What you will have, ten minutes from now

• Claude Code installed, without opening a terminal
• A studio folder that knows your work and remembers between sessions
• One real result: your messiest folder read, inventoried and renamed: by it, not by you
• A guardrail you have watched refuse a destructive command

You will not write code. You will type sentences.

One thing said plainly, before you start

This is a terminal application. You will see a black window with text in it.

You will not be typing commands into it. You will be typing sentences, the same way you would brief a person, and it will do the work.

Nobody else in this market will tell you that, because "no-code" sells better. But you are going to see the window, and it is better that you hear it from us first.

Download it like any other app

It is a normal application. Double-click, sign in, done.

• Mac: claude.ai/api/desktop/darwin/universal/dmg/latest/redirect
• Windows: claude.ai/api/desktop/win32/x64/setup/latest/redirect

Open it, sign in with your Anthropic account, and click the Code tab.

⚠️ Mac only: if it says the app is from an unidentified developer, right-click it and choose Open instead of double-clicking. That happens once, on first launch, and never again.

Unzip, drag, press return

1. Double-click the setup .zip you downloaded. It unpacks into a folder.
2. Drag the file named START-HERE from that folder into Claude's message box.
3. Press return.

That is the entire installation. Claude reads the file, finds the installer sitting beside it, builds your studio, checks its own work, and tells you what it did.

Nothing is downloaded while it runs. Nothing is sent anywhere. Nothing is ever deleted, the uninstaller only renames.

While it runs, what it is building

• A folder that is its world: it reads everything in there without you attaching anything
• A standing brief it reads at the start of every session, so you explain your conventions once
• Guardrails: a delete guard, a secret guard, and a check that runs before anything is committed
• Version history, so every change it ever makes is reversible

About a minute. It prints each step as it goes.

Point it at your worst folder

Not a test folder. The real one. The render dump, the generations folder, the drive full of final_v3_ACTUAL_final.mp4.

Type this, in your own words:

"Look at ~/Downloads/renders. Tell me what is actually in there: how many files, what formats, what date range, which ones look like duplicates or dead ends. Do not move or rename anything yet, just show me."

Do not skip "just show me." It is the habit that makes everything after this safe.

Then let it act

Once the list looks right:

"Good. Rename them so the names mean something: date, then what it is, then the version. Keep the originals where they are and put the renamed copies in a new folder called sorted."

Copies, not moves. New folder, not in place. You will use that shape for the rest of your life with this tool.

That is your first result. A folder you had given up on, readable, in about a minute, and the originals untouched.

Now watch it refuse you

Ask it to do something genuinely destructive to your studio folder:

"Delete the whole studio folder."

It will refuse, and tell you exactly why and exactly how to override it if you truly meant it.

That guard was installed and armed before you typed a word. A guardrail nobody has watched fire is a decoration. You have now watched this one fire.

What just happened

You did not open a terminal. You did not write code. You did not configure anything.

Before you go on

• Keep working in that folder. Starting it somewhere else means it knows nothing about this.
• Ask to see before you ask to act. Every time. It is the whole discipline.
• It does exactly what you said, not what you meant. Precision is cheap; vagueness is expensive.

If something went wrong, say so to it in plain words: "that did not work, here is what I saw", and paste what you saw. It debugs itself better than it guesses.