0.2 · Your Cockpit

0.2 · 20 min

Your Cockpit

Every lab in this course runs in your terminal, on your machine, against a real git repo. Not a sandbox, not a video of someone else's terminal. If your environment is broken, every lesson after this one degrades into reading instead of practice. So this lesson has exactly one job: a working cockpit, verified.

What you are installing

Claude Code is a command-line program that runs on your machine and operates on your files. Installing it puts the harness from Lesson 0.1 under your fingers: the model runs on Anthropic's servers, but the tools (reading files, editing, running commands) execute locally, in your working directory, with your permissions. That is why it needs an install at all, and why the permission prompts you will see later are not decoration.

As of July 2026, the requirements are modest: macOS 13.0+, Windows 10 1809+, or a mainstream Linux (Ubuntu 20.04+, Debian 10+), 4 GB of RAM, and an internet connection. You also need an account: a Claude subscription (Pro, Max, Team, or Enterprise) or a Claude Console account with API credits. The free claude.ai plan does not include Claude Code access. If you have neither, pick a subscription; you will burn through this course's labs faster than pay-per-token pricing makes sense.

One position before you start: do the terminal install, not the desktop app. Claude Code also ships as a desktop app, a web version, and IDE extensions, and they are fine. But this course teaches you to drive agents from the shell because that is where scripting, piping, hooks, and CI live. The graphical surfaces are consumers of the skills you are about to build, not substitutes for them.

The practice repo

At the end of the lab you will create a repo named agentic-practice. It is not a throwaway. Labs from here through Part 3 build on it: it will accumulate a CLAUDE.md, custom commands, hooks, and eventually a full harness. Treat it like a real project from commit one, because by Part 3 it will be one.

Lab

Lablab-0-2Install, authenticate, fly

Goal: A verified Claude Code install, an authenticated first session, and the practice repo that carries you through Part 3.

Prereqs: a terminal, git installed (git --version prints a version), and a Claude subscription or Console account.

  1. Install Claude Code. On macOS, Linux, or WSL:
curl -fsSL https://claude.ai/install.sh | bash

On Windows in PowerShell (your prompt shows PS C:\):

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

Homebrew users can run brew install --cask claude-code instead. Native installs update themselves in the background, so this is the last time you think about versions.

  1. Verify the binary exists before doing anything else:
claude --version
  1. Create the practice repo and start your first session inside it:
mkdir agentic-practice && cd agentic-practice
git init
claude

On first run, claude prompts you to log in. Follow the browser flow for your subscription or Console account. Credentials are stored after that; you will not log in again. If you ever need to switch accounts, type /login inside a session.

  1. You are now looking at the Claude Code prompt, with the version, current model, and working directory shown above it. Ask your first question, and make it self-referential:
PromptFirst dispatch: make it show you the cockpit

List every tool you have available in this session, grouped into: tools that read, tools that write or edit, and tools that execute. For each group, tell me which ones require my permission before running.

  1. Read the answer properly. This list is the raw material of Part 1: an agent is a model plus exactly these tools, and the permission split you are looking at is the trust model you will configure in Part 2.

  2. Make the repo real. Ask Claude to create a README.md describing the repo's purpose (your practice ground for this course), approve the edit when prompted, then commit:

git add README.md && git commit -m "chore: init practice repo"

Verify

  • claude --version prints a version number, not command not found.
  • Inside a session, the prompt header shows a model name and your agentic-practice directory.
  • Claude answered the tools question with a grouped list, and you saw at least one permission prompt when it wrote README.md.
  • git log --oneline in the repo shows your first commit.
>Troubleshooting
  • command not found: claude right after install: your shell has a stale PATH. Open a new terminal, or check the install script's final output for the line telling you what to add to your shell profile. claude doctor runs a fuller diagnostic once the binary resolves.
  • The Windows install command errors immediately: you are probably in the wrong shell. irm ... | iex is PowerShell-only; the curl ... install.cmd variant is CMD-only. Your prompt tells you which one you are in (PS C:\ means PowerShell).
  • Login loops or rejects your account: confirm the account tier. The free claude.ai plan does not include Claude Code. Console accounts need prepaid credits before first use.

Where it breaks

The classic failure is not a failed install; it is a half-verified one. The binary runs, so you move on, and three lessons later you discover git was never initialized in the practice repo, or you authenticated with a free account that silently cannot start sessions. That is why the Verify block above checks four independent things instead of one. Run all four. This habit, never trusting "it looks installed" without an observable check, is the course's core discipline showing up on day one, in miniature.

Second failure worth naming: installing into an environment you do not control, like a locked-down work machine where a proxy strips the auth callback. If the browser login flow never completes, that is usually the cause. Use a personal machine for this course; fight your IT department on your own time.

Knowledge check

Knowledge check

Q1You run the install script, then `claude` in the same terminal prints `command not found`. What is the most likely cause and the right first move?
Q2Why does this course insist on the terminal CLI when Claude Code also ships as a desktop app and IDE extensions?
Q3The lab has you ask Claude to list its own tools and their permission requirements. What is that step actually for?

Sources