Claude Code

By the end of this lesson, you’ll have Claude Code installed and working — and you’ll have built your first web page by describing it in plain English.

What is Claude Code?

Claude Code is Anthropic’s command-line AI agent. It runs in your terminal, can inspect files in your project directory, and uses a large context window to reason across many files — planning, writing, editing, and debugging code autonomously.

Key features:

• Understands your codebase through a 1M-token context window — Sonnet 5 on every plan; Opus 5 and Fable 5.1 at 1M on Max, Team, and Enterprise (on Pro, Opus at 1M and Fable 5.1 at all need usage credits)
• Creates, edits, and deletes files
• Runs commands and tests
• Iterates on errors automatically
• Respects your project’s style and conventions

Installation

Step 1: Install Claude Code

Anthropic’s recommended path is the native installer — one command, no Node.js required, and it keeps itself up to date in the background.

curl -fsSL https://claude.ai/install.sh | bash

curl -fsSL https://claude.ai/install.sh | bash

irm https://claude.ai/install.ps1 | iex

curl -fsSL https://claude.ai/install.sh | bash

The installer puts a claude launcher in ~/.local/bin. Verify it installed:

claude --version

You should see a version number (something like 2.1.261 (Claude Code)). claude doctor prints a read-only health check of the install if anything looks off.

Step 2: Authenticate

Run Claude Code for the first time:

claude

It will walk you through authentication. You have two options:

Option A — Claude subscription (easiest): If you have a Claude Pro ($20/mo), Max (from $100/mo — see claude.com/pricing), Team, or Enterprise plan, Claude Code logs in through your browser and uses that plan. The free claude.ai plan does not include Claude Code.

Option B — API key (pay per token):

1. Go to the Claude Console at console.anthropic.com
2. Create an account if you don’t have one
3. Go to API Keys and create a new key
4. When Claude Code asks, paste your key

export ANTHROPIC_API_KEY="sk-ant-your-key-here"
claude

Your first interaction

Let’s do a “hello world” — the simplest possible task to make sure everything works.

Step 1: Create a project folder

mkdir hello-claude && cd hello-claude

Step 2: Start Claude Code

claude

You’ll see a welcome screen with a prompt. Claude Code is now an interactive session — you type messages, Claude responds, and you go back and forth like a conversation.

Step 3: Describe what you want

Type this into the Claude Code session:

I need a simple HTML page that says "Hello from Claude Code!" with a
dark background and centered white text. Make it look clean and modern.

Watch what happens:

1. Claude plans what to create
2. It creates an index.html file
3. It writes the HTML, CSS, and content
4. It tells you what it did

Step 4: Refine it

Without leaving the session, type a follow-up:

Add a subtle gradient to the background and a footer with today's date.

Claude updates the file. This back-and-forth conversation is how you’ll use Claude Code for real work — not one-shot commands, but an ongoing dialogue.

Step 5: Check the result

Press Ctrl+C to exit Claude Code, then inspect what it built:

ls                    # See the file(s) Claude created
cat index.html        # Look at the contents

open index.html

explorer.exe index.html

xdg-open index.html

You should see a clean webpage with your message. You just built a website by having a conversation in plain English.

The Real Workflow

Knowing the commands is one thing. Knowing how experienced practitioners actually use Claude Code is another. Here’s the session lifecycle that makes the difference.

Launch with intention

Claude Code has permission modes that decide how much it asks before acting. Press Shift+Tab to cycle through them; the current mode shows in the status bar.

• Auto — the starting mode on Pro, Max, and Team plans, in a terminal or the VS Code extension, from v2.1.228 onward (v2.1.233+ on native Windows). A built-in classifier approves routine reads and edits on its own and stops to ask you about anything risky. Good default for everyday interactive work. It is not the starting mode for claude -p, the Agent SDK, Enterprise plans, or Console API keys — those always start in Manual, regardless of version.
• Manual — Claude asks before every file edit and command. This is the starting mode for API-key/Console accounts, Enterprise plans, claude -p, and the Agent SDK, and the one to pick when you’re working on code you care about and want to see every step.
• Accept edits — file edits go through without asking; commands still ask.
• Plan — read-only. Claude reads your project and proposes a strategy without touching anything. Start here for anything non-trivial: claude --permission-mode plan, or press Shift+Tab until the bar says plan.

There is also claude --dangerously-skip-permissions for prototyping in throwaway folders. It removes every confirmation and deliberately has no short alias — typing it in full is part of the safety friction. Never use it near important code.

