Notes6 min read

Your first Claude Code skill in 15 minutes

Make a folder at .claude/skills/<name>/ with a SKILL.md file inside: a short header with the skill's name and a description of when to use it, then the steps in plain words. Run it, write down whatever Claude had to guess, and run it again.

From the course · lesson 0.6

Making a Claude Code skill takes about fifteen minutes. A skill is a written recipe that Claude follows on request. You describe a job once, in plain words, and save it in a file called SKILL.md. After that you can run it by typing / and its name, and Claude can also pick it up on its own when your request matches.

Anthropic's docs give a simple rule for when to make one. If you keep pasting the same instructions, checklist or steps into chat, that's a skill waiting to be written. If the other terms are still fuzzy, agents, skills, MCPs and hooks explained with Figma maps each one to something you know.

Pick a weekly chore

Good first skills are small, frequent, and annoying to explain. Some examples:

  • Turning raw notes into the same summary format every week
  • Checking a page against your design rules before you share it
  • Writing release notes from a list of changes

This article uses one from Kudos, a made-up testimonial app. Every Friday, someone reads the week's customer feedback, picks the three strongest testimonials, trims them into pull quotes, and writes a short roundup for the newsletter. It's a 30-minute job with a lot of small judgment calls, which makes it a good skill.

How to make a Claude Code skill: the parts of a SKILL.md

A skill is a folder, and the folder's name becomes the command. For a project skill, the path is:

.claude/skills/testimonial-roundup/SKILL.md

The .claude folder sits at the top of your project. Put the skill in ~/.claude/skills/ instead if you want it in every project on your computer. Skills follow an open format called Agent Skills, so the same folder can work in other AI tools that support it.

Inside, SKILL.md has two parts.

1. The header (frontmatter). This is a short block between two --- lines, and it has to be the very first thing in the file. Two fields matter for now:

  • name is the command you'll type. Use lowercase words joined by hyphens. If you leave it out, the folder name is used.
  • description says what the skill does and when to use it. Claude reads this to decide whether a request matches, so include the words you'd actually say, like "weekly roundup" or "best testimonials."

2. The body. These are the instructions, written in Markdown. Write them the way you'd brief a new teammate. Say what to read, the steps, what the output should look like, and the rules that never bend.

Here's a complete skill you can copy:

---
name: testimonial-roundup
description: Picks the three strongest testimonials from this week's customer feedback in feedback/inbox/ and writes a short roundup. Use when the user asks for the weekly roundup, the best testimonials, or quotes for the newsletter.
---

# Weekly testimonial roundup

## What to read
Every file in `feedback/inbox/`. Each file is one customer message with
their name, their company, and a line that says `consent: yes` or `consent: no`.

## Steps
1. Skip any message that doesn't say `consent: yes`.
2. Pick the three most specific messages: a concrete result, a before and
   after, or a named feature. Skip vague praise like "great product."
3. For each one, write a pull quote of 25 words or fewer. Use only the
   customer's own words. You may cut words. Never add or change them.
4. Under each quote, add one line on why you picked it.

## Output
Save the roundup to `roundups/` as a file named with today's date
(for example `2026-10-03.md`), in this format:

> "Pull quote" (Name, Company)
> Why: one line

## Rules
- Never invent a quote, a name, or a result.
- If fewer than three messages qualify, say so. Don't pad the list.

The rules section is the part that helps most. It's where your judgment goes, written down so it doesn't depend on you being in the room. Directing a crew of agents shows how that habit grows.

CodifyStandby
A messy weekly chore, written down once as a SKILL.md, then run the same way every week.

Try it: 15 minutes

  1. Minutes 0–3. Do the chore once with Claude, in normal chat. Notice every time you had to explain something.
  2. Minutes 3–8. Create the folder and the SKILL.md. You can write it yourself or ask Claude to draft it from the conversation you just had. Either way, read every line and fix it.
  3. Minutes 8–10. Start Claude Code in your project and type /testimonial-roundup. You can also ask in plain words, "do the weekly roundup," and see whether Claude picks the skill up by itself.
  4. Minutes 10–15. Read the output the way you'd critique a design. Where did Claude guess? Write each answer into the skill and run it again.

Run it, fix it, reuse it

The first run shows you what you knew but never wrote down. Maybe Claude picked a quote that mentions a competitor. Maybe it counted "consent: Yes" with a capital letter as missing. Each surprise becomes one more line in the skill. After two or three runs, the skill usually matches how you'd do the job yourself.

If Claude doesn't pick the skill up when you ask in plain words, the docs suggest three fixes. Put the words people would naturally say into the description. Ask "What skills are available?" to make sure it loaded. Or run it by name with /testimonial-roundup.

Two more options, once the basics work:

  • Keep it manual-only. Add disable-model-invocation: true to the header, and the skill runs only when you type its name. The docs recommend this for anything with side effects, such as publishing or sending.
  • Add reference files. A skill folder can hold other files, such as examples.md with three roundups you loved. Link to them from SKILL.md, and Claude reads them only when it needs them, much like keeping reference frames next to your design.

The automation ladder

A skill is the second rung of a ladder. Each rung needs less of you.

RungWhat it isKudos example
PromptYou explain it every timeTyping the roundup brief each Friday
SkillWritten once, run on command/testimonial-roundup
Skill + hookA rule that runs every time, with no exceptionsA hook that blocks edits to the raw feedback files
RoutineRuns on a schedule, even when your laptop is closedThe roundup drafted every Friday at 9am

Routines run on Anthropic's cloud, and you can create one by typing /schedule in Claude Code. A desktop scheduled task does the same on your own machine. One thing to watch: the docs note that disable-model-invocation: true also stops a skill from running when a scheduled task fires with it as the prompt. Leave that line out of a skill you plan to put on a schedule.

In Chief of Agents, this is lesson 0.6. It's the first skill you write, and most of what you build later is skills like this one.

Sources

Learn this inside Claude Code.

Join the waitlist for Chief of Agents. You'll get the free beginner's handbook and skills right away, and hear first when the course opens.

One email to confirm, then the handbook and skills. Unsubscribe anytime.

Keep reading

  1. 6 min read

    Claude Code for designers: what it is, and why it isn't just for engineers

    What a coding agent does, why directing one is a design skill, and a safe first hour for designers new to Claude Code.

  2. 7 min read

    Write your taste down: a DESIGN.md your agents follow

    How to get AI to follow your design taste: write a DESIGN.md of short, checkable rules your agents read, then review screens against it.

  3. 7 min read

    Write the press release before you build

    How to know an idea is worth building: write the press release first. Amazon's PR/FAQ, a skeptic agent and the 11-star exercise.

All notes →

© 2026 Chief of Agents · Notes · PrivacyIdeate. Build. Sell. You lead, the agents work.