On how does the mocking work, that's a really interesting question. We do a lot of AST parsing - for both code and build configuration languages. Even then, we

On how does the mocking work, that's a really interesting question. We do a lot of AST parsing - for both code and build configuration languages. Even then, we still have to rely on the LLM to figure out a lot of the details. Making this work reliably for non-frontier models and codebases that don't have existing test harnesses is where a lot of the design work goes in.