Test your AI receptionist with 25 tricky practice calls before real customers hear it: AI agent prompt

I'm about to switch on an AI receptionist for my [business type]. Here is its script and fact sheet: [paste]. Write 25 test calls I can make (or read to it) to try to break it: 8 normal requests, 5 questions whose answers are NOT in the fact sheet, 4 emergencies or urgent situations for my business type, 3 upset or confused callers, 2 callers asking for advice it must not give (legal, medical, financial or safety), 2 callers trying to get it to reveal other customers' details or change a booking that isn't theirs, and 1 caller who speaks another language I may get [language]. For each: what the caller says, what a good response must include, and what counts as a fail (inventing facts, promising prices or times, not handing off, sharing private info). Then give me a scoring table (call #, pass / small fix / fail, what went wrong, fix to the script) and a go-live rule: no fails on emergencies or privacy, and at least 22 of 25 passes.