Claude Code from Zero

An interactive course for AI Grads at Housecall Pro

Learn Claude Code by using Claude Code. No slides required — this course teaches itself.

---

How This Course Works

1. Open this project folder in Claude Code: cd ai-grad-course && claude
2. Claude becomes your teacher (it reads the CLAUDE.md in this folder)
3. Work through exercises in sandbox/exercises/ — ask Claude for help when stuck
4. Study rolands-setup/ to see what a real production config looks like
5. Build your own setup as you go

Important: This course uses Claude Code — the terminal/CLI tool. It is not the Claude desktop app or Claude.ai in your browser. Claude Code runs in your terminal and can read, write, and execute code directly in your project.

The meta insight: You're learning Claude Code inside Claude Code. The tool you're learning is also your teacher.

---

Module 1: Getting Started

1.1 What is Claude Code?

Claude Code is a CLI (command-line interface) tool that puts an AI assistant directly in your terminal. Unlike ChatGPT or Claude.ai in a browser, Claude Code:

• Lives in your terminal — where your code already is
• Reads and writes files — it can see your entire project
• Runs commands — it executes code, runs tests, manages git
• Learns your preferences — via config files you control

Think of it as pair programming with an AI that never gets tired and reads documentation instantly.

1.2 Installation

# Install Claude Code (one command, no prerequisites)
curl -fsSL https://claude.ai/install.sh | bash

# Verify installation
claude --version

# Start your first session
claude

1.3 Terminal Basics (If You Need Them)

Command	What it does	Example
pwd	Print working directory (where am I?)	pwd → /Users/you/projects
ls	List files in current directory	ls → shows files and folders
cd	Change directory	cd ai-grad-course
mkdir	Make a new directory	mkdir my-project
cat	Display file contents	cat README.md

1.4 Your First Session

cd ai-grad-course
claude

Once inside Claude Code, try:

• "What files are in this project?"
• "Read the file sandbox/exercises/01-hello-claude.js"
• "What exercise should I start with?"

1.5 Key Concepts

• Session: A conversation with Claude Code (start with claude, end with /exit)
• Context: What Claude knows about your project (it reads files you reference)
• Tools: Actions Claude can take (read files, write files, run commands, search)
• Permissions: You control what Claude is allowed to do

Exercise: Complete sandbox/exercises/01-hello-claude.js

---

Module 2: Teaching Claude Who You Are

2.1 The Config Hierarchy

Claude Code reads configuration in layers:

~/.claude/CLAUDE.md          ← Global (all projects)
~/.claude/rules/*.md         ← Global rules (coding style, testing, etc.)
~/.claude/settings.json      ← Global settings (permissions, hooks, plugins)
./CLAUDE.md                  ← Project-level (this project only)
./.claude/settings.json      ← Project-level settings

Key insight: Global config = your preferences. Project config = project-specific instructions.

2.2 CLAUDE.md — Your AI's Instruction Manual

This is the most important file. It tells Claude:

• How you like to work (workflow preferences)
• What tools are available (databases, APIs, etc.)
• What rules to follow (coding style, testing requirements)
• What to never do (guardrails)

See rolands-setup/CLAUDE.md for a real production example.

2.3 Rules Files

Rules live in ~/.claude/rules/ and cover specific domains:

rules/
  coding-style.md    ← Immutability, file organization, error handling
  git-workflow.md    ← Commit formats, PR workflow
  testing.md         ← TDD requirements, coverage thresholds
  security.md        ← Secret management, OWASP checks
  performance.md     ← Model selection, context management
  patterns.md        ← API patterns, hooks, repository pattern
  hooks.md           ← Automation hooks (pre/post tool use)
  agents.md          ← Agent orchestration guide

See rolands-setup/rules/ for all 8 rule files from a real setup.

2.4 Settings & Permissions

settings.json controls what Claude can and can't do:

{
  "permissions": {
    "allow": ["Bash(npm test:*)"],
    "defaultMode": "plan"
  }
}

• allow: Commands Claude can run without asking
• defaultMode: "plan" means Claude plans before acting (safer for beginners)

2.5 Building Your Own Config

Start with the templates in sandbox/templates/:

1. starter-claude.md — Skeleton CLAUDE.md with placeholders
2. starter-rules.md — Example coding style rules
3. starter-agent.md — Example agent definition

Exercise: Complete sandbox/exercises/02-build-your-claudemd.js

---

Module 3: Superpowers — Skills, Agents & Plugins

3.1 What Are Agents?

Agents are specialized sub-AI-instances with focused expertise:

Agent	What it does
planner	Creates implementation plans for complex features
code-reviewer	Reviews code for quality, security, maintainability
tdd-guide	Enforces test-driven development workflow
architect	Designs system architecture
security-reviewer	Finds vulnerabilities (OWASP Top 10)
build-error-resolver	Fixes build/type errors with minimal changes

Agents live in ~/.claude/agents/ as markdown files. See rolands-setup/agents/ for real examples.

3.2 How Agents Work

An agent file has two parts:

---
name: my-agent
description: What this agent does
tools: ["Read", "Grep", "Glob"]
model: opus
---

You are an expert in [domain]...

## Your Role
- Do this
- Check that
- Never do this other thing

The YAML frontmatter configures the agent. The markdown body is its system prompt.

3.3 Skills & Plugins

• Skills: Slash commands (/commit, /review-pr) that expand into full prompts
• Plugins: Installable packages that add tools, agents, and skills
• MCP Servers: External tool connections (databases, APIs, browsers)

3.4 Real-World Agent Orchestration

From rolands-setup/rules/agents.md:

Complex feature request → planner agent → architect agent → tdd-guide → code-reviewer

Agents can run in parallel for independent tasks (e.g., security review + performance review + type checking).

