# Your first conversation with Claude Code **Author:** Ivan Misic **Published:** 2026-01-09 **URL:** https://ivanmisic.net/blog/ai-tools/first-conversation-with-claude-code **TL;DR** Use a separate practice folder for the first Claude Code session. Three small prompts let you inspect the folder, create a file, and edit it while checking each result. - Start Claude Code inside the practice folder so the intended file scope is clear - Ask what files are present before requesting a change - Create `hello.txt`, read it back, then change its message and inspect the file after each step - Permission behavior depends on the active mode; a prompt reduces risk but does not prove the action or result is correct - Use precise filenames and expected results, then keep Git or checkpoints when the files matter Use a new practice folder for your first session. With three prompts, you will inspect the folder, create a text file, and edit it while checking each result. ## Starting Claude Code Do not start the first session in your whole Documents folder. Use a separate folder so the files in scope are clear. In Windows PowerShell: ```bash mkdir "$HOME\Documents\practice" cd "$HOME\Documents\practice" ``` On macOS or Linux: ```bash mkdir ~/Documents/practice cd ~/Documents/practice ``` Now start Claude Code: ```bash claude ``` Type instructions at the prompt in plain language, but name the file and expected result when you can. ## Three Starter Prompts Let's do three things, each one building on the last. **Prompt 1: Ask a question** Type this: ```text What files are in this folder? ``` Claude should inspect the folder and report that it is empty. The useful point is that it checks the local folder instead of answering from the conversation alone. **Prompt 2: Create something** Type this: ```text Create a file called hello.txt that says "This is my first file created by AI" ``` In the default permission mode, Claude asks before creating the file. If your interface is in `acceptEdits`, `auto`, or another configured mode, the prompt can differ. For this exercise, use default mode and review the proposed change before approving it. You may see something like: ```text I'll create the file hello.txt with that content. > Create file: hello.txt Allow? [y/n] ``` Approve the action, then open the practice folder and confirm that `hello.txt` exists with the expected text. **Prompt 3: Change something** Type this: ```text Read hello.txt and change the message to say "I told an AI to create this file, and it actually did" ``` Review the proposed edit before approving it, then reopen `hello.txt` and verify the exact sentence. Those three prompts cover the basic loop: describe an action, review the proposed change, and check the file afterward. ## Check the Three Results The exercise produced three concrete results: - Created a real file on your computer by describing what you wanted - Modified that file by describing the change in English - Didn't write a single line of code, HTML, or anything technical If you've used web-based AI chats like ChatGPT, Gemini, or even Claude on claude.ai, you know the workflow: copy content into the chat, attach a document, get a response, copy it back out, paste it into your file, repeat. It works, but it's clunky. Especially when you're working with multiple files or making iterative changes. Claude Code can read and change files in the working folder, which removes the copy-and-paste step. You still need to review the proposed action and inspect the file after the change. The same loop applies to larger work: describe one outcome, review the proposed action, check the result, then continue. Larger tasks also need clearer constraints, testing, and recovery points. ## The Permission System In default mode, file changes and most shell commands require approval. Read-only file access and a built-in set of read-only shell commands do not. Other modes can auto-approve more actions. A permission prompt reduces risk, but it does not prove the change is correct. Read the proposed action, inspect the result, and use Git or checkpoints when the files matter. ## Exiting Claude Code Type `/exit`, or use the exit shortcut documented for your terminal. In the current CLI, `Ctrl+C` exits after a second press when nothing is running, while `Ctrl+D` asks for a second press within a short interval. Your files stay, and the conversation is saved locally. Run `claude` for a new session, `claude --continue` for the latest session in this directory, or `claude --resume` to choose one. ## Try a Few More Things Before we move on, try these prompts. Each one shows you something Claude Code can do: - "Create an HTML page with a dark background and the text 'Hello World' in large white letters" - "What's in hello.txt?" - "Create a folder called notes and put a file in it called ideas.txt with three random project ideas" - "List all the files in this folder and its subfolders" You do not need to understand the HTML yet, but open the result after each prompt and check whether it matches what you asked for. ## My First Session In my first Claude Code session, I asked for a simple HTML page with a dark background and some text. I opened the file in a browser, then asked it to increase the text size, add my name, and create a section with three skills. Thirty minutes in, I had a basic personal page, and I never touched HTML. I learned to describe what I wanted clearly enough for Claude to build it. The skill is describing a result clearly enough to review it. Claude can write the code, but you still decide whether the result is correct. > Next up: which files Claude Code can access and how the working folder affects that scope.