Most Claude Code guides assume you've already used it a bit. This one starts from zero.
What you need
- Claude Code installed (from Anthropic — it runs in your terminal)
- A project directory with code in it
- An Anthropic API account with credits, or a Max subscription
That's the full list. No plugins, no IDE setup, no additional tools required.
Your first session
Open your terminal in a project directory. Run claude. You're in.
For your very first task, pick something small and self-contained. "Add a comment explaining what this function does" or "Write a test for this function." Not "refactor the whole authentication system."
Small tasks let you see how Claude works, understand the review flow, and build trust before you give it more latitude.
The one file to create first
Before you do anything else: create a CLAUDE.md file in your project root. It doesn't need to be good yet. Just add:
## Build commands
[how to run tests in your project]
[how to build your project]
## Language and stack
[what language and framework you're using]
Without CLAUDE.md, Claude guesses at your project's conventions. With it, Claude follows them. This is the single biggest quality improvement and it takes 5 minutes.
What to expect in week 1
You'll be slower than usual. This is normal. You're learning what to delegate, how to write prompts, and how much to trust the output.
You'll have some sessions where Claude produces exactly what you wanted. You'll have others where you spend more time fixing it than you would have writing it yourself.
Both are useful. The first shows you what works. The second shows you where the limits are.
The most important habit
Read every diff before committing. Always. Claude's code is usually correct. When it's not, catching it before commit saves you from debugging it in production.
That habit, applied consistently, is the difference between productive Claude Code use and frustrating Claude Code use.
Top comments (0)