Skip to content

Setup and First Session

Level: Beginner · Time: ~40 min · Exam relevance: CCAR-P 7.1, 7.2, 7.3

In plain English

Getting started takes three steps: install the claude command, sign in, and run it inside a project folder. The rest of this page is about working well in that first session: letting Claude explore before it edits, using plan mode for bigger changes, knowing how to stop and undo, and learning the dozen commands you will use every day.

Why it matters

  • Good habits from day one (explore, plan, implement, verify) produce better results than jumping straight to edits.
  • A team rollout starts with a repeatable install and a known authentication path (CCAR-P-7.1).
  • Knowing the diagnostic commands (claude doctor, /doctor, /status, /context) is the first step in supporting other developers (CCAR-P-7.3).

How it works

System requirements (as of 2026-10-24)

Requirement Supported
Operating system macOS 13.0+; Windows 10 1809+ or Windows Server 2019+; Ubuntu 20.04+; Debian 10+; Alpine Linux 3.19+
Hardware 4 GB+ RAM; x64 or ARM64
Shell Bash, Zsh, PowerShell or CMD
Network Internet connection; see the network configuration page for proxies and firewalls
Account Claude Pro, Max, Team or Enterprise; Claude Console; or Amazon Bedrock, Google Cloud, Microsoft Foundry

Install

The native installer is the recommended route; it auto-updates in the background.

# macOS, Linux, WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell
irm https://claude.ai/install.ps1 | iex

Alternatives (these do not auto-update by default):

brew install --cask claude-code        # Homebrew, stable channel (claude-code@latest tracks the latest channel)
winget install Anthropic.ClaudeCode    # Windows Package Manager

Verify the installation in a new terminal:

claude --version     # prints a version number
claude doctor        # read-only diagnostics: install health, settings validation, warnings

Note: Piping a script from the internet into a shell is a trust decision. Organisations often prefer package-manager installs, a pinned version, or a corporate launcher distributed by device management. minimumVersion and requiredMinimumVersion managed settings let administrators control versions (page 8).

Sign in

Run claude in a project folder. On first use it prompts you to log in through the browser for a Claude subscription or Console account. With a Console account, a "Claude Code" workspace is created automatically for cost tracking. To switch accounts later, use /login; /logout signs out. Cloud-provider authentication (Bedrock, Google Cloud, Foundry) is configured with environment variables or the /setup-bedrock and /setup-vertex commands.

A productive first session

cd ~/src/acme-checkout
claude
  1. Orient. Ask questions before asking for changes: "what does this project do?", "how are the tests run?", "where is the payment logic?". Claude reads files and runs read-only commands to answer.
  2. Create project memory. Run /init. Claude analyses the codebase and writes a starting CLAUDE.md with build commands, test instructions and conventions. Review it and commit it (page 3).
  3. Plan before large changes. Press Shift+Tab to cycle to plan mode (or type /plan). Claude explores and proposes a plan without editing source files. Approve or refine the plan, then let it implement.
  4. Give verifiable goals. "Make test_applies_loyalty_discount pass without changing the test" is better than "fix the discount".
  5. Review changes. /diff shows what changed; /code-review and /security-review analyse the diff.
  6. Undo if needed. Esc interrupts the current action. Esc twice (on an empty prompt) or /rewind rolls code and conversation back to a checkpoint.
  7. Commit. Ask Claude to commit with a descriptive message, or do it yourself.

Everyday commands and shortcuts

Command or key What it does
/help List commands
/init Generate or improve CLAUDE.md
/clear Start a new conversation with empty context (use between unrelated tasks)
/compact [focus] Summarise the conversation to free context, optionally with a focus
/context Visualise what is using the context window
/resume, claude -c, claude -r Resume a previous conversation
/rewind Roll back code and/or conversation to a checkpoint
/model, /effort Switch model or effort level
/permissions View and edit allow, ask and deny rules
/memory Open memory files and toggle auto memory
/mcp Manage MCP servers and authentication
/usage (alias /cost) Token usage and plan usage
/status Version, model, account, and which settings sources (including managed) apply
/doctor Setup checkup inside a session
/diff Review working-tree changes
Shift+Tab Cycle permission modes (for example manual, accept edits, plan, auto)
Esc Interrupt Claude; Claude keeps work done so far
! at the start of input Shell mode: run a command yourself and share its output with Claude
@ Mention a file path with autocomplete
Ctrl+O Toggle the detailed transcript view

