Getting started with Claude Code (for creatives, not developers)

I'm not a traditional software engineer. I spent twenty years in finishing suites. Claude Code is the tool that let me build my own content console and this entire website anyway. It's an AI that works inside a folder on your computer: it reads files, writes code, runs commands and checks its own work, while you direct.

The mindset that makes it work: you are the director, it is the crew. You own taste, decisions and review. It does the typing.

Install it

Open Terminal (on a Mac: Spotlight → "Terminal") and run the installer:

curl -fsSL https://claude.ai/install.sh | bash

If you already use Node.js, npm install -g @anthropic-ai/claude-code works too. Close and reopen Terminal, then type claude and follow the login. You can use a Claude subscription or an API key from the Anthropic Console.

Give it a home

Claude Code works on whatever folder you start it in. Make one for your project and start there:

mkdir ~/Projects/my-first-tool
cd ~/Projects/my-first-tool
claude

Everything it creates lives in that folder. Nothing touches the rest of your machine unless you allow it.

Write the brief file first

Type /init. Claude creates a CLAUDE.md file, which is its standing brief for this project, read at the start of every session. Open it and add what a new crew member would need:

# Project: shot-list generator
- Who it's for: me, a VFX supervisor, prepping commercials
- Style: plain language in the UI, no jargon
- Never delete files in /footage
- Ask before installing anything paid

Think of it as the production bible. The better it is, the less you repeat yourself.

Start in plan mode

Press Shift+Tab to cycle permission modes until you see plan mode. In plan mode Claude researches and proposes but doesn't change anything. Describe what you want, then read the plan like a bid: is the scope right? Is anything missing? Is it building more than you asked for?

Good first request:

Build a simple web page where I paste a script and get a shot list back as a table: shot number, description, lens, notes. Keep it to one page. Plan it first.

When the plan is right, switch out of plan mode and tell it to go.

Let it build, then review like dailies

Claude will create files and run commands. It asks permission for anything risky, and you can approve once or always. While it works:

• Esc stops it mid-step if it's heading the wrong way.
• @filename points it at a specific file ("fix the layout in @index.html").
• !command runs a terminal command yourself (for example !open index.html).

Then look at the result yourself. Open it, click through it, break it. Your notes are the most valuable input in the whole loop.

Save your work (version control)

Ask Claude to set up git. That's version history for your project: every good state is a snapshot you can return to.

Initialise git, commit what we have with a clear message, and from now on commit after each working change.

This is the single biggest safety net. When an experiment goes sideways, you roll back instead of starting again.

Keep sessions clean

Long sessions get cluttered. Two commands help:

• /clear starts fresh, for a new task in the same project. Your CLAUDE.md and files are still there.
• /compact summarises the conversation so far and keeps going.

One task per session is a good habit. It's the same reason you'd open a new comp per shot.

Your first-hour checklist

1. Installed, logged in, started in a project folder
2. /init run, and CLAUDE.md edited with your rules
3. One small build: planned, built, reviewed, fixed
4. git set up and first commit made
5. A note in CLAUDE.md about anything it got wrong, so it doesn't repeat it

Next: how to give notes that actually land.