Meet Claude Code - Your AI First Mate

Ahoy @name, and welcome to the final module! Captain Redbeard here. Every great pirate captain needs a trusted first mate at the helm - and yours is Claude Code, the official command-line crewmate from Anthropic. 🏴‍☠️🤖

What is Claude Code?

Claude Code is an agentic coding tool that lives in your terminal. Unlike a simple chatbot, it can read your files, search your codebase, edit code, run commands, and complete multi-step tasks - all from the command line. Think of it as a tireless first mate who knows every plank of your ship.

Claude Code vs Cursor

We sailed with Cursor through the earlier modules. Both tools are powered by AI, but they steer different ships:

Cursor - the AI-first IDE:

• A full graphical editor (GUI)
• Inline edits and

  Tab

  autocomplete as you type
• Agent / Composer for multi-file changes inside the editor

• @

  -mentions to pull files into context

Claude Code - the terminal-native first mate:

• A command-line tool (CLI) and SDK
• Runs autonomous, multi-file tasks from the terminal
• Perfect for automation, scripts, and CI pipelines
• Works alongside any editor you like

They are complementary. Cursor is your captain's wheel; Claude Code is the crew that works the whole deck. Many pirates use both.

Boarding the Ship: Installation

Requirements

• Node.js 20.x (check with:

  node --version

  )
• npm

Install Claude Code

1# Install globally with npm
2npm install -g @anthropic-ai/claude-code
3
4# Then launch it
5claude
6
7# No global install? Run it on demand with npx
8npx @anthropic-ai/claude-code

Signing the Articles: Authentication

The first time you run

claude

, you sign the pirate's articles - you log in. Run:

1# Inside the Claude Code REPL
2/login

This opens an OAuth flow. You can sign in with:

• A Claude Pro or Claude Max subscription, OR
• Anthropic Console / API credits

If you prefer to authenticate with an API key instead of OAuth, set an environment variable:

1# Linux/Mac - add to ~/.zshrc or ~/.bashrc
2export ANTHROPIC_API_KEY="sk-ant-..."

Choosing Your Crew: Models

Claude Code can steer with different Claude models, each suited to a different voyage:

• claude-opus-4-8 - the most capable; best for the hardest treasures
• claude-sonnet-4-6 - balanced and the default; great everyday crewmate
• claude-haiku-4-5 - the fastest; perfect for quick questions

Switch models from inside the REPL with the

/model

slash command, or pick one when you launch:

1# Launch with a specific model
2claude \
3  --model sonnet

1# Or switch mid-voyage, inside the REPL
2/model

Setting Sail: Running Claude Code

There are several ways to call your first mate:

1# 1) Interactive REPL - a full conversation in the terminal
2claude
3
4# 2) Interactive, but start with a prompt
5claude "Explain what TypeScript is"
6
7# 3) Non-interactive - print the answer and exit (for scripts and pipes)
8claude -p "List the npm scripts in this project"
9
10# 4) Continue your most recent session
11claude -c
12
13# 5) Resume a specific past session
14claude -r

The

-p

(

--print

) flag is the secret to automation: Claude answers and exits, so you can use it in scripts and pipelines. You can even feed it data through a pipe:

1# Pipe a file into Claude for review
2cat src/auth.ts | claude -p "review this for bugs"
3
4# Pipe a git diff for a summary
5git diff | claude -p "summarize these changes"

Pointing at the Treasure: File Context

Here is a crucial truth, matey - Claude Code has no separate file flag at all. It reads and searches your files on its own. To point it at a specific file in a prompt, use the

@

symbol:

1# The @ symbol references a file right in the prompt
2claude -p "explain @src/auth.ts"

Need Claude to see another directory? Add it with the

add-dir

flag:

1# Give Claude access to an extra directory
2claude \
3  --add-dir ../shared-lib

The Captain's Toolbelt: Key Flags

These are the real flags Claude Code understands. Memorize this short list - there are no others to invent:

1# Pick the model
2claude \
3  --model sonnet
4
5# Add a directory to context
6claude \
7  --add-dir ./packages
8
9# Control how permissions are granted
10claude \
11  --permission-mode plan
12
13# Restrict which tools Claude may use
14claude \
15  --allowedTools "Read,Edit"
16
17# Change the output format (great for scripts/CI)
18claude -p "review @src/app.ts" --output-format json
19
20# Load extra MCP servers
21claude \
22  --mcp-config ./mcp.json
23
24# Append extra instructions to the system prompt
25claude \
26  --append-system-prompt "Always write TypeScript"

Beware the kraken of made-up flags! Many tutorials invent flags that simply do not exist - a "file" flag, a "temperature" flag, a "max-tokens" flag, a "glob" flag, an "output" flag. None of these are real. If a guide shows them, it is sailing by a false map. The genuine flags are exactly the ones listed above.

Slash Commands in the REPL

Once you are inside the interactive REPL, slash commands steer the session. Here are the real ones:

1/help       # show available commands
2/clear      # clear the conversation history
3/compact    # summarize the conversation to free up context
4/init       # generate a CLAUDE.md for this project
5/model      # switch the Claude model
6/config     # open settings
7/login      # authenticate
8/cost       # show token usage and cost for the session
9/review     # review code changes
10/agents     # manage subagents
11/mcp        # manage MCP servers
12/memory     # edit memory / CLAUDE.md files

Custom skills also appear as slash commands - a skill named

deploy

is invoked with

/deploy

.

Some tutorials invent slash commands for attaching files, saving output, or managing context. Those do not exist. To reference a file, use

@

in your prompt. To save Claude's work, just ask it to write the file - it can edit files directly.

The Ship's Logbook: CLAUDE.md

CLAUDE.md

is a markdown file in your project root that records your conventions, architecture, and common commands. Claude Code loads it automatically every session, so your first mate always knows the rules of the ship.

• Generate one instantly with

  /init

• There is also a global

  ~/.claude/CLAUDE.md

  for rules across all your projects
• Nested

  CLAUDE.md

  files can live in subdirectories

A good

CLAUDE.md

is the single best way to make Claude Code feel like a crewmate who has sailed with you for years.

Try It Yourself

Open the sandpack and study the example commands. Notice how every command is one of the real forms you just learned - the REPL,

-p

for scripts,

@

for files, and the genuine slash commands.

Summary 🎓

✅ Claude Code - an agentic CLI + SDK from Anthropic that reads, edits, and runs code ✅ Install -

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

, then

claude

✅ Auth -

/login

(Pro/Max or Console), or the

ANTHROPIC_API_KEY

env var ✅ Models -

claude-opus-4-8

,

claude-sonnet-4-6

(default),

claude-haiku-4-5

; switch with

/model

✅ Running -

claude

,

claude "prompt"

,

claude -p

(non-interactive),

-c

/

-r

to continue/resume ✅ File context -

@file

in the prompt and the

add-dir

flag; there is no separate file flag ✅ Real flags -

--model

,

--add-dir

,

--permission-mode

,

--allowedTools

,

--output-format

✅ Slash commands -

/help

,

/clear

,

/compact

,

/init

,

/model

,

/review

,

/agents

,

/mcp

,

/memory

✅ CLAUDE.md - auto-loaded project memory; generate with

/init

In the next exercise we hoist the sails on real vibecoding workflows!

See you on deck! 🚀