← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

3 Ways To Build
A Claude Skill

There are three ways to build a Claude skill. Two of them are guessing. The third is the one I trust to run in my actual business, and here’s exactly how to do it.

A skill is a saved set of instructions that teaches Claude to do a repeatable job the same great way every time, so you never have to re-explain it. Under the hood it’s a folder with a SKILL.md file inside.

It’s like writing down exactly how you want a task done, once, so a new hire nails it forever. The question is how you write those instructions, and the three ways are not equal.

Way 1: Copy Someone Else’s Skill

The fastest way. You grab a skill someone already made, from Anthropic’s own skills repo, a GitHub search or one of my free guides, and upload it to Claude.

It’s a great place to start. A well-built skill for something generic, like formatting a weekly report or cleaning up a spreadsheet, works for almost anyone. But the second the job touches how your business runs, it starts to miss. A stranger’s skill knows nothing about your offers, tools or clients.

Use It For

Getting started fast and generic tasks. Weak for anything specific to how you operate.

Way 2: Describe It And Let Claude Build It

A big step up. You describe the skill you want and Claude writes it, using Anthropic’s built-in skill-creator. It asks about your process, then drafts the SKILL.md for you.

The gap is that you’re describing how you think you work, from memory, in one sitting. You forget the edge cases, the small judgment calls, the "except when it’s this kind of client" moments. Claude builds a clean skill for the version of the work in your head, not the real one.

Use It For

A solid first draft when you can’t demo the task live. Weak for capturing the calls you make without thinking.

Way 3: Do It Live, Then Save It

This is the method I trust for my own business. Instead of describing the work, you do the work with Claude, once, correcting it in real time. Then you tell it to save that whole session as a skill.

The skill gets built from a real run of the task, with your real corrections and your real standard of good baked in. Here’s the four-step loop.

01

Run the real workflow with Claude

Pick a task you actually do and do it with Claude on a real example. Not "imagine you’re writing an email," an actual email you need to send today.

02

Correct it as it goes

Every time Claude does something slightly off, say so. "Too formal, I’d never say that." "Skip that part." "This is exactly right, keep doing it." You’re teaching it the difference between good and bad by your standard.

03

Get it to nail the result

Keep going until the output is something you’d use without editing. That final version is your bar.

04

Save the whole process as a skill

Now tell Claude to turn everything you just did into a skill. It writes up the steps and every correction you made, so next time it does it your way on the first try.

Paste this to kick off Step 1, fill in your task, and run your real workflow with it.

Prompt 1: Run The Real Workflow
I want to build a reusable skill by doing a real task with you first, then saving how we did it. Do not build the skill yet. We are in training mode. THE TASK I WANT TO TURN INTO A SKILL: [describe the repeatable job in one or two sentences, e.g. "reply to a customer support email" or "turn my rough notes into a LinkedIn post"] A REAL EXAMPLE TO WORK ON RIGHT NOW: [paste or attach the actual input: the real email, notes, data or brief you're working on today, not a made-up one] HOW TO WORK WITH ME: 1. Do the task on this real example, start to finish, the way you think is best. 2. Go one section or one decision at a time. After each one, pause and ask "is this right, or what would you change?" before moving on. 3. When I correct you, treat it as a permanent rule for this kind of task, not a one-off fix. Restate the rule back to me in one line so I know you caught it. 4. Watch for the small judgment calls: tone, what to include vs cut, formatting, and what "done" looks like to me. Ask about them instead of assuming. 5. Keep a running list of the rules you've learned. Don't offer to save anything as a skill until I say we're finished. Start by attempting the task on the example above, then pause for my first round of feedback.

Once the output is good enough to send as-is, run this one. It’s Step 4.

Prompt 2: Save It As A Skill
That final version is exactly what I wanted. Turn everything we just did into a Claude skill I can reuse on new inputs and get this same quality on the first try. FORMAT (follow it exactly so it installs): - A folder named with a short lowercase-hyphen name, e.g. support-email-reply. - Inside it, a SKILL.md file that starts with YAML frontmatter: name: (the same as the folder name) description: (under 200 characters, saying what the skill does AND when to use it, since that's how you decide to load it) - If you can create files, package the folder as a ZIP with the folder at the root so I can upload it. If you can't, show me the full SKILL.md so I can copy it. SECTIONS INSIDE SKILL.md: 1. What it does: one or two sentences on the job. 2. Inputs: exactly what I have to give you each time, and what to do if one is missing. 3. Process: the steps we actually followed, in order, written as instructions to yourself. 4. My rules: every correction I made today as a permanent rule (tone, include vs cut, formatting, anything I said "always" or "never" about). This is the most important section. Do not lose a single one. 5. What good looks like: the quality bar of the final version we landed on, so you can check your own work. 6. Mistakes to avoid: what you did on the first pass that I corrected. GUARDRAILS: - Base it only on what happened in this chat, not generic best practices. If you're unsure whether something was a real rule, ask me instead of inventing it. - Write it so a version of you with zero memory of today could produce the same result. - Keep it tight and skimmable. When you're done, show me the skill and ask the one or two questions that would make it more reliable.
Install And Test It
  1. Turn on code execution. Skills need it. You’ll find it under Settings, then Capabilities. Skills work on every plan, including Free.
  2. Save or upload it. If Claude hands you a skill file in the chat, save it from there. Or go to Customize (being renamed Plugins), then Skills, click + or Add, and upload the ZIP. Don’t unzip it first.
  3. Test it in a new chat on a fresh example. If it doesn’t kick in when you expect, rewrite the description, since that’s what Claude reads to decide when to use it.

On a Mac with Pro, Max, Ultra or Team, Cowork can also record you doing the task on screen and propose a skill from it, which is Way 3 on autopilot. (Anthropic’s skill guide)

Want Way 3 done for you? My free Lock It In skill turns any chat you loved into a tested skill when you say "lock it in."