Skip to content

00-setup.md

Latest commit

 

History

History
183 lines (125 loc) · 5.58 KB

00-setup.md

File metadata and controls

183 lines (125 loc) · 5.58 KB

Module 0: Setup & First Run

Duration: 40 minutes | Level: Beginner

TL;DR

Install Claude Code, authenticate, run your first prompt, and understand the basic interaction model.

0.1 Installation

Install via npm

npm install -g @anthropic-ai/claude-code

Verify installation

claude --version

Authenticate

claude auth login

You can authenticate with:

  • Anthropic API key — direct API access
  • Claude Max/Pro subscription — consumer plan usage
  • OAuth — enterprise SSO

First launch

cd your-project
claude

Claude Code launches in your terminal with an interactive prompt. It sees the working directory and can read/write files, run commands, and interact with git.

Updating

npm update -g @anthropic-ai/claude-code

Claude Code updates frequently. Check your version with claude --version and update regularly to get new features and bug fixes.

0.2 Where Claude Code Runs

Claude Code is available in multiple environments:

Environment How to access
Terminal CLI claude command in any terminal
Desktop App Mac and Windows native app
Web App claude.ai/code
VS Code Extension from marketplace
JetBrains Plugin from marketplace

All environments share the same core engine. Key differences:

Environment Strengths Limitations
Terminal CLI Most full-featured, scriptable, supports all flags Text-only interface
Desktop App Native OS integration, easier onboarding Fewer advanced flags
Web App No install needed, accessible anywhere Requires internet, no local file access without setup
VS Code Inline in your editor, sees open files Subset of CLI features
JetBrains Inline in your IDE, project-aware Subset of CLI features

The CLI is the reference environment for this course. IDE extensions are useful for quick tasks without leaving your editor.

0.3 The Interaction Model

Claude Code is not a chatbot. It is an agent that:

  1. Reads your codebase
  2. Plans changes
  3. Edits files directly
  4. Runs commands (with your permission)
  5. Iterates until the task is done

You give it tasks, not questions. Instead of "How do I add a login page?", say "Add a login page with email/password using NextAuth."

0.4 Permission Modes

Claude Code asks for permission before taking actions. You control how strict this is:

Mode Behavior
default Asks for most file writes and all shell commands
auto Allows reads, writes, and safe commands automatically
bypassPermissions No prompts (use with caution)
plan Requires plan approval before implementation

Set at launch:

claude --permission-mode plan

0.5 Essential First Commands

Once inside Claude Code, try these:

Command What it does
/help Show available commands
/status Show current session info
/model Switch the model
/clear Clear conversation context
/compact Compress conversation to save context
Ctrl+C Cancel current operation
Escape Exit Claude Code

Exercise 0.1: Hello Claude Code

  1. Open a terminal and navigate to any project directory
  2. Run claude
  3. Type: What files are in this project? Give me a brief summary.
  4. Observe how Claude reads the directory and responds

Exercise 0.2: Your First Edit

  1. In Claude Code, type: Create a file called hello.txt with a greeting
  2. Approve the file write when prompted
  3. Check that the file was created: cat hello.txt
  4. Delete the file: rm hello.txt

Exercise 0.3: Permission Modes

  1. Exit Claude Code (Escape)
  2. Restart with: claude --permission-mode plan
  3. Ask Claude to make a change — notice it creates a plan first
  4. Approve or reject the plan

0.6 Security Basics

Trusting Projects

When you open Claude Code in a cloned repository, it may contain .claude/settings.json with pre-configured permissions, hooks, and MCP servers. Always review these before trusting a project:

cat .claude/settings.json   # Check what permissions the project requests
cat .mcp.json               # Check what MCP servers it configures

A malicious project could configure hooks or settings that execute arbitrary commands. Claude Code shows a trust prompt for new projects -- read it carefully.

API Key Safety

  • Never put API keys in CLAUDE.md or settings.json (both may be committed to git)
  • Use settings.local.json (gitignored) or system environment variables for secrets
  • Use the env field in MCP configs for connection strings, not hardcoded values

Permission Discipline

  • Start with restrictive permissions and expand as needed
  • Read permission prompts before approving, especially for shell commands
  • Use --permission-mode plan when working with unfamiliar code

Common Pitfalls

  • Running Claude outside a project directory: Claude works best when launched from a project root with meaningful code. Running it from ~ gives it no context.
  • Not reading the permission prompts: Always read what Claude wants to do before approving, especially shell commands.
  • Treating it like ChatGPT: Give tasks and instructions, not open-ended questions.

Key Takeaways

  1. Claude Code is an agent that operates on your codebase, not a chatbot
  2. Always launch from your project directory
  3. Permission modes let you control how much autonomy Claude has
  4. The CLI is the most full-featured environment