If I understand correctly, this is a two-stage generation.
The first stage generates the web page structure information and description in function determineWikiStructure.
The second stage generatePageContent recalls the code in the codebase based on the web page description and then generates the markdown web page.
However, since the <file_path> is already provided in the web page structure information, why not directly recall the corresponding code first, but perform a retrieval (and when retrieving, embed the entire system prompt rather than page.title and page.file_path)?
I also wonder if the information is enough only with page.title and page.file_path. When generating a page, is it better to generate a page description as well, like WikiStructure?