Your AI helper doesn't ask before it deletes
Why an AI that writes code needs brakes, and a $27 add-on that makes it stop and ask first.
If you let an AI write code on your computer, it can also run commands on your computer. Most of the time that is the point. Once in a while it deletes the wrong folder, wipes a database, or pushes unfinished work to the live site, and it does not ask first.
The problem in plain terms
An AI helper has no sense of what you cannot afford to lose. To it, deleting a test file and deleting your only copy of a project are the same kind of command. A reminder in the prompt can be forgotten a few hundred steps into a long session. A rule that sits outside the AI, and checks each command before it runs, cannot be forgotten.
What the Claude Code Seatbelt Kit does
The kit installs a set of rules between your AI coding helper and your computer.
- The most destructive commands are refused outright. Examples: dropping a database, deleting everything, force-pushing over your main branch, adding your secrets file to a repository.
- The risky ones make the AI stop and ask you first. Examples: other recursive deletes, resetting your work, pushing or deploying, editing secrets files.
- Every 25 tool calls it restates your standing rules, so a long session does not drift away from what you asked.
- Every decision is written to a log on your own machine.
It runs locally. There is no account and no subscription.
What it cannot do
It matches commands and paths against its rules, so it only catches what the rules describe. You can edit the rules. It does not make an AI smarter. It is a seatbelt, not a guarantee, so keep backups.
Who it is for and what it costs
It is for anyone who lets an AI write or run code every day. It works with Claude Code first. Other tools are listed on the product page. It needs Python 3.10 or newer, and the installer offers to install that for you. It is $27 once, with a 14-day money-back guarantee.