Describe, don’t command

Bad: “Create a file called form.html with a form element containing three input fields.”

Good: “I need a contact form for a small business website. It should collect name, email, and message, with validation. Dark theme.”

Give context and intent. Let Claude figure out the implementation details — that’s what it’s good at.

Start by asking questions

The fastest way to get value from Claude Code on an unfamiliar codebase isn’t to have it write code — it’s to have it explain the code. Before reaching for autonomous edits, ask:

• “How does authentication work in this project?”
• “Where is the checkout total calculated, and how is that function used?”
• “Why does this function take so many arguments? Check the git history.”

Claude reads your actual files (and your git history) to answer — it traces real usage rather than doing a plain text search, and nothing is indexed or uploaded to a server. This is how teams at Anthropic onboard new engineers onto a codebase, and it doubles as a low-risk way to learn what Claude can do before you trust it with changes.

The mode dance

This is the skill that separates casual users from effective ones:

1. Plan mode (Shift+Tab until plan, or start a prompt with /plan) — Claude reads your project and proposes a strategy without editing anything. Use this when you need to think through a problem.
2. Execute (Auto or Manual) — Claude builds, edits, and runs commands. This is where the work happens.
3. The rhythm: Plan → think → approve → build → checkpoint → repeat. The back-and-forth between planning and executing IS the skill.

Context is finite

After extended sessions, Claude starts losing earlier context. Two tools help:

• /compact — condenses the conversation, preserving key decisions and dropping noise. Use when responses start getting less accurate.
• /clear — fresh start for a new task. Always commit your work before clearing.

Checkpoint constantly

Between major steps, save your work with git (covered in Module 7):

git add -A && git commit -m "checkpoint: feature X working"

Git is your safety net. If Claude makes a mess in the next step, you can always roll back. Don’t worry about this command yet — you’ll learn exactly what it does in Module 7.

Investing in Your Tools

Claude Code is a professional tool, and the most effective way to use it involves a subscription. Here’s how to think about the cost:

Max 5x ($100/month) or Max 20x (a higher fixed-cost tier — see claude.com/pricing for current rates): generous Claude Code usage, Opus 5 by default, 1M-token context, and access to Fable 5.1 — the model Anthropic positions for demanding reasoning and long-horizon agentic work (released 2026-09-01; the fable alias now resolves to it, and it needs Claude Code 2.1.255 or later) — subject to current plan terms. The predictable monthly cost makes it easier to experiment and iterate. You can experiment freely, iterate without watching a meter, and build the habits that make you productive. This is what most serious users should consider once they see the value.

The ROI framing: If Claude Code saves you 5 hours per month on tasks you’d do manually, the subscription has paid for itself many times over. Anyone who bills by the hour can run that arithmetic against their own rate. Researchers who can prototype analyses in hours instead of weeks. Students investing in a skill that compounds across their career.

API key ($5-50/month typical): Pay per token. Good for light or occasional use. Use /usage to monitor spending during a session.

No free Claude Code tier: Claude Code requires a Pro, Max, Team, or Enterprise plan (or an API key). The free claude.ai chat plan does not include Claude Code access. If you want to evaluate before committing, the cheapest path is one month of Pro at $20.

Useful Claude Code commands

Once inside a Claude Code session, these are helpful:

Tips for effective use

1. Start in your project directory. Always cd into your project folder before running claude. This gives it access to your files.

2. Be specific. “Make it look nice” → vague. “Use a dark theme with #09090b background, rounded corners, and subtle border” → specific and predictable.

3. Iterate. Don’t try to get everything right in one prompt. Start with the basic structure, then refine: “Now add a navigation bar” → “Make the nav sticky” → “Add a mobile hamburger menu.”

4. Let it run. In Manual mode Claude will often plan multiple steps and ask for confirmation. When it asks “Should I proceed?”, say yes unless something looks wrong. In Auto mode it only stops for the risky steps — read those prompts carefully.

Power features

Once you are comfortable with the basics, these features make Claude Code significantly more capable.

Plan mode

Press Shift+Tab until the status bar says plan (or start any prompt with /plan). In this mode, Claude reads your project and proposes a strategy without editing any files. This is useful when you want to think through an approach before committing to changes. Press Shift+Tab again to cycle back to an executing mode and let it build. Especially valuable for beginners: use plan mode first to understand what Claude would change and why before letting it touch your files.

