If it's all vibe coded, how do you know — without review — that the new tests, for a new feature, test anything useful at all?

When I was in a test-driven development environment, one of our rules was that you had to see the test fail. You had to prove that it would actually test what you were trying to test.