Start Here: Your First Session with Claude Code

Claude Code is an agentic coding tool: you describe an outcome, and it reads your codebase, edits files, runs commands, and iterates until the outcome is real — all from your terminal, IDE, desktop app, or browser.

That one sentence hides a mental-model shift. You are not autocompleting lines anymore; you are directing an agent that acts. This guide gets you from zero to a productive first session and installs the right instincts from the start.

What Claude Code actually is

Under the hood, every session is a loop: Claude reads your request, gathers context with tools (reading files, searching, running commands), takes an action, observes the result, and repeats until done. Everything else — memory files, skills, hooks, subagents — exists to make that loop smarter, safer, and faster.

The loop you are now part of

Your job in the loop is judgment: define the outcome, supply constraints, and verify results. Claude's job is execution. The biggest beginner mistake is treating it like a search engine — issuing vague one-liners and accepting the first answer. Treat it like a sharp new teammate instead: give it the context you would give a person.

Pick your install path

The native installer (curl … | bash) is the recommended route and the only one that auto-updates in the background. [V] Every other method — Homebrew, WinGet, Linux package managers, npm — installs the same native binary but leaves you to upgrade on your own schedule. [V] Pick the row that matches how you already manage software on the machine.

Drive the session from the keyboard

Most of your speed gains come from four keystrokes, not from typing longer prompts. The two that matter most on day one are the permission-mode cycle and the rewind. [V]