Claude Code — Part 3 — Creating and Navigating Your First Project

Introduction

Part 2 got Claude Code installed and verified. This post covers the part that actually matters day to day: starting a session in a real project folder, letting it understand what is there, and making your first change.

There is no dashboard, no “New Project” button, and no configuration wizard to get through first. A project is a directory, and a session starts the moment you run claude inside one.

Starting a Session

Open a terminal in the folder you want to work in and start Claude Code from there.

# Move into the project or scripts folder you want to work in
cd C:\Scripts\ADHealthCheck

# Start an interactive session
claude

The prompt that appears shows the current version, the active model, and the working directory above it, so you always know what scope a session is running in. If you are not logged in yet, Claude Code prompts you to authenticate in a browser the first time. To switch accounts or re-authenticate later, run /login inside a running session.

Letting Claude Explore First

Before asking for a change, it is worth spending the first message or two just asking what is actually there. This is closer to how you would brief a new colleague than how you would use a search engine.

what does this project do?
what technologies does this project use?
where is the main entry point?
explain the folder structure

This matters more in operations folders than it sounds. A scripts directory built up over several years rarely has a README explaining which script is still in use, which is a leftover, and which naming convention actually means what. Asking Claude to explain the folder structure first, before asking it to change anything, catches that kind of drift early.

Making Your First Change

Once Claude has some context, describe a real task in plain language rather than a rigid command syntax.

add error handling to the SMB connectivity check function

For a task like that, Claude Code finds the relevant file, shows you the proposed change as a diff, and asks for approval before writing anything. You can approve individual changes one at a time, or turn on an accept-all mode for the rest of the session if you trust the direction it is taking.

Git, Conversationally

Common git operations work the same way, described rather than typed out command by command.

what files have I changed?
commit my changes with a descriptive message
create a new branch called fix/smb-timeout
show me the last 5 commits

This is useful for engineers who use git occasionally rather than daily. You do not need to remember the exact flag for “show the last N commits in one line” if you can just ask for it.

Everyday Navigation

A small set of commands covers most sessions. Shell commands are run from your terminal, before or instead of an interactive session. Session commands are typed inside Claude Code once it is running.

A few keyboard habits are worth building early: Tab completes commands and paths, the up arrow recalls command history, and Shift+Tab cycles through permission modes, useful once you want to move faster than approving every single edit individually.

Final Thoughts

There is no ceremony to starting a project in Claude Code. The directory you are in is the project, the first few messages are worth spending on exploration rather than instructions, and every edit stays visible before it is applied.

Once a session feels natural, the next step is making that context persist, so Claude does not need to re-learn your conventions every single time you open a new session.