Your first few conversations will hit a moment that feels like a record scratch. Claude stops mid-task and a panel slides in asking you something.
Before doing anything that changes your files or runs something on your computer, Claude pauses and asks. You click to allow it or not. Nothing happens until you do.
That’s the permission prompt. It is the most important safety feature in the app, and the one people are most tempted to switch off. Don’t.
What you’ll see
You ask for something — say, deleting an empty file. Instead of doing it, a panel appears showing:
- What Claude wants to do — the exact action.
- Where — which file or folder it lands on.
- Your choices — allow once, allow from now on, or decline.
You click. Claude waits. This is the app’s Ask permissions mode, and Anthropic’s own docs recommend it for new users. Leave it on.
Three answers, three sizes of consequence
- Allow once. Just this time. The same thing next time will ask again. This is your default.
- Allow from now on. Skip the prompt for similar actions in future. Save this for things you’d cheerfully do yourself without thinking — like reading a file.
- Decline. Don’t. Then tell Claude what to do differently, in the chat.
Prefer allow once for now, for two reasons. You haven’t yet built the instinct for which actions matter. And “allow from now on” is a one-way door: each one removes a future chance to catch a mistake.
Why this is good rather than annoying
Claude makes mistakes. Wrong file, wrong folder, right idea applied in the wrong place. The permission prompt is the gap where you catch those before they happen instead of after.
The annoying kind of safety is the kind that doesn’t prevent anything. This one prevents real damage, every day, for anyone who actually reads the panel before clicking.
So read the panel. It takes two seconds and it is the entire point.