What prompts work best for generating pytest tests from an OpenAPI spec?

I have a FastAPI service with a well-maintained OpenAPI spec (auto-generated by FastAPI). I want to generate a meaningful first-pass pytest test suite from it using an LLM. My current prompt approach: paste the full spec JSON + ask for tests. Result: tests that compile but assert nothing meaningful (just status code 200 on the happy path). Better prompts I've tried: - Specifying test categories (happy path, validation, auth, edge cases) explicitly - Asking for one endpoint at a time with the schema pasted inline - Including an example of a test I've already written Anyone have a prompt structure that consistently produces tests with meaningful assertions, not just structural placeholders?