Live quota readings for the AI coding agents I run: how much of each plan's window is used, and the pace that would spend it by the reset.