Claude Code for beginners: install, first launch, CLAUDE.md

Claude Code is a command-line tool from Anthropic: you talk to it in a terminal, it reads your project, proposes changes, and runs commands. Unlike a typical chatbot, it doesn’t just answer; it explores, plans, and acts while you watch, correct, or step away. Here’s the exact path, verified against the official documentation, to get started without getting lost.

Installing Claude Code

On Windows, open PowerShell and run:

irm https://claude.ai/install.ps1 | iex

This is the “native” install method the official documentation recommends: it updates itself in the background. A WinGet alternative also exists (winget install Anthropic.ClaudeCode), but it doesn’t auto-update. On macOS, Linux, or WSL, the equivalent command is curl -fsSL https://claude.ai/install.sh | bash.

Once the install finishes, open a new terminal window and confirm it worked:

claude --version

If your terminal doesn’t recognize the claude command, the install directory isn’t on your PATH yet; the documentation has a dedicated page to fix that for your system.

Your first launch

Move into a project folder (even a small personal one), then simply run:

claude

You’ll need to log in with a Claude account (Pro, Max, or a Console account) the first time; after that, you won’t have to again. Once you’re in the session, the documentation recommends starting with an exploratory question rather than diving into a task: try “what does this project do?”, “what technologies does this project use?”, or “explain the folder structure.” Claude reads the files it needs and answers, without you having to explain everything yourself first.

If you get stuck at some point, three options exist outside the session itself: type /help inside Claude Code, browse the official documentation, or take the free “Claude Code 101” course on Claude Academy. A Discord community also exists if you’d rather ask other users directly.

Don’t feel like you need a “real” software project to try this on. A small personal script, a class exercise, or a folder you’re just starting to build all work fine for your first session; the point at this stage is to get comfortable with the back-and-forth, not to produce something impressive on day one.

A few essential commands worth knowing from day one:

Command	What it does
claude	starts an interactive session
claude "task"	starts a session with an initial request already typed
claude -c	resumes the most recent conversation in this folder
/clear	clears the current conversation history
/help	shows available commands
/exit	exits Claude Code

The CLAUDE.md file: your project’s memory

Every session starts with a blank memory. The CLAUDE.md file exists to give Claude instructions it reads again at the start of every new session: coding conventions, test commands, things to never do. The official documentation sums it up well: it’s the place to write down “what you’d otherwise re-explain every time.”

You can place it in two spots depending on the scope you want:

• ./CLAUDE.md, at the project root: shared with anyone who picks up the project (useful if you’re working with others, or want to keep it with the project through Git).
• ~/.claude/CLAUDE.md, in your personal folder: applies to every project on your machine.

You don’t have to write it by hand from scratch: the /init command, run from inside a session, analyzes your project and generates a starting CLAUDE.md with the commands and conventions it detects. Keep it short; the documentation suggests under 200 lines, since an overly long file means Claude ends up ignoring part of what’s in it. A minimal example:

# Code style
- Use ES modules (import/export), not CommonJS (require)

# Workflow
- Always run the type checker after a series of changes
- Run individual tests rather than the whole suite, for speed

To confirm your CLAUDE.md actually got picked up, type /context at the start of a session: Claude Code lists the memory files it actually loaded. That’s the habit to build if it ever feels like Claude is “forgetting” a rule you’re sure you wrote down.

Permissions: who decides what

Claude Code doesn’t just do whatever it wants without checking in: a “permission mode” defines what it can run without asking first. The documentation lists several, from most cautious to most autonomous:

• Manual (the default mode): Claude reads files, but asks for your approval before changing anything or running a command.
• Accept edits: file edits get applied directly, handy when you’re going to review the result with git diff anyway.
• Plan: Claude explores and proposes a detailed plan, without changing anything until you approve it.
• Auto: a separate classifier model reviews each action on your behalf and blocks anything that looks risky, cutting down on interruptions during long tasks.

You can switch modes at any time with Shift+Tab. As a beginner, stay in Manual mode, or use Plan mode for anything touching several files at once: it forces you to read what Claude is proposing before it becomes irreversible.

Good habits for getting started

• Move in small steps. The official documentation recommends clearly separating exploration, planning, and execution, rather than asking for everything at once.
• Read every diff. Don’t approve a change you don’t understand; ask for an explanation if you need one.
• Use Git. Claude Code can commit, create branches, and write a descriptive commit message; ask it in plain language rather than leaving the session to type the commands yourself.
• Correct early. If Claude heads in the wrong direction, press Esc to stop it immediately; the context is preserved, so you can redirect without losing everything.
• Don’t get stuck on the first attempt. Per the documentation, it’s often better to clear the context (/clear) and start over with a more precise request than to keep correcting the same failed approach forever.

If you’re using Claude Code for graded work, check your institution’s own AI usage rules first: how you’re expected to disclose this kind of help varies from one school to another.

Key takeaways

• On Windows, the recommended install is one PowerShell command: irm https://claude.ai/install.ps1 | iex.
• A session starts with claude from your project’s folder; begin with an exploratory question rather than a task.
• The CLAUDE.md file (generated with /init, then refined by hand) gives Claude a persistent memory of your project, as long as it stays short and precise.
• Permission modes (Manual, Accept edits, Plan, Auto) determine what Claude can do without asking; Shift+Tab cycles through them.
• Move in small steps, read every diff, and use Plan mode as soon as a task touches more than one file.

For a starting CLAUDE.md already structured for a student project, the claude-md-projet skill gives you a template to adapt.