Your First Session

Tier 0 · Total Beginner 8 min read

This is the page where the tool finally does something. We’ll start a session, give it a goal in plain English, watch it work, and end the session. You don’t need a coding project yet — any folder will do.

What “a session” means

A session is one running conversation with the agent. You start it, you go back and forth, and when you’re done you end it. While a session is open, the agent remembers what you’ve said in that session and can see the folder you started it in.

Important to understand early: when you end a session, that working memory is gone. Anything that needs to last — notes, decisions, code — has to be written to a file on disk. (This “files, not chat memory” idea is the backbone of everything in Tier 1 and beyond. For now, just know that closing a session is like closing a notebook: only what you wrote down survives.)

Step 1 — Pick a folder and start

The agent works inside whatever folder you launch it from — its working directory. Let’s make a scratch folder so there’s nothing to break:

Make and enter a practice folder:
Terminal window
```
mkdir ~/claude-practice
cd ~/claude-practice
```
Start a session:
Terminal window
```
claude
```
You’ll see a prompt. Above it, Claude Code shows its version, the current model, and the working directory it’s pointed at — confirm it says ~/claude-practice.

Step 2 — Ask it something

You talk to the agent in plain English — no special syntax. Type this and press Enter:

What files are in this folder?

The folder is empty, so it’ll tell you so. That’s fine — the point is you just watched it use a tool (it looked at the folder) rather than guessing.

Now give it a real, tiny task:

Create a file called notes.txt with three things a beginner should remember about Claude Code.

Step 3 — Watch it work

This is the part worth slowing down for. The agent doesn’t just reply with text — it takes actions, and it shows you each one. You’ll typically see:

A short plan or statement of what it’s about to do.
A tool call — here, a request to write a file. Claude Code will show you the file and its contents and ask for your permission before it actually writes anything. (The next page, Permissions & Safety, is entirely about these prompts.)
The result — once you approve, it writes the file and confirms it’s done.

Approve the write when asked. Then check that it really happened — ask:

Show me what's in notes.txt

It reads the file back to you. You just completed the full loop: a goal in English → the agent chose tools → it acted → you saw the result. That loop is the whole game.

Step 4 — End the session

When you’re done, leave cleanly:

exit

(Or press Ctrl+D.) The session closes. The conversation memory is gone — but notes.txt is still sitting in your folder, because it’s a file. That’s the lesson in one move: the chat evaporated, the file survived.

Coming back later

You don’t have to start from scratch next time. From the same folder:

claude -c continues your most recent conversation in that directory.
claude -r lets you pick an earlier conversation to resume.

Try it: run claude -c from ~/claude-practice and ask “what did we just do?” — it’ll remember the notes file.

What just happened, in one sentence

You gave a plain-English goal, the agent picked the right tools, asked permission for the risky one, did the work, and you verified the result on disk. Every more advanced thing on this site is that same loop with more tools, more memory, and less hand-holding.

Next: Three short pages go deeper on what you just saw — The Working Directory & First Edits, Permissions & Safety, and Reading the Output. Start with the working directory.