Getting Started with Claude Code

The First Command: /init

Bash

cd your-project
claude
> /init

/init inspects your project — reads your package.json, tsconfig, directory structure, framework conventions — and generates a CLAUDE.md file. This is the AI's onboarding document for your codebase.

What a generated CLAUDE.md looks like:

MARKDOWN

# Project Overview
A Next.js 15 e-commerce application with TypeScript.

## Tech Stack
- Next.js 15 (App Router)
- TypeScript (strict mode)
- Tailwind CSS
- Prisma + PostgreSQL
- Vitest for testing

## Project Structure
src/
  app/        → Next.js App Router pages and layouts
  components/ → Reusable UI components
  lib/        → Utilities, database client, helpers
  types/      → Shared TypeScript type definitions

## Common Commands
npm run dev      → Development server
npm run test     → Run test suite
npm run build    → Production build
npm run lint     → ESLint check
npx prisma studio → Database UI

## Notes
- Uses "@/" import alias for all imports
- Components follow naming convention...

This generated file is a starting point, not a final product. Edit it. The AI can infer your stack and structure, but it can't know:

Which patterns are intentional vs. historical accidents
What decisions were made deliberately and should not be changed
What the team tried and decided was wrong
Your security constraints
Domain-specific rules

Spend 10 minutes on it after generation. Time invested in CLAUDE.md compounds — every future session benefits.

Slash Commands Reference

Plain text

/help          → list all available commands
/status        → context usage, current model, token count
/model         → switch models mid-session
/init          → (re)generate CLAUDE.md for this project
/compact       → compress conversation history to save context
/exit          → quit Claude Code

/model — Switching Mid-Session

Plain text

> /model haiku    → fast and cheap for simple tasks
> /model sonnet   → the default daily driver
> /model opus     → maximum reasoning for hard problems

A pattern that works:

Plain text

Use Sonnet to implement a feature
Hit a tricky bug that Sonnet misses multiple times
> /model opus
"Think hard: why doesn't this work?"
Opus figures it out
> /model sonnet   ← switch back for the next task

Don't use Opus for everything. Use it surgically.

/status — Your Context Gauge

Plain text

> /status
Model: claude-sonnet-4-6
Context: 24,108 / 200,000 tokens (12.1%)
Session: 47 minutes

Before starting a large task, check /status. If you're at 60% or higher, compact first or start a new session. You want enough context headroom to complete the work without running out mid-task.

Asking Questions vs. Making Changes

Claude Code does both, and the distinction matters.

Plain text

> what does the auth module do?              ← read-only, safe to start with
> how does this project handle sessions?     ← read-only
> refactor the auth module                   ← makes changes
> implement Google OAuth                     ← makes changes

A pattern worth building:

Plain text

Ask questions first: understand what exists
Ask what would need to change for your goal
Review the AI's understanding
Then ask for the implementation

Plain text

> @src/auth/ explain how session management works here

> given that structure, what would I need to change to add
  Google OAuth support?

> [after reviewing the plan] implement the Google OAuth flow
  starting with the callback handler

The AI's answer to the second question is your plan. The third message executes it. This prevents the most common failure mode: the AI implementing a solution that conflicts with patterns you didn't know existed.

Referencing Files

Plain text

> @src/components/UserCard.tsx fix the TypeScript error
> @src/auth/ how does this module work?
> @package.json @tsconfig.json what TypeScript version and features are active?
> @src/ give me a high-level tour of the codebase structure

Use @ to attach files and directories. Be specific about what's relevant to the task. Attaching @src/ for a task that only touches one file wastes context on everything else.

The First Session Checklist

A productive pattern for starting with Claude Code on any project:

Plain text

1. Run /init → generates CLAUDE.md

2. Edit CLAUDE.md (10 minutes):
   - Fix anything incorrect
   - Add conventions the AI couldn't infer
   - Add "do not" rules based on past mistakes
   - Add your testing approach

3. Explore before touching:
   > @src/ give me a high-level tour
   > what's the most important thing I should understand
     about this codebase before making changes?

4. Start small:
   Not: "build the authentication system"
   Yes: "add email format validation to the LoginForm component"

5. Review the output carefully
   Read every changed line

6. Commit if it's good
   git add -p → git commit

What CLAUDE.md Actually Does

Every Claude Code session in your project opens by reading CLAUDE.md. It's the first thing in context. Everything in it shapes how the AI interprets your subsequent messages.

Without CLAUDE.md: the AI guesses your conventions from what it can see, defaulting to patterns from its training data.

With a good CLAUDE.md: it knows your stack, your conventions, what to avoid, and what you've already decided. You don't re-explain it session after session.

The compounding effect: a better CLAUDE.md makes every session better. Update it whenever:

The AI does something wrong that a clear rule would have prevented
You add a new library or change a pattern
You make a decision that shouldn't be undone

One line in CLAUDE.md can prevent the same mistake from recurring indefinitely.