L04 · Setup · 5 min

The terminal, without fear

In one sentence

The terminal is a text window for running direct instructions on your computer, and while your coding agent runs almost every command for you, it explains each one in a single plain sentence before asking your approval.

Think of it like this

Using a normal app with menus and buttons is like pointing at photos on a restaurant menu. Using the terminal is like passing a short written order straight through the kitchen hatch—and your coding agent is the bilingual server who writes the slip and shows it to you before handing it in.

How it works

A DefinitionterminalA window where you type commands instead of clicking. Your agent mostly uses it for you.Open in the glossary → (built into the Claude desktop app with ` Ctrl + ``, or called Terminal on Mac and PowerShell on Windows) is simply a text conversation with your computer. Instead of clicking folders and buttons, you type a DefinitioncommandAn instruction typed in the terminal, such as `git --version`. Your agent explains each one before running it.Open in the glossary →—a one-line instruction—and press Enter.

You don't need to memorize commands. Your Definitioncoding agentAn AI that works in your project folder: it reads files, writes code and runs commands, with your permission.Open in the glossary → types and runs them for you, following a strict rule: before running any command, it explains what the command does in one plain sentence and waits for your permission in Manual mode.

Harmless check commands just ask a tool to introduce itself: git --version, node --version, and gh --version print the installed version numbers of DefinitionGitThe tool that keeps your repo's memory: it records every commit on your computer.Open in the glossary →, Node.js, and the GitHub tool without changing anything on your computer. Other commands start your app locally (npm run dev on DefinitionlocalhostYour own computer acting as a server while you test, usually at `http://localhost:3000`.Open in the glossary →) or save checkpoints.

When the terminal is running a live process—such as your local test server at http://localhost:3000—it stays busy watching for changes. Whenever you want to stop a running command in the terminal and get the prompt back, press `Ctrl + C` (even on a Mac, it's Ctrl, not Cmd).

1 · ProposalPlain explanationThe agent shows the command (e.g. `git --version`) and explains it in one sentence.
2 · PermissionYou approveIn Manual mode, nothing runs until you review the sentence and click Allow.
3 · OutputTerminal repliesThe computer prints the result (or a clear error message) for you and the agent.
FIG. 1 · How terminal commands run with an agent: one-sentence explanation first, your approval second, and clear text output third.

What it looks like in your project

During ① Setup, your agent runs git --version, node --version, and gh --version to see which tools are already on your computer and which ones need to be installed. If you just installed a tool and the terminal says “command not found”, simply close and reopen the Claude app so it refreshes its tool list.

Copy for your agent

Check whether git, node (22.20+), and gh are installed on my computer. Explain each command in one plain sentence before running it.

Quick check · 1 question

What should happen before your coding agent runs a command in the terminal?

Go deeper

Next lessonWhat is a repo? Git and checkpoints

Last reviewed Nov 8, 2026