Projects & coding · Tutorial 01

Claude Code: set up your first project session

Open a practice folder, inspect its files, and establish how the agent will make and check changes before building.

A paper coding arm edits repository file cards as failed and passing code strips circle a test loop.
Reading time
6 min
Last updated
September 2026

0 of 3 complete

Complete & next →

Essentials · Step 2 of 3 · View the route

Last checked and updated: September 8, 2026. Editorial and source review; exercises below are authored practice, not reported benchmark runs.

A coding agent can work with project files and run commands. That makes the first useful skill knowing which project it is working in and how you will check what it changes.

Bring the SPEC.md from the project-scope lesson. You will finish this lesson with a working session in a practice folder and a reviewed implementation plan.

Install and verify access

Follow Anthropic’s official quickstart for the installation method that matches your operating system. It covers terminal installation, authentication, and supported access arrangements. Follow the account route you actually use; a subscription and API billing are different arrangements.

After installation, run this in your terminal:

claude --version

It should print the installed version. If it fails, use the quickstart’s installation troubleshooting before continuing. Do not solve an installation failure by pasting unrelated commands from a different platform.

Create a new folder named expense-practice with your file manager. Put only SPEC.md inside it. Open a terminal in that folder, then start:

claude

Check the working directory shown by the app. Follow the sign-in flow if needed. Installation and access requirements are documented in the quickstart; check them before making a purchase.

Inspect the workspace before writing

Send this first:

Read SPEC.md in this practice folder.
Do not create, change, move, or delete files yet.
Tell me the project directory, the files you found, the expected total,
and the behavior required for empty or invalid input.

Compare the response with the file you wrote. The sample project’s grand total is 35.00. If the agent reports a different directory or cannot find the file, correct that before proceeding.

A workspace is useful context, not proof that every possible tool action is confined to it. Inspect the app’s active permission mode. Review consequential operations, especially commands affecting other folders or external services.

Ask for a small implementation plan

Plan a one-page local expense summary from SPEC.md.
Use the smallest approach that can demonstrate the requested behavior.
List the files to create, how to open the result, and how to run the checks.
Explain any software installation you think is necessary before installing it.
Do not add accounts, external services, deployment, or extra features.
Do not implement yet.

For this exercise, a plan that proposes a database, an account system, or a paid service has expanded the task. Ask the agent to return to the brief.

Keep the plan understandable: what reads the sample rows, what calculates the totals, and what displays them? You do not need to know every syntax rule to notice that the plan has added a server you never asked for.

Agree on what evidence to keep

Ask the agent to retain the check command and its actual result, plus instructions for opening the page. “Tests pass” is less useful than a command you can run again.

Anthropic describes a workflow of exploring the repository, planning when needed, implementing, and verifying. It also documents CLAUDE.md for persistent project instructions. For this small exercise, a clear SPEC.md and direct instructions are enough. Claude Code best practices.

Ready to build

You are ready when the agent is in the intended folder, has read the correct brief, and has proposed a local implementation with concrete checks. Save the plan beside the brief.

Continue to build the expense summary. If you already use another coding agent, you can follow that exercise with it. The required evidence stays the same.

For historical analysis of the September 2026 usage-limit announcement, see the dated usage-limit benchmark record. It is background material, not a prerequisite or a current quota promise.

Check your understanding

Q1.The agent reports a different working folder. What comes next?
Q2.The plan adds an account service to the local expense exercise. What should you do?
Q3.What evidence should accompany a passing test claim?
Q4.What is the deliverable for this setup lesson?