There was a paper recently that tackled this:

https://tiger-ai-lab.github.io/TheoremExplainAgent/