SHORT ANSWER
Yes. Claude Code, Codex CLI, and Gemini CLI all use a permission system by default — they ask before running commands, editing files, or doing anything meaningful on your machine. You approve or deny each action. The scary stories you may have seen online almost always involve fully autonomous agents running with permissions turned off. That is not what this course does.
How the permission system works
When Claude Code wants to read a file, write a file, run a shell command, or visit a website, it pauses and asks. A prompt appears in your terminal with the specific action it wants to take. You choose whether to allow it.
If you deny the action, the tool explains what it was trying to do and asks for different instructions. Nothing happens on your machine that you didn't explicitly approve.
This is the most important thing to understand about CLI agents: you are in the driver's seat at every step. The tool is watching you as much as you're watching it.
Four rules that keep the blast radius small
These come straight from an answer Joe gave in the Questions for the instructor forum. They apply no matter which tool you're using.
1. Keep the default permission settings
All three tools ship in "ask before acting" mode. Don't turn that off. There is a flag called --dangerously-skip-permissions that lets Claude Code run without stopping to ask — the name is the warning. You won't need it for this course, and you probably won't need it for most of what you do after this course either.
2. Work in a dedicated folder for the course
Create one folder — something like ~/mooc-prompt-eng/ — and launch your sessions from inside it. That way if you ever say yes to something you didn't mean to, the blast radius is whatever files are in that folder. Your documents, your photos, your tax records, your browser profile — none of that is reachable unless you explicitly point the tool at it.
3. Read what the tool is asking to do before you approve it
The whole point of the permission prompt is to give you a chance to say no. It's not a "click through to continue" dialog. Glance at the command or the filename. Does it match what you asked for? If it does, approve. If it doesn't, deny and tell the tool what you actually wanted.
4. You don't need a special IDE
Your regular terminal is fine. You don't need VS Code, Cursor, or any other editor to use the CLI tools safely. The videos in this course run the tools in a plain terminal and that's the expected default. If your IT department blocks the terminal, see the IDE explainer for alternatives.
What to do if something feels off
Close the terminal. That's it. Closing the terminal window kills the session immediately — no ongoing process, no background daemon, no file half-written. The tool stops the moment the window closes.
If you want to double-check nothing is still running, open a fresh terminal and run:
ps aux | grep claude
You should see only the grep command itself in the output. If you see a claude process, note its PID (the number in the second column) and run kill <PID> to stop it. You almost certainly won't need to.
Where the horror stories come from
The scary accounts you might have read online almost always involve one of three things, none of which this course asks you to do:
- ›Fully autonomous agents running with
--dangerously-skip-permissionsenabled, iterating for hours without any human in the loop. - ›Agents given root or sudo access to system files — which they should never need for journalism work.
- ›Long-running unattended sessions on production servers, where nobody is watching and the blast radius is a whole company's infrastructure.
Interactive use inside a dedicated folder — which is everything Module 1 through Module 4 asks of you — has a blast radius measured in "the files you can see scrolling by in your own terminal." That is as safe as AI agents get right now.
ONE MORE THING
If anything ever feels off while a session is running, close the terminal. That kills the session immediately. You don't have to figure out the right command to stop it — just close the window.
READY? Head back to Module 1 to install the tools and run your first CLI command. Or read about what a CLI actually is if you're still not sure what "the terminal" means.