Claude Code & Agentic Development · Lesson 1

Installation and First Session

Install Claude Code, authenticate, start a first session, learn permission modes, and how it differs from Cursor/Copilot.

20 min read4 questions in quizReady prompt includedIn progress

Practical exercise

What to do after this lesson

Install Claude Code globally, run claude auth, start a session in any of your repos, and ask the agent to explain what one of the files does. Then restart with --permission-mode plan and confirm it only proposes a plan.

Ready-to-use prompt

Template for this lesson

Copy and adapt to your context. Text in angle brackets should be replaced.

I just installed Claude Code and started a first session in a project.
Project stack: <…>
What I want to understand to start: <…>

Briefly explain this repository's structure from the files you can see,
and propose 3 safe first tasks you could perform.

Common mistakes

What people get wrong

Running claude not from the project root — the agent doesn't see the relevant files.
Enabling auto-approve immediately on an unfamiliar project and losing control of edits.
Treating it like Copilot autocompletion and dictating line-by-line instead of setting a goal.

Pro tips

What works but no one documents

First session

Go to the project root and just run:

cd my-project claude

An interactive REPL opens right in the terminal. This is not an editor plugin — it is an agent with access to the filesystem, shell, and git within the working directory. Describe a task in plain language:

Read src/server.ts and add error handling to the /api/users route

Claude finds the file itself, reads related modules, proposes changes as a diff, and asks for confirmation before writing.

Permission modes

Claude Code does not perform dangerous actions silently. There are three behaviors:

default — asks for confirmation on file writes and shell commands. The safe default.

auto-approve (acceptEdits) — auto-accepts file edits but still asks for potentially dangerous shell commands.

manual / plan — the agent only proposes a plan and changes nothing until you explicitly allow it.

Choose the mode at launch:

claude --permission-mode auto-approve

You can also switch the mode inside a session via the permission UI (Shift+Tab toggles the edit-acceptance mode).

How it differs from Cursor and Copilot

Copilot — autocompletion in the editor. Suggests the next line, doesn't see the whole project, doesn't run commands.

Cursor — an editor with chat and edits; the human drives every step in the IDE.

Claude Code — an autonomous agent in the terminal: it reads dozens of files, runs tests, makes git commits, and drives a multi-step task to a result. You set the goal, not dictate the steps.

This changes how you work: you describe what you need, not how to type it. That's why the rest of the course is about how to phrase tasks and keep the agent under control.

Сообщить об ошибке

Installation and First Session

Квиз — 4 вопроса

Installation

Authentication

First session

Permission modes

How it differs from Cursor and Copilot