An agent spec is a contract. An eval is the proof the contract is still being kept. This two-page workbook walks you through writing one eval for one agent, using a refund agent as the worked example. Page 1 explains what an eval checks (outcome, tools, path, cost, rules), who checks it (script, AI judge or person), and why evals keep running to catch Quiet Drift. Page 2 is the sheet you fill in: the clause, the test, the pass checks, the never line, the threshold and the remedy. You finish with an eval ready to file under property 6 of the agent spec. Built on Trusted Evals, the first of the Four Pillars of GOVERN/ASSURE.
Join Salim Ismail for The 10x Shift โ a live, hands-on workshop where you'll apply the same exponential frameworks used by the world's fastest-growing companies.