If you've used Claude in a browser, you already know how to talk to it. Claude Code is the same model with hands: it can open the files on your computer, change them, and run commands. That's the whole difference, and it's a big one.
This hour gets you from nothing installed to one real job done, without breaking anything you care about.
What you need
- A paid Claude plan or a Console account. Claude Code works with Pro, Max, Team and Enterprise subscriptions, or a Claude Console account with pre-paid API credits.
- A terminal. That's the text window where you type commands: Terminal on a Mac, PowerShell on Windows.
- A real folder. Blog drafts, a website export, a spreadsheet of orders. Something you'd actually like help with.
Never opened a terminal? Anthropic has a terminal guide for new users that walks through it. And if you'd rather skip the terminal entirely, the Claude desktop app has a Code tab that runs the same thing with buttons.
Install it
Open your terminal, paste the line for your system and press Enter.
On macOS, Linux or WSL:
curl -fsSL https://claude.ai/install.sh | bash
On Windows, in PowerShell:
irm https://claude.ai/install.ps1 | iex
The installer shows no progress bar while it downloads, so give it a minute. This "native" install updates itself in the background, which is why it's the recommended route.
If you prefer a package manager, brew install --cask claude-code works on a Mac and winget install Anthropic.ClaudeCode on Windows. Neither updates itself, so you'll run brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode now and then. There's also an npm package if you already use Node.js (version 22 or later).
Then open a new terminal window and check it worked:
claude --version
You should see a version number followed by (Claude Code). If the terminal says claude isn't found, the install folder isn't on your PATH yet (the list of places your terminal looks for programs). The troubleshooting page has the fix.
Sign in and open a folder
Claude Code works inside whichever folder you start it from. So move into your folder first, then start it:
cd ~/Documents/newsletter
claude
Replace the path with your own. The first time, Claude Code asks you to log in and opens a browser window for it. Sign in with your Claude account and you're done: it remembers you. If you ever need to switch accounts, type /login inside a session.
Next, it asks whether you trust the files in this folder. Say yes for your own folders. Start from the project folder itself, not your whole home folder: trust in your home folder isn't saved, so you'd get asked every launch, and you're handing it far more than the job needs.
Ask before you assign
Start with a question, not a task:
what's in this folder? give me a short tour.
Claude reads the files it needs and summarizes. You don't have to paste anything in. This tells you two things: what it understood, and whether you're in the right folder.
To point at a specific file, type @ followed by the file's path and Claude Code includes that file in your message.
Give it one finished job
The most common first-hour mistake is a vague request. "Clean this up" gets you a guess. A finished job says what the result is, where it goes, and what's off-limits:
Read every draft in drafts/. Create content-calendar.md with a table:
title, status (idea, draft or ready), word count, and a one-line summary.
Don't edit the drafts themselves.
That prompt has a file name, a format, and a rule. Claude can tell when it's done, and so can you.
Watch the first minute. If it heads somewhere you didn't mean, press Esc. That stops it mid-action and keeps everything done so far, so you can redirect with a sentence instead of starting over.
Read the plan before you approve
For anything bigger than one file, ask for the plan first:
Before changing anything, tell me your plan in five bullets and wait.
Read it like you'd read a contractor's quote. Which files will it touch? Will it delete anything? Did it understand the goal? If something's off, say so now; it's the cheapest moment to fix a misunderstanding.
Claude Code has a dedicated mode for exactly this, where it can research but can't edit until you approve. That's plan mode, and it gets its own article.
How permission prompts work
A permission mode decides what Claude can do without asking you first. You'll see the current one in the status bar under the prompt, and Shift+Tab cycles through them.
| Mode | What runs without asking |
|---|---|
Manual (default) | Reading files only. Edits and most commands ask first |
| Accept edits | Reading, file edits and common file commands in your folder |
| Plan | Reading and researching. No edits until you approve a plan |
| Auto | Everything, with a separate safety model checking actions first |
Recent versions start new terminal sessions in auto mode, where a second model (Anthropic calls it the classifier) reviews actions instead of you. It's built to stop the things you'd want a person to stop, like piping a script from the internet straight into your shell, or a production deploy. It's also, in the docs' own words, not a guarantee of safety.
For your first hour, press Shift+Tab until the status bar reads ⏸ manual mode on. Now Claude stops and asks before each edit or command, and you get to see how it works. A prompt looks like this:
- Yes: run it once.
- Yes, and don't ask again for a command: approve that command for this project from now on. Claude Code usually saves the rule in
.claude/settings.local.jsonat the top of the project. Approving file edits this way only lasts until the session ends. - No: refuse. Press
Tabfirst to attach a note, like "no, put it in a new file instead," and Claude adjusts.
Start narrow. Read each prompt and approve one action at a time, and think twice before you say "don't ask again" for anything that deletes, sends or publishes. You can review and remove saved rules anytime with /permissions, and when you're ready to cut down the prompts safely, read permissions without babysitting.
The commands worth knowing today
Inside a session, anything starting with / is a command. Type / on its own to see the full menu, and keep typing to filter it.
| Command | What it does |
|---|---|
/help | Shows help and the commands available to you |
/clear | Starts a fresh conversation. Use it between unrelated jobs |
/rewind | Rolls files and conversation back to an earlier point |
/usage | Shows session cost and how much of your plan you've used |
/model | Switches the model |
/resume | Picks up an earlier conversation |
/exit | Leaves Claude Code (or press Ctrl+D twice) |
From your normal terminal, claude -c continues the most recent conversation in that folder.
How to undo
Every message you send creates a checkpoint, a snapshot of the files Claude is about to change. If you don't like the result, run /rewind (or press Esc twice on an empty prompt), pick the message you want to go back to, and choose:
- Restore code and conversation: back to how things were, both files and chat.
- Restore code: undo the file changes, keep the conversation.
- Restore conversation: rewind the chat, keep the files.
Two limits are worth knowing. Checkpoints track edits Claude makes with its own file tools; a file that was only changed by a terminal command (say, moving files with mv) isn't restored. And checkpoints are short-term: by default, snapshots are cleaned up about 30 days after a session. For anything you'd hate to lose, keep real backups or use Git, the version history tool most developers use. Claude can set that up for you if you ask.
What to try next
You've installed it, given it a job, approved its actions and learned the undo button. Here's where to go from here:
- Stop repeating yourself. Run
/initand Claude writes a starterCLAUDE.md, a file it reads at the start of every session. Here's how to write a good one. - Turn a task you explain often into a skill. A skill is a written procedure Claude follows whenever the job comes up.
- Make a shortcut. Custom slash commands turn a prompt you keep retyping into
/something. - Plan the big stuff. Before any job that touches lots of files, use plan mode.
Pick one folder, one job, one hour. That's enough to know whether this changes how you work.
Sources
- Quickstart, Claude Code docs
- Advanced setup, Claude Code docs
- Terminal guide for new users, Claude Code docs
- Choose a permission mode, Claude Code docs
- Configure permissions, Claude Code docs
- Checkpointing, Claude Code docs
- Commands, Claude Code docs
- Security, Claude Code docs