Track 6: Practitioner Setup

Your First Agentic Coding Session in 30 Minutes

Installation to first project in under 30 minutes.

This is the hands-on tutorial. Not an overview of what Claude Code can do — a step-by-step walkthrough of actually setting it up, configuring your project, and running your first agentic coding task. Concrete commands, real examples, what to expect at each step.

Prerequisites:Node.js 18+npm or npxOpenAI API keyTerminal access

Step-by-Step Tutorial

From zero to first agentic task

01

Install Claude Code

Claude Code is a Node.js CLI. You need Node 18+ installed. One command gets you there.

terminal
# Install globally via npm
npm install -g @anthropic-ai/claude-code

# Verify installation
claude --version
# → claude 1.x.x
NOTE

You'll need an OpenAI API key. Get one at console.anthropic.com.

02

Authenticate

Run the login command. Claude Code will open a browser window to authenticate with your Anthropic account, or you can provide an API key directly.

terminal
# Authenticate with your Anthropic account
claude login

# Or set API key directly in your environment
export ANTHROPIC_API_KEY=sk-ant-...

# Verify auth works
claude "Say hello"
NOTE

For team environments, use ANTHROPIC_API_KEY in your CI/CD secrets.

03

Create your CLAUDE.md file

CLAUDE.md is how you give Claude persistent context about your project — tech stack, conventions, commands, and anything it should always know. Create it in your project root.

terminal
# In your project root
cat > CLAUDE.md << 'EOF'
# My Project

## Tech Stack
- Next.js 15, TypeScript, Tailwind CSS
- Supabase (postgres), Vercel deployment

## Key Commands
- `npm run dev` — local dev server
- `npm run build` — production build
- `npm test` — run test suite

## Conventions
- Use TypeScript strict mode
- Components in src/components/
- API routes in src/app/api/
- Never use `any` type
EOF
NOTE

Claude reads CLAUDE.md at the start of every session. Update it as your project evolves.

04

Run your first agentic task

Now the interesting part. Describe a task in plain language. Claude will read your project, plan the changes, and execute them.

terminal
# Start Claude Code in your project directory
cd my-project
claude

# Or send a one-shot task
claude "Add input validation to all form fields
in src/components/ContactForm.tsx. Use zod.
Include error messages under each field."

# Claude will:
# 1. Read your project structure
# 2. Plan the changes
# 3. Edit the file(s)
# 4. Show you a diff to review
NOTE

For risky operations, Claude will ask for confirmation before executing.

05

Master the key commands

Claude Code has a small set of commands that unlock the full workflow — slash commands for common operations, keyboard shortcuts for navigation.

terminal
# Inside a Claude Code session:

/clear          # Clear context, start fresh
/compact        # Summarize context to save tokens
/cost           # Show token usage for this session
/review         # Ask Claude to review its own output

# Exit modes
Ctrl+C          # Cancel current operation
/exit           # End session

# One-shot from terminal
claude -p "explain this codebase structure"
claude --continue  # Resume last session
NOTE

Use /compact frequently in long sessions to avoid context overflow.

06

Run tests and iterate

Give Claude your test command. It will run the suite, read failures, fix them, and re-run until passing. This is the core loop that makes agentic coding powerful.

terminal
claude "Add unit tests for the auth utilities
in src/lib/auth.ts. Run the tests after writing
them and fix any failures before stopping."

# Claude will:
# 1. Read src/lib/auth.ts
# 2. Write test file
# 3. Run: npm test
# 4. Read failures
# 5. Fix implementation or tests
# 6. Re-run until green
NOTE

Always specify your test command in CLAUDE.md so Claude knows how to run them.

Pro Tips

What experienced users do differently

Write your CLAUDE.md before starting any session

A well-written CLAUDE.md saves dozens of minutes per session. Include your test commands, build commands, linting setup, and any conventions Claude should follow. Treat it like onboarding documentation for a new developer.

Be specific about success criteria

"Add a button" leaves too much ambiguous. "Add a primary CTA button with the text 'Get Started', orange background, border-radius 24px, linking to /sign-up, above the fold on mobile" gives Claude what it needs to succeed on the first attempt.

Use git checkpoints before large tasks

Before you give Claude a big refactor or a multi-file task, commit what you have. If Claude's changes go in an unexpected direction, you can reset cleanly. Treat Claude like a pair programmer, not an undo button.

Review diffs, not just outcomes

Claude shows you its changes as a diff before applying them. Read the diff. Not because Claude makes mistakes often, but because reviewing what changed is how you stay in the driver's seat and catch anything that wasn't what you intended.

What to build first

The best first project is something you actually want — a small tool, a script that automates something tedious at work, or a feature in an existing project. Start with something you understand well enough to review Claude's output intelligently.

Good first tasks: add validation to a form, write tests for an existing function, refactor a component that's grown too large, add a database query with pagination. Bad first tasks: “build me a startup” or anything where you can't evaluate whether the output is correct.

Go deeper with Track 6

Interactive lessons cover everything this tutorial introduces — plus slash commands, memory, hooks, and advanced workflows.