hyphens (up to 64 characters) and cannot include the words "anthropic" or "claude"; the description can be up to 1,024 characters.

Step 4: Write a description that triggers at the right time

The description is the most important line in the skill. Claude sees only the name and description of every installed skill, and uses them to decide which one to load. A vague description means the skill is ignored or used at the wrong time.

Step 5: Add examples and templates

One excellent finished example teaches more than a page of rules. Add example-email.md with an email you would be proud to send (with client details changed). If the output has a fixed structure, add a template too.

Keep SKILL.md itself short and focused (Anthropic's guidance for Claude Code is to keep it under about 500 lines) and move long reference material into separate files that the instructions point to.

Skills can also include scripts for steps that need exact, repeatable results, such as calculating figures or checking a file's format. That is optional; many useful skills are instructions only.

Step 6: Test it on real tasks

In the Claude app

1. Zip the skill folder (the zip should contain the folder itself).
2. In Claude's settings, open the skills section and upload the zip. Make sure it is turned on.
3. Start a new conversation and paste real call notes: "Here are my notes from today's call with a client. Draft the follow-up."

In Claude Code

Place the folder in ~/.claude/skills/ to use it in all your projects, or in .claude/skills/ inside a project to share it with everyone working there. Then ask for the task, or type /discovery-follow-up to run it directly.

What to check

• It loads when you ask in different ways ("follow-up," "recap email," "next steps").
• It does not load for unrelated requests.
• The output follows every rule, every time, across five or more real examples.
• It asks for missing information instead of inventing it.

Step 7: Refine, then share

Every time you correct the output, ask: would this correction apply next time? If yes, add it to the skill's rules. After a few rounds, the skill produces work you barely need to edit.

To share it, send the zip to teammates for the Claude app, or commit the folder to your team's repository for Claude Code.

Common mistakes

• A description that only says what, not when. Include the phrases people use to ask for the task.
• One skill for everything. Split "all our client emails" into separate skills per email type.
• Rules without an example. Add one finished example of great output.
• Secrets in the skill. Never put passwords or API keys in skill files.

Common questions

Can a skill use my company's documents?

Yes. Include reference files in the skill folder, or use the skill inside a Claude Project that holds the documents. See how to build an AI knowledge base.

How many skills can I have?

Many. Claude keeps only each skill's short description in mind until one is needed, so installed skills cost very little.

Are skills safe?

Skills you write yourself are as safe as their instructions. Treat skills from others like software: read every file before installing, and only use sources you trust.