Well, I can understand your frustration, but this is basically a pipeline that takes some code written by an LLM and attempts to prove its correctness using the Lean 4 theorem prover.
I'm sorry if it was unclear. My comment was just a validation of your assertion. It does use Claude - or any openai compatible LLM - to perform the extraction.