How to Use Claude Code: From First Launch to Your First Question (August 2026)

Ever searched "how to use Claude Code"? The results split two ways. One is about using the claude.ai chat window, plain old Claude chat. The other is about using the coding agent you run from a terminal, Claude Code. Today is about the second one. Type claude in your terminal to open a session, then actually ask it something, like "what's in this folder?" and see what comes back. A question that only reads files gets an answer with no permission prompt in the way. If you finished installing Claude Code and logging in last lesson, today's task is that one question. Just here to check whether asking something triggers a permission prompt? This lesson alone will answer that.

Running claude in your terminal

You already logged in last lesson, so there's no new authentication needed today. Open your terminal (or PowerShell) and type this.

claude

Once the session opens, you'll see the version number, the model you're currently using, and the folder you're working in, all displayed near the top. Below that is an input box with a blinking cursor, ready for you to type. If you're in a folder you already trusted last lesson, you'll land straight at the input box with no extra prompt. Did you happen to run this from a folder you're opening for the first time? Then you might see that "do you trust this folder" prompt from last lesson again. Remember, it's a safeguard separate from login. If it's the folder you just opened yourself, trust it and move on.

There's one exception, though. If you open your terminal and run claude immediately, right from your home folder (/Users/your-name on Mac, your user profile folder on Windows), the trust confirmation won't get saved. That means the same prompt reappears every time you run it from there. Don't worry if you see it again and again, nothing's wrong. But if you'd rather not deal with it repeatedly, move into a subfolder first, something like the cd Documents you tried last lesson, and run claude from inside there instead. The official docs also recommend starting from a project subfolder rather than your home folder.

If you're curious, try typing /help into the input box right now. It shows you the list of available commands. Today, just two are worth remembering: /help for the command list, and /exit to end the conversation. Pressing Ctrl+D twice (within 0.8 seconds) does the same thing as /exit.

Asking your first question

Type this into the input box.

What's in this folder?

Press Enter, and after a short wait, an answer comes back. If you ran this from a folder with code in it, you could ask something more specific too, like "what does this project do?" It's fine even if the folder you're in only has documents or photos. If you took the earlier suggestion and ran this from a subfolder like Documents, even better. Whatever's actually in the folder, Claude Code opens it and describes what it finds.

One question isn't going to make a visible dent in your usage. But it's worth knowing that Claude Code runs within your subscription's usage limits too. It shares the same limit as Claude chat, and that limit resets on a rolling window every five hours. It's not unlimited, but a question or two like today's is nowhere near enough to worry about.

You might be wondering something here. If it's about to open a file, shouldn't a permission prompt show up? It doesn't. Read-only actions inside your working folder run immediately with no confirmation, regardless of which permission mode is currently on. Editing files or running commands is a different story. Right now, in the default manual mode, Claude Code asks for permission every single time. But starting August 14, 2026, new sessions on Pro, Max, and Team default to auto mode instead. In this mode, only a background safety check runs. Unless an action is judged risky, it just proceeds with no permission prompt. If an action is judged risky, instead of asking for permission on the spot, it's blocked first and a different approach is tried. Only after getting blocked repeatedly, three times in a row or twenty times total in a session, does it fall back to asking for permission. Later on, in the Claude Code track, you'll ask it to actually edit a file for you. What happens then depends on which permission mode is on. In manual mode, you'll see the permission prompt directly. In auto mode, you'll watch it proceed quietly. Either way, that's the moment you first encounter the "edit and run" layer, something a read-only question alone couldn't show you.

My own first question wasn't a read-only one like this. I'd started out trying to build something, so I described what I wanted, and Claude Code just started creating files. On the web, the result had always stayed inside the chat window. Here, it turned into an actual file sitting in my own folder. Watching that happen was the strange part.

Once you get a response: what just happened

Let's break down what just happened. Claude Code didn't receive your question and immediately spin up a plausible-sounding paragraph. It actually opened the folder, checked the names of the files and folders inside one by one, and built its answer from that. Compare that to asking a regular chatbot, with no photo attached, "what's in my fridge?" It has no choice but to make something up, since it's never actually opened your fridge door. Claude Code is different. It opens the door first, then answers.

This is what it means for an AI to not just chat, but to read files, run commands, and chain multiple steps together, checking things for itself along the way as needed. That's agentic AI. The one question you asked today only showed you one piece of that, reading. Later in the Claude Code track, you'll see this same agent actually edit files, run commands, and carry out multi-step work.

What you should have

Today's task was one thing: ask Claude Code a first question and get a response. That's enough. What matters isn't the content of the question so much as seeing with your own eyes that the AI actually looked inside your folder before answering.

What felt different wasn't the speed, it was the location. On the web, I'd get a result and then have to move it somewhere myself. Here, the work happens inside my own folder from the very start.

That didn't make it any less unsettling. If anything, that's exactly why it felt unsettling. That same feeling from the last lesson, folders appearing and disappearing on their own, shows up again here. So at first, I made one empty folder and only ever gave it instructions there. Having a spot where nothing valuable is at stake, even if something breaks, makes it a lot easier to relax.

Up next: wiring in a connector

What you saw today is the shallowest layer of agentic AI, reading. The next lesson covers wiring a connector you've already used on the web into Claude Code on your terminal. Editing files, running commands, and chaining multiple steps together comes after that, in the Claude Code track. No need to rush. Once today's screen feels comfortable, that's when to open the next lesson.