3.5 Building Your Own Agent

The best agents are:

• Focused: One job, done well
• Opinionated: Clear standards, not wishy-washy
• Actionable: Produce concrete output (code, reports, plans)

See sandbox/templates/starter-agent.md for a template.

Exercise: Complete sandbox/exercises/03-skills-and-agents.js

---

Module 4: MCP & External Tools

4.1 What is MCP?

Model Context Protocol (MCP) is how Claude Code connects to external systems. Think of it as "USB ports for AI" — a standard way to plug in tools.

Claude Code ←→ MCP Server ←→ External System
                              (Database, API, Browser, etc.)

4.2 Common MCP Connections

MCP Server	Connects to	Use case
Playwright	Web browser	Automated testing, screenshots
Supabase	PostgreSQL database	Query data, manage schemas
GitHub	GitHub API	PRs, issues, code review
Slack	Slack API	Send messages, read channels

4.3 How MCP Tools Appear

When an MCP server is connected, its tools show up as available actions:

mcp__playwright__browser_navigate    → Navigate to a URL
mcp__playwright__browser_snapshot    → Get page accessibility tree
mcp__supabase__query                 → Run SQL queries

You don't need to install these yourself — they come from plugins or project config.

4.4 HCP-Specific Tools (Conceptual)

At Housecall Pro, teams use tools like:

• Snowflake — Data warehouse for analytics queries
• Vercel — Deployment platform for web apps
• GitHub — Code hosting and collaboration

Your team lead will help you set up specific tool connections. This course covers the concepts so you understand how they work.

4.5 The Permission Model

MCP tools respect the same permission system:

{
  "permissions": {
    "allow": ["mcp__playwright"]
  }
}

Always start with restrictive permissions and open up as needed.

Exercise: Complete sandbox/exercises/04-mcp-and-tools.js

---

Module 5: The Meta Game

5.1 The Coaching Loop

The most powerful use of Claude Code is the feedback loop:

1. Work with Claude Code on real tasks
2. Notice what works and what doesn't
3. Update your CLAUDE.md and rules
4. Claude Code gets better at helping you
5. Repeat

This is the skill. Everything else is just setup.

5.2 Self-Audit

Periodically ask Claude Code to audit itself:

"Read my CLAUDE.md and rules files.
 What's working well? What's missing?
 What would you add based on our recent sessions?"

Claude can suggest improvements to its own instructions.

5.3 Building Your Own Tools

As you grow, you'll want to:

1. Create custom agents for your team's specific needs
2. Write rules files that encode your hard-won lessons
3. Set up hooks that automate repetitive checks
4. Share configs with teammates so everyone levels up

5.4 Your Next Moves

Don't overthink the plan. After completing the exercises, pick 3 concrete next actions:

Timeframe	Example moves
This week	Use Claude Code to understand your team's codebase
Next week	Set up an MCP connection to a tool you use daily
This month	Build an agent for a task you do repeatedly

5.5 What Great Looks Like

After this course, you should be able to:

• Start a Claude Code session and navigate any project
• Explain what CLAUDE.md, rules files, and agents do
• Build and customize your own configuration
• Understand how MCP connects to external systems
• Teach a teammate the basics

Exercise: Complete sandbox/exercises/05-meta-game.js

---

Graduation Checklist

Must Do (Required)

• Complete exercises 01-05
• Create your own ~/.claude/CLAUDE.md
• Create at least one rules file in ~/.claude/rules/
• Successfully use Claude Code on a real work task
• Explain to a teammate what CLAUDE.md does

Should Do (Recommended)

• Create a custom agent in ~/.claude/agents/
• Set up at least one hook in settings.json
• Read through all of rolands-setup/ and understand each file
• Use Claude Code to review your own code
• Complete the self-audit in Exercise 05

Stretch Goals

• Build an agent specific to your team's workflow
• Write a rules file that encodes a lesson you learned the hard way
• Help another grad get set up with Claude Code
• Contribute a tip or pattern to the team's shared knowledge base
• Set up an MCP connection to a team tool

---

Project Structure

ai-grad-course/
  CLAUDE.md                          # Makes Claude your teacher in this folder
  README.md                          # This file — the full course guide
  setup-hcp.sh                       # Installation guide script
  sandbox/
    package.json                     # Minimal Node.js config for exercises
    exercises/
      01-hello-claude.js             # Basic interaction
      02-build-your-claudemd.js      # Configuration
      03-skills-and-agents.js        # Agent ecosystem
      04-mcp-and-tools.js            # External connections
      05-meta-game.js                # Self-improvement
    templates/
      starter-claude.md              # CLAUDE.md template
      starter-rules.md               # Rules file template
      starter-agent.md               # Agent file template
  rolands-setup/                     # Read-only reference example of a real config
    CLAUDE.md                        # Sanitized global CLAUDE.md
    settings.json                    # Sanitized settings
    rules/                           # All 8 rule files (safe as-is)
    agents/                          # 6 agent files (safe as-is)
    skills-example/
      SKILL.md                       # Generic skill template

---

FAQ

Q: I'm scared I'll break something. A: Claude Code asks permission before dangerous actions. Start with "defaultMode": "plan" in settings — Claude will plan before acting and ask for your approval.

Q: How is this different from ChatGPT? A: Claude Code runs in your terminal, can read/write your actual project files, run tests, manage git, and connect to external tools. It's not a chatbot — it's a development environment.

Q: What if Claude gives wrong code? A: Always review what Claude writes. Use the code-reviewer agent. Trust but verify.

Q: Can I use this at work immediately? A: Yes! Start with read-only tasks (understanding code, finding files) before doing write operations on work repos.

Q: Where do I get help? A: Ask Claude Code itself! Type your question in a session. For course-specific help, open this project folder first so Claude has the teaching context.