Claude Code: where to start
5 min czytaniaClaude Code is Claude working inside your project: it reads files, edits code, runs commands and commits. Not a hint in your editor — someone you brief in words.
How it differs from a chat
- In a chat you copy code back and forth. Here the agent opens the files itself, finds the right spot and edits it.
- It sees the whole project, not the fragment you pasted. So it knows where else a change will land.
- It can check itself: run the tests, build the project, open a page in a browser and look at the result.
Installing
- You need Node.js 18 or newer. One command: npm install -g @anthropic-ai/claude-code
- Then go to your project folder and run claude. On first launch it asks you to sign in.
- There's also a no-install route: claude.ai/code runs in the browser and works with your GitHub repositories.
The first thing worth doing
- Ask it to explain the project: “work out what this project is and tell me briefly”. You'll see how well it understands your codebase and get a description that's useful to you too.
- Create a CLAUDE.md — a file of project rules the agent reads at the start of every session. There's a separate guide on it.
- Give it a small real task, not an invented one. Fixing a known bug shows what it can do far better than “write a sorting function”.
How to brief it
- Talk about the outcome, not the steps. “Make the button scroll to the section below” beats “open page.tsx, find line 114”.
- Ask it to verify: “run the tests”, “show me a screenshot on mobile”. It can do this itself, but it won't always guess that you care.
- Don't hesitate to undo. If you don't like it, say “revert” — that's cheaper than editing someone else's edit.
What not to expect
- It doesn't read your mind about the product. What to build stays your call.
- It makes mistakes, like anyone. So ask it to verify the result rather than trusting the report.
- It doesn't replace knowing your project. The better you explain the context, the less you'll redo.