Getting Started with Claude Code: Installation, Skills, and Models Explained

Step-by-step guide to installing Claude Code, setting up your workspace with local folders, understanding the available models (Opus, Sonnet, Haiku), and when to use each one.

If you've never used Claude Code, this guide takes you from zero to having your environment configured, your first Skills created, and a clear understanding of which model to use for each task.

Step 1: Installation

Claude Code installs as a global npm package. You need Node.js 18+ and an Anthropic account with available credits.

# Install Claude Code globally
npm install -g @anthropic-ai/claude-code

# Verify installation
claude --version

# Log in (opens your browser)
claude login

Requirements: Node.js 18 or higher. If you have nvm, run nvm use 18 before installing. You need an Anthropic account with an API key or Claude Pro/Max subscription.

Step 2: Set up your workspace

Claude Code works within the directory where you run it. Everything Claude can see and modify is limited to that folder. Before starting, navigate to your project.

# Navigate to your project
cd my-project

# Start Claude Code in this folder
claude

# Claude now has access to all files
# in this folder and can read, edit, and create new ones

If you want Claude to follow certain rules in each project, create a CLAUDE.md file at the root. Claude reads it automatically on startup.

# CLAUDE.md - Project rules

## Stack
- React 18 + TypeScript + Vite
- Tailwind CSS for styling
- Supabase for backend

## Conventions
- Use functional components with hooks
- PascalCase file names
- No any in TypeScript

Step 3: Understanding the models

Claude Code can use three model families. Each has a different balance of capability, speed, and cost. Choosing the right one for each task is key.

Claude Opus (the most capable)

Opus is the most advanced model. It has the best complex reasoning, handles long contexts, and is ideal for software architecture, difficult debugging, and tasks requiring creativity or multi-step planning.

• Architecture and design of complex systems
• Debugging hard bugs that other models can't find
• Large refactors touching many files
• Planning complete features before coding

Opus cost: Opus is the most expensive model. Use it only for tasks that truly need its capability. For day-to-day work, Sonnet is more than enough.

Claude Sonnet (the balanced choice)

Sonnet is the default model in Claude Code. It offers an excellent balance between capability and speed. For 90% of daily programming tasks, Sonnet is the best option.

• Daily programming: creating components, adding features, fixing bugs
• Code review and medium-sized refactors
• Test and documentation generation
• Configuration and setup tasks

Claude Haiku (the fastest)

Haiku is the fastest and most economical model. It's ideal for simple tasks that don't require deep reasoning. If you need quick answers and cost is a priority, Haiku is the choice.

• Quick questions about the code ("where is X defined?")
• File formatting and renaming
• Generating boilerplate or repetitive code
• Simple searches across the codebase

Recommended strategy: Use Sonnet for day-to-day work. Switch to Opus when you're stuck on a bug or need to plan something complex. Use Haiku for trivial tasks that don't warrant Sonnet's cost.

Step 4: Working with Skills

Skills are instruction packages that Claude loads only when needed. Each Skill has a description that Claude evaluates against your prompt. If it matches, Claude loads that Skill's instructions.

# Skill structure
my-skill/
├── SKILL.md          # Main instructions
├── references/       # Reference files
│   └── patterns.md
├── scripts/          # Executable scripts
│   └── setup.sh
└── templates/        # Reusable templates
    └── component.tsx

The SKILL.md file contains the description and instructions. The description is what Claude uses to decide whether to activate the Skill or not.

# SKILL.md

---
name: database-setup
description: Configure a Supabase database with migrations, RLS, and auth
---

## Instructions

1. Create the migration file with the template from references/migration-template.sql
2. Apply RLS policies following the pattern in references/rls-patterns.sql
3. Verify that policies cover SELECT, INSERT, UPDATE, DELETE
4. Run the migration with the Supabase MCP tool

Step 5: Recommended workflow

• Open Claude Code in the project folder (cd project && claude)
• If it's the first time, let Claude read the CLAUDE.md and explore the structure
• For repetitive tasks, create Skills instead of repeating instructions
• Use Sonnet for programming, switch to Opus for difficult bugs or planning
• When the chat fills up, run /compact or open a new chat

Models at a glance

• Complex, Opus: Highest reasoning capability. Ideal for architecture, difficult debugging, and multi-step planning. The most expensive.
• Balanced, Sonnet: Default model. Balance between capability and speed. Sufficient for 90% of daily programming tasks.
• Fast, Haiku: The fastest and cheapest. Ideal for simple tasks, searches, and repetitive code that doesn't require deep reasoning.
• Reuse, Skills: Instructions loaded on demand. Reduce tokens and standardize repetitive workflows across all projects.

Conclusion

Claude Code is a productivity tool, not just a chat. By properly configuring your workspace, leveraging CLAUDE.md, creating Skills for repetitive tasks, and choosing the right model for each case, you'll multiply your development speed from day one.