Build a Real Skill — Companion Guide

Before we start

Prerequisites

Tap each one as you confirm it. If something is missing, flag it in chat and we will get you set up.

✓Claude Code is running. The claude command works in your terminal.
✓A code editor is open on a folder (VS Code or Cursor) so you can see the files.
✓You have one recurring task in mind you can describe step by step (market report, listing package, weekly email, call follow-up).
✓You have a ~/.claude/.env file for secrets. API keys live here, never in the skill itself.

The one rule

Map the workflow first

You do not start by writing the skill. You map every step you take, the order, and where a human decision has to happen. Then you hand that map to your AI and it builds the skill around it.

"If you can map it, you can build it. If you cannot map it, you do not understand your own process well enough yet, and that is the real work."

Live build · 30 minutes

The six steps

Copy each prompt into Claude Code when we reach that step. Do not rush ahead, we move together.

Pick the task and map it

6 min

Pick one recurring, multi-step task. Before touching Claude, map it on paper or in a list. For each step write what happens, in what order, and where you have to decide or approve something.

Paste into Claude Code, then answer its questions

I want to build a skill for [your task]. Before we write anything, help me map the workflow. Ask me questions until we have every step, the order, and every point where I need to approve something before it continues.

✓ Done when you have a numbered list of steps with at least one "I approve here" marked.

Scaffold the skill

5 min

Turn the map into a real skill folder with a SKILL.md. The description is how the AI decides when to use it, so make it clear.

Paste into Claude Code

Turn this map into a Claude Code skill. Create the folder in ~/.claude/skills/, write the SKILL.md with a name, a clear description, and trigger phrases, and set up the steps as plain-English instructions. Do not write code yet.

✓ Done when you have a SKILL.md with a name, a description, and a trigger phrase. Read your description out loud, is it obvious when this skill should fire?

Break it into steps and phases

6 min

Lay the task out as numbered steps. If it is big, split it into two phases with their own triggers, one that prepares and one that sends or publishes, just like the newsletter skill.

Paste into Claude Code

Lay out the steps in order. If part of this should wait for my go-ahead, split it into two phases: one that prepares and one that sends or publishes.

Add a human checkpoint

4 min

This is the most important add of the day. Put one checkpoint where the skill stops, shows its plan, and waits for your yes before anything you cannot undo.

Paste into Claude Code

Before the step that [sends / publishes / spends money / is hard to undo], add a checkpoint: print what you are about to do and wait for me to confirm. Never do that step without a yes.

Nobody moves on until they can point to the line in their skill where it stops and asks.

Add a references file

4 min

If your task has a template, a format, or a set of rules, move it into a references/ file so the main skill stays short and readable.

Paste into Claude Code

I have a template/format/rules for this. Create a references/ file for it and have the SKILL.md point to it instead of holding all of it inline.

Add memory and test it

5 min

Give the skill a memory file, then run it for real. When it gets something wrong, correct it once and have it write the fix into LEARNED.md. That is the loop that makes it better every week.

Paste into Claude Code

Create a LEARNED.md in this skill's folder and tell the SKILL.md to read it at the start of every run. Then let's do a test run of the skill on a real example right now.

✓ Done when you have run your skill at least once and started a LEARNED.md.

Reference

Dependencies, mapped

The newsletter skill only works because these exist. Your skill will have its own version of each. Every service needs an API key, and every key goes in ~/.claude/.env.

The newsletter skill uses	Your equivalent
Notion (curated content queue)	Where your data lives: Notion, a Google Sheet, a folder
Gmail (source feed)	Any inbox or feed you pull from
Web search	Built into Claude already
Acumbamail (sends the email)	Your email tool with an API: Mailchimp, MailerLite
Slack / Google Tasks	Optional: a "done" notification and a reminder

After today

Level up next, and homework

Scripts

For exact, repeatable steps like sending or formatting.

API connections

Wire it to your real tools. Keys in .env.

Calling other skills

One skill triggers another as your workflow grows.

A quality check

The skill scores its own output and fixes low scores.

✓Finish your skill so it runs your real task end to end.
✓Correct once: after the next real run, add one fix to LEARNED.md.
✓Share in Skool: what task you turned into a skill, and how much time it saved.