tool desk

The prompt

Build an airline support desk. Each round is one customer and one request, such as moving a flight and refunding the difference in fare. I click through the customer record, the flights, and the payments. I should book what was asked, and I should issue the refund once. Doing the refund a second time should fail the task even if the customer ends up with more money. Forgetting the booking should also fail.

Show the desk tools and the score when I say I am done. Score the records, not the order of my clicks. Different customers. The same customer should start from the same records every time.

# AI : shared Action contract — the page posts {"move": "..."}; keep this shape.
from pydantic import BaseModel


class Action(BaseModel):
    move: str