How well do agents use test/verification techniques?

Definitely interesting but I don’t think this has enough detail to directly replicate. Apart from the prompts themselves I’d want to know if this is one shot or iterative.