Permission mode on first launch

With Claude Code v2.1.283 or later, interactive terminal and VS Code sessions start in auto mode by default (a background classifier reviews actions such as shell commands and blocks risky ones), unless settings say otherwise. claude -p and the Agent SDK start in default (manual) mode. Organisations can change the starting mode or disable auto mode with managed settings. Page 4 explains each mode.

Example

A realistic first hour on a fictional repository, written as the prompts a developer would type:

> what does this project do, how is it structured, and how do I run the tests?
> /init
> [Shift+Tab to plan mode]
> We need to add a "gift card" payment method alongside card and PayPal. Look at how the existing
  methods are implemented and propose a plan, including tests. Don't edit anything yet.
> Plan looks good, but put the provider client in src/payments/providers/ like the others. Go ahead.
> run the full test suite and the linter, and fix anything you broke
> /diff
> /security-review
> commit this on a new branch called feature/gift-cards with a descriptive message

Enterprise use case

Woodgrove Bank (fictional) prepared a "first day with Claude Code" guide for 400 engineers. Instead of the curl installer, IT distributed Claude Code through its software portal with a pinned version and a managed settings file already in place. The guide asked every engineer to run claude doctor, then /status to confirm the Setting sources line showed the bank's managed policy, before opening any repository. The first exercise was a read-only exploration of a sample service, then /init on a sandbox repository, then a plan-mode change reviewed with /diff. Support tickets in the first week dropped compared with an earlier tool rollout, because the most common problems (PATH issues, wrong account, missing policy) were caught by claude doctor and /status.

Common mistakes

  • claude: command not found after installing: the install directory is not on your PATH yet; open a new terminal or follow the PATH fix in the troubleshooting guide.
  • Signing in with a personal account on a work machine. Organisations can enforce the login method and organisation with managed settings (forceLoginMethod, forceLoginOrgUUID).
  • Editing before exploring, especially on unfamiliar code.
  • Never clearing context, so one long session accumulates unrelated history.
  • Using Homebrew or WinGet and forgetting to upgrade. Those installs do not auto-update by default.
  • Approving prompts without reading them. Prompt fatigue is a real risk; tune allow rules instead.

Security considerations

  • Install from official sources only, and prefer managed distribution in organisations.
  • Start in repositories you trust. The workspace trust dialog lists permission rules a repository wants to grant; read it.
  • Avoid pasting secrets into prompts; keep .env files out of reach with deny rules (page 4).
  • ! shell-mode commands you type run outside the sandbox in most sessions, because the sandbox applies to commands Claude runs.

Troubleshooting

Symptom Likely cause Fix
claude not found PATH not updated Open a new terminal; see the install troubleshooting page
Install command shows no progress Normal; the installer downloads silently Wait, then run claude --version in a new terminal
Logged into the wrong account Cached credentials /logout then /login
Settings seem ignored Invalid JSON or overridden by a higher level claude doctor; /status to see sources
Search fails ripgrep missing (some Linux distributions) Install ripgrep per the setup page

Check your understanding

  1. Which command gives read-only installation and settings diagnostics without starting a session?
  2. What does /init produce, and what should you do with it?
  3. How do you get Claude to propose an approach before it edits anything?
  4. What is the difference between /clear and /compact?
Answers
  1. claude doctor (or /doctor inside a session for a setup checkup).
  2. A starting CLAUDE.md with build and test commands and project conventions; review it, refine it with knowledge Claude could not discover, and commit it for the team.
  3. Switch to plan mode with Shift+Tab (or /plan) and ask for a plan; approve it before implementation.
  4. /clear starts a new conversation with empty context; /compact keeps the same conversation but summarises it to free space, optionally with a focus.

Further reading