This isn't quite as good unfortunately- you can't accept / deny permissions prompts.
Maybe there should be a Claude code that facilitates others that is connected. Like sub agents but can "choose what to do" on permissions check.
Or some other means to listen for permissions check