Score the same checks per arm (a hook/skill/rule on vs off) with a real model. Spends money.
⚠️ paid_ overstates slightly: an injected evalDriver replaces the billed path. The DEFAULT path bills.
paid_
evalDriver
Score checks across arms against the real claude CLI.
claude
Score the same checks per arm (a hook/skill/rule on vs off) with a real model. Spends money.
⚠️
paid_overstates slightly: an injectedevalDriverreplaces the billed path. The DEFAULT path bills.