THE VAULT / START HERE / GUIDE + MARKDOWN
The Workspace Setup Kit
Lovable and Bolt are great training wheels, but the browser owns your work, meters your credits, and hides your files. This kit is the 30-minute bridge to a real local setup: install Claude Code (or Cursor), open a terminal for the first time, and run your first session. Every command is copy-paste, every step tells you what you should see, and every scary word comes with a plain-English translation.
LAST VERIFIED 2026-07-09
Leaving the browser is worth 30 minutes of mild discomfort
If you've shipped something in Lovable, Bolt, or Replit, you already know how to build: describe what you want, watch it appear, iterate. This kit doesn't change that. It changes where it happens, and that changes three things that matter.
| What changes | In the browser tools | On your machine |
|---|---|---|
| Cost control | Credits that vanish fast; heavy build days can run $50+ in top-ups | Claude Code on the $20/month Pro plan covers a few focused hours of building every day, flat |
| Lock-in | Your app lives on their platform; leaving means exporting and hoping | No platform to leave. Your project works with Claude Code today, Cursor tomorrow, whatever wins next year |
| Your files | Code hidden behind a preview pane you can't fully touch | Real files in a real folder you own, back up, and open with anything |
NOTE
Use this page as a prompt reference. Every file below is built to be handed straight to an AI. Download it, drop it into Claude, ChatGPT, Cursor, or Lovable, and say "use this as my playbook." The prompts on this page are copy-paste ready.
The only obstacle is the terminal (a text window where you type commands to your computer), and the obstacle is 90% reputation. You will use about 6 commands, all copy-paste, all in this kit. Think of it as a chat box for your computer: you type, it answers in text. You already talk to computers in text all day; this is the same skill with a worse font.
The setup: 5 steps, 30 minutes, Mac or Windows
The full walkthrough with every command, every "what you should see," and every "what if I don't" lives in setup-guide.md below. Here's the shape of it:
- 1
Pick your tool (2 min)
Claude Code is the recommended default: a coding agent in your terminal, included with the $20/month Claude Pro plan ($17/month billed annually). Cursor is the alternative if you want to see your files in a sidebar: $20/month Pro with a free Hobby tier. Prices verified July 2026. Pick one; you can switch later because your files are yours now.
- 2
Open a terminal (3 min)
Mac: Cmd + Space, type Terminal, Enter. Windows: Windows key, type PowerShell, Enter. You should see a blinking cursor waiting for you. Type
echo hello, press Enter, and the computer sayshelloback. That's the whole skill. - 3
Make a projects folder (2 min)
Run
mkdir -p ~/projectsthencd ~/projects(Windows:mkdir ~/projectsthencd ~/projects). One folder where everything you build lives. The terminal says nothing when it works; silence is success. - 4
Install and verify (5 min)
Mac:
curl -fsSL https://claude.ai/install.sh | bashand Windows:irm https://claude.ai/install.ps1 | iex. Then open a fresh terminal and runclaude --version. Any version number means you're in. The setup guide covers the two errors that actually happen. - 5
Start it and log in (5 min)
Run
claudeinside your projects folder. First run opens your browser to log in with your Claude account. You land back in the terminal with an input box. Talk to it in plain English; that's the interface.
WATCH OUT
The one universal fix: when any step throws an error, copy the exact error text and paste it to an AI with "explain this in plain English and give me the single most likely fix as a copy-paste command." That one move resolves most setup problems in one round trip.
The CLAUDE.md: house rules that make the AI beginner-safe
Claude Code reads a file called CLAUDE.md from your project folder at the start of every session. It's a standing set of house rules, so you never have to repeat yourself. The beginner version in this kit is 5 rules and a diary habit, each one there because of a specific way beginners get burned:
| The rule | The burn it prevents |
|---|---|
| Explain what changed after every step | Sessions where things happen and you learn nothing |
| One change at a time | 12 changes at once; when one breaks, every change is a suspect |
| Ask before deleting anything | The one mistake that's genuinely hard to undo |
| Never touch .env without telling me (the file that holds secrets like API keys) | Leaked keys and surprise bills, the expensive kind of oops |
| Keep a PROGRESS.md diary | The AI forgets everything between sessions; the diary is its memory and yours. Next session opens with "read PROGRESS.md" and you're back where you left off |
The file is annotated: every rule has a comment explaining why it exists, so you're learning the reasoning, not just copying rules. Rename it to CLAUDE.md, drop a copy in every project folder, done.
Your first 15 minutes with Claude Code: three prompts
Setup done, CLAUDE.md in place. These three prompts, in order, take you from "it's installed" to "something I made is on the internet." Paste them exactly.
Prompt 1: the tour (minutes 0-3)
Prompt 2: build a tiny page (minutes 3-10)
That last line is the point of the exercise. Open the file, ask Claude where the background color lives, change it yourself, refresh the browser. The moment your edit shows up on screen is the moment this stops being magic and starts being yours.
Prompt 3: put it on the internet (minutes 10-15)
PAYOFF
15 minutes in, you have: a working local setup, a page you built and edited, a live URL you can text to someone, and a PROGRESS.md diary already doing its job. That's the entire browser-to-local bridge, crossed.
What breaks most often (and the exact fix)
| Symptom | What it means | The fix |
|---|---|---|
command not found: claude after installing | Your terminal hasn't noticed the new install yet, or it went somewhere the terminal doesn't look | Close the terminal completely and open a fresh one. Still failing? Re-run the installer and paste its final output into a Claude chat: "what do I run to finish this install?" |
| Red "execution policy" error on Windows | PowerShell being cautious about downloaded scripts | Run Set-ExecutionPolicy -Scope CurrentUser RemoteSigned, answer Y, retry the install command |
| Claude Code asks permission for everything | Working as designed; it asks before touching files and running commands | Say yes to reads and file creation. Keep the asks for deletes; your CLAUDE.md demands that on purpose |
| Login loop: browser says success, terminal disagrees | The handshake between browser and terminal dropped | Ctrl + C to quit, run claude again. Second attempt almost always lands |
| The AI forgot everything from yesterday | Normal. Sessions don't share memory (each has a limited context window, the amount it can keep in mind) | Start every session with: "Read PROGRESS.md and tell me where we left off." This is why the diary rule exists |
When you outgrow this kit
Two signs you're ready for the next rung. First: your CLAUDE.md starts feeling thin because you now have real commands, real rules, and files that must not be touched. Graduate to the CLAUDE.md Starter, the pro version of the file you just installed. Second: you've built something you'd be sad to lose. That's your cue for GitHub for Normal People, which gives every project a memory and an undo button.
NOTE
The move: run the 30-minute setup today, then spend your first 15 minutes on the three prompts above. Everything after that is just building.
Get the next drop
Every new Vault system ships to the list first. Twice a week, free.