You've built a website, then an app — and each time, you've repeated many of the same preferences: your stack, style, and how you like Claude to work. In this lesson, you'll set up two things that make Claude remember: CLAUDE.md and a skill. From here on, Claude already knows how to build an asset to make it your way. Let's get started!

Your language tracker works. The design holds together. A friend sees it and says, "Could you build me one like that, but for sleep?" You definitely can build one, but you don't want to re-explain everything to Claude when starting from scratch: "keep it simple, plain HTML, save data locally, deploy to Netlify with a one-line ask."

There are two things that end the re-explaining trap:

You'll build one of each, and your friend's sleep tracker will be the test.

A rule is a fact. A process is a workflow.

Rules don't change between tasks; they shape every task. Processes turn into the same kind of work every time you trigger them.

CLAUDE.md is just a text file named "CLAUDE.md" at the root of your project folder — the one you choose for every distinct asset you're creating. Claude automatically reads it at the start of every session in that folder.

You don't have to write it by hand. Type /init in the chat, and Claude scans your project and drafts a starter CLAUDE.md:

From there, you add your own rules by asking Claude directly, "Add these rules to CLAUDE.md..." or by using /memory command.

Consider adding these details in your prompt to create a useful CLAUDE.md file:

The more specific the rules, the more time you save in every session that follows.

Open a fresh session in the same folder and ask Claude to start something small — anything. Notice what happens. You won't have to specify any of your usual preferences. Claude opens with your stack already chosen, your storage approach already set, and your deploy plan already in mind. If it misses something, open the file through /memory and make the instruction clearer.

CLAUDE.md is great for what's always true. But your friend asking for a sleep tracker is a recurring kind of work. Every time someone wants a small variation of what you built, you do roughly the same thing:

That's not a rule, but a process that goes into skills.

A skill is a folder Claude can load when its job matches what you're doing. The folder contains a SKILL.md file: a short description of what the skill does and when to use it, followed by the steps Claude should take when triggered. You don't write SKILL.md by hand. Anthropic ships a skill called skill-creator that interviews you about your workflow and writes the file for you.

In the chat, type /skill-creator. It walks you through a short interview: what the workflow is, when it should run, what inputs you'll give it, what output you want, and any edge cases. You answer in plain language. It writes the skill, sets up the folder, and installs it. You can also create skills from Customize → Skills → Create skill, which uses the same flow in a visual form.

Notice what skill-creator just spent the most time on: the description. If the description is vague — "helps build apps" — Claude won't reliably trigger it. But if it's specific — naming the phrases users actually say, the kind of input expected, the output produced — the skill activates when it should and stays out of the way when it shouldn't.

Your friend wants a sleep tracker. You don't need to re-specify any recurring instructions. You just describe what the app does. Three pieces of information do all the work:

You typed a functional brief and got a working, deployed, on-brand app. Three things lined up to make that possible:

Nice work! Your friend has a sleep tracker, you didn't re-explain a single rule, and Claude now opens every session in this folder already knowing how you work. Next, you'll go further: instead of Claude waiting for you to ask, you'll set up Routines you already met before — work that runs on a schedule, on its own. The CLAUDE.md and the skills you just built come with it. Let's keep going!