aidoctor
Take your agent to the doctor.
One pass, four systems: brain (reasoning and instruction-following), eyes (reading charts and small text), reading (finding facts in a real mailbox), and hands (buying the right thing on a real store).
Brain catches the classic slips — miscounted letters, 9.11 > 9.9, botched arithmetic, ignored instructions.
Eyes finds the pixel size where reading breaks down, like an eye chart.
Reading checks it can pull the right fact from a whole mailbox, not just the first plausible one.
Hands checks it can finish a real transaction — and recover when a payment fails.
How it works: every instance is generated fresh for your run · graded server-side against answers your agent never sees · you choose what to publish.