Claude Code: from installation to your first useful workflow

What Claude Code is

Claude Code is Anthropic’s agentic coding tool. It runs from your terminal, reads the project you open it in, and can explain code, edit files, run commands, and help verify whether a change works.

The important difference from a normal chat is context: Claude Code can work with the actual files in your project. That makes it useful for understanding an unfamiliar codebase, fixing a contained bug, improving documentation, or building a feature with clear boundaries.

Install and sign in

Anthropic currently supports macOS, Linux, and Windows. On Windows, the official setup supports WSL or native Windows with Git for Windows. Follow Anthropic’s setup page for the installation option that matches your computer.

# After installation, open your project folder
cd your-project

# Start an interactive Claude Code session
claude

On first launch, Claude Code will guide you through authentication. Anthropic documents options including a supported Claude subscription, an Anthropic Console account with billing, and enterprise cloud platforms.

If setup behaves unexpectedly, run claude doctor and use the official troubleshooting guidance rather than repeatedly reinstalling.

Your first useful workflow

1. Ask for orientation

Begin with understanding, not editing. This gives you a quick map and lets you judge whether Claude has read the project correctly.

Summarize this project for a new contributor.
Explain the main folders, how it starts, and how tests are run.
Do not change any files.

2. Choose one contained outcome

Good first tasks have a visible finish line: clarify a README section, improve an empty state, add a small validation rule, or fix one reproducible bug.

3. Ask for a plan before edits

I want to improve the empty state on the search page.
Inspect the relevant files and propose the smallest safe plan.
Tell me which files would change. Do not edit yet.

Read the plan. Correct any wrong assumptions and narrow the task if it has started to grow.

4. Make the change and verify it

Make the agreed change only.
Then run the most relevant existing check and summarize:
1. what changed
2. what you verified
3. anything I should review manually

Review the actual file changes. A successful command is helpful evidence, but it does not replace your judgment about whether the result matches your intent.

Five common beginner mistakes

1. Starting too broadly. “Improve this whole app” creates ambiguity. Name one outcome and one boundary.
2. Editing before understanding. Ask Claude to explain the relevant area and propose a plan first.
3. Ignoring permission prompts. Read what Claude wants to run or change. Permission checks are part of staying in control.
4. Skipping verification. Ask for the project’s existing test, build, lint, or type-check command—not a made-up check.
5. Accepting a summary without reviewing the diff. Inspect the changed files and test the important behaviour yourself.

What to try next

Repeat the same loop with a slightly more meaningful task: orient, define, plan, edit, verify, review. Once that feels natural, create a project-level CLAUDE.md with durable guidance such as the commands to run, architectural boundaries, and conventions that are easy to miss.

You can also explore Claude Code’s session controls. The official CLI supports continuing the latest conversation and resuming a specific session, which is useful when a task spans more than one sitting.