I’ve thought of this too. If prompts allow the ability of saving of data that goes onto a public website like a dashboard without sanitizing output then you can do the traditional XSS hacks.

Another solution could be to make a system that attempts to recognize malicious input somehow .