Extended thinking

Use /effort to control how deeply Claude thinks before responding:

/effort auto puts it back to the model’s default. Haiku 4.5 has no effort control.

Context management

Long conversations can cause the AI to forget earlier decisions. Use /compact to condense a conversation (keeping key context), /clear to start fresh, or /init to generate a CLAUDE.md project memory file. These are all listed in the command table above.

The .claude directory

Claude Code stores configuration and project memory in a .claude directory in your project root and in ~/.claude for global settings. The /init command generates a CLAUDE.md file that captures your project’s conventions — Claude reads this at the start of every session.

Inline shortcuts: @, #, and !

Three single-character prefixes speed up everyday work inside a session:

• @ — reference a file or folder by path. Type @src/ and Claude tab-completes; the file is pulled into context exactly when needed instead of you pasting it.
• # — start a line with # to tell Claude to remember something. It appends the note to your CLAUDE.md automatically, so it persists into future sessions. This is the lowest-friction way to grow your project memory: when you catch Claude doing something the wrong way, correct it once with # and it sticks.
• ! — start a line with ! to run a shell command without leaving Claude. The command runs locally and its output goes into the context window, so Claude can act on it next turn. Handy for long-running commands or pulling specific output into the conversation.

Two keys are worth muscle memory too: Esc interrupts whatever Claude is doing without killing the session — use it to course-correct mid-action instead of Ctrl+C, which exits entirely — and Ctrl+O toggles the verbose view so you can see exactly which files and commands Claude is touching.

Working from images and mocks

Claude Code is multimodal. You can hand it an image — a design mock, a screenshot, a diagram, an error dialog — and have it work from what it sees:

• Drag and drop an image file into the terminal session
• Paste an image straight from your clipboard
• Point to a path: “Implement the layout in ./designs/dashboard.png”

This is one of the fastest ways to go from idea to working UI: drop in a mock, say “build this,” then iterate. It also speeds up debugging — paste a screenshot of a broken layout and ask what’s wrong.

Autonomous mode

Running claude --dangerously-skip-permissions lets Claude execute without asking for confirmation on file edits or commands. The name is intentional — it is dangerous. Use it only for throwaway prototypes in a fresh directory, never near important code or data. (Auto mode is the safer everyday alternative: it approves the routine steps and still asks about the risky ones.)

Newer features worth knowing about

These capabilities have been added in recent releases. You don’t need them on day one, but they’re good to know about:

• Desktop app — Claude’s desktop app (macOS, Windows, Linux beta) bundles Claude Code in a “Code” tab: no terminal and no install required, same CLAUDE.md and settings.
• Voice mode (/voice) — Hold spacebar to speak commands instead of typing. Great for non-developers who think faster than they type, or for describing complex ideas conversationally.
• Auto-Memory — Claude automatically saves project notes in a MEMORY.md index (check the /memory command), so it remembers key decisions and patterns across sessions without you having to repeat yourself.
• Skills — Reusable workflows defined as markdown files in .claude/skills/. Think of them as saved recipes Claude can follow (the older .claude/commands/ files still work — they’ve been merged into skills).
• Remote control — Monitor and interact with your local Claude Code sessions from claude.ai or the mobile apps. Useful for kicking off long tasks and checking on progress from your phone.
• Claude Code on the Web — A cloud-hosted version at claude.ai/code that connects to your GitHub repos for full project editing in the browser, no local install needed.
• Import from other tools — claude import codex or claude import gemini pulls MCP servers and settings over from Codex CLI or Gemini/Antigravity, handy once you run more than one tool.

Verify it works

If you already ran claude --version during installation and saw a version number, you’re set. Now confirm Claude Code can actually reach the API:

claude

Type a simple question like “What is 2 + 2?” and you should get a response. If you do, congratulations — Claude Code is fully working. Press Ctrl+C to exit. You’ll use it for real projects starting in the next module.

Key Takeaways

• Install with curl -fsSL https://claude.ai/install.sh | bash (no Node needed) and verify with claude --version
• Authenticate with your Claude subscription (easiest) or an API key (pay per token; handy for remote/SSH)
• Always start from your project directory — Claude Code reads the files around it to understand context
• Know your mode — Auto approves the routine steps, Manual asks about everything, Plan touches nothing; Shift+Tab cycles them
• Use it as a conversation — describe what you want, refine with follow-ups, don’t try to one-shot everything
• The real workflow: plan → think → execute → checkpoint → repeat
• Invest in your tools — Max subscriptions remove per-token anxiety and let you experiment freely