Your First Game: The Pixel Lab · Lesson 5 ⏱ ~30 min

Make it yours, give it a memory, ship it (or don't).

Three closing moves: put your fingerprints on the lab, teach the project to remember itself between work sessions, and make your first grown-up decision about what belongs on the public internet.

Step 1 — fingerprints (three quick rounds)

Freeform time. Some crowd favorites, one prompt each:

  • The character: "Give the character my hair — short and dark — and a green hoodie." (Describe yourself. It's your lab.)
  • The sign: "Put a wall sign above the desks that says AUSTIN'S LAB in chunky pixel letters." (Your name, presumably. Austin's Lab is taken.)
  • The clutter that makes it a place: "Add a coffee mug on one desk, a crumpled paper ball on the floor, and one more plant. Plants are important."

Two or three rounds, no more — the finish line is close and this step has no bottom. Commit when you like it.

Step 2 — the memory: your first CLAUDE.md

Here's a wall you'll hit next week without this step. You'll open the folder, start Claude Code, and ask for a change — and it won't remember any of your decisions. Not the no-frameworks rule, not how lab-data works, not what /refresh-lab is for. Sessions don't share memory. You'd have to re-explain your own project to your own assistant.

The professional fix is almost embarrassingly simple: a file in the folder named CLAUDE.md, which Claude Code automatically reads at the start of every session. Whatever it says is just known, every time, before you type a word:

The prompt

Write a CLAUDE.md for this project. Cover: what this is (a pixel-art lab game where each computer shows one of my real projects), how the pieces fit (the game code reads lab-data and draws one computer per entry — content changes belong in the data file, not the code), the /refresh-lab command and what it does, and the rules I want kept: single HTML file, plain JavaScript, no frameworks, chunky pixel look, and never invent fake projects in my data.

Read what it writes — it's your project's constitution, and you're allowed to edit it in plain English, forever. Then feel free to test it: type exit, run claude fresh, and ask "what is this project and what are its rules?" It knows. That eerie little moment is the difference between owning a folder of code and owning a project.

🌱This is not a game trick either. Every serious project I run has a CLAUDE.md — including the website this course is published on: mine records how the site deploys and which folder is the real one, so any session on any machine knows the rules before it touches anything. From now on, every project you start should get one the day it stops being a toy.

Step 3 — the shipping decision

Course 01 ended with "put it on the internet," and you could absolutely do that here — same GitHub + Vercel moves, and Claude Code walks you through them on request. But this project deserves ten seconds of adult thought first:

🔒Your lab-data file is a map of your computer. Real folder names, real paths (which usually include your username), real descriptions of your work. None of it is catastrophic — but "who exactly needs to see this?" is the exact question professionals ask before every deploy, and answering it before publishing rather than after is what separates the careful from the apologetic.

Three legitimate answers, all correct:

  • Keep it local. Completely valid endpoint. It's a personal dashboard of your work that greets you when you open it — plenty of the best software in the world is an audience of one.
  • Ship a sanitized copy. Tell Claude: "Make a copy of lab-data for the public version — keep the names and summaries, strip the folder paths." Then deploy exactly as in Course 01, lesson 7. You get the show-off link; the internet doesn't get your file system.
  • Ship it all as-is. Also fine if you've read the file and you're comfortable — it's your call to make. The point was never "don't ship"; it's "decide on purpose."

Whichever you pick: show a human being. Locally over a shoulder or by link, the reaction you're farming is the same — "wait, those are your actual projects?" Yes. Yes they are.

What you actually learned (it wasn't a game)

Strip away the pixels and look at the skills, because each one maps straight onto real work:

  • State, input, the loop → the anatomy of every interactive thing you'll ever build or buy — booking widgets, calculators, dashboards.
  • Content in a data file → you can now demand this in everything ("keep the content where I can edit it") and update your own projects forever, free.
  • Claude reads reality and writes data → point the same trick at invoices, notes, inboxes, any folder of mess that should be a tidy summary.
  • A custom slash command → you extended your tool. Deploy checklists, report generators, weekly summaries — all this same mechanism.
  • CLAUDE.md → your projects now survive gaps, machine changes, and your own forgetting.

Course 01 taught you the loop: describe, build, look, fix, ship. This course taught you the other half of the craft: structure — data where it belongs, automation with names, memory that persists. Loop plus structure is honestly most of what I get paid for. The rest is reps.

Where to take it (steal any of these)

  • The lab comes alive (this one's written). Course 03 moves an AI-powered NPC into this exact lab — live Claude API calls, an assistant that knows your real projects, and a door that generates infinite rooms. Your lab-data file is about to become somebody's memory.
  • A goals room. Door on the right wall, second room, each plant is a habit you're growing — water them by checking in. The data file pattern, pointed at your life.
  • The quiz cabinet. An arcade machine in the corner that quizzes you from a data file Claude generates out of your real notes or business FAQ. Reality → data → screen, again.
  • Live links. Projects that are deployed get a "visit site" button on their panel — your lab becomes a portfolio you walk around.
  • The updating lab. Run /refresh-lab after every project you ever finish, and quietly accumulate the most motivating dashboard you'll ever own: a room that keeps needing new desks.

If you're building for the love of it: pick one of those while the lab is still warm — extending a project you understand teaches more than starting fresh. If you run a business and just did the math on your own hours: you're now the client every consultant fears in the best way — one who knows what a data file is and can't be dazzled. When you want this built for you, start with the free audit or tell me about your week.

Two courses, zero dollars, no gatekeeping. That was the deal. Go build something weirder.

You can now…

  • Write a CLAUDE.md so a project remembers itself across sessions
  • Make a deliberate call on what data goes public — and sanitize before shipping
  • Map every skill in this course onto non-game, real-world work
  • Extend the lab — or any project — on your own steam

Running a business and would rather this get built for you? That's literally my job — start with the free audit.