deepgrounding/recursive-self-improvement ? reverse-engineered prompt

Build me a reproducible research repo for a paper about recursive self improvement in AI, with the manuscript source, figures, bibliography, and the paper corpus all in one place.

I want a clean setup where I can regenerate the paper from the data, including the 1,250 paper corpus, the taxonomy labels, the summary tables, and the main manuscript. Please include scripts that can reclassify the corpus, rebuild the bibliography from the corpus, regenerate the key figures and table, and export the manuscript into a compile ready paper PDF.

Keep the organization simple, with the data files, the draft source, and the helper scripts separated in a way that is easy to understand. If something depends on live paper metadata, make it work with sensible defaults and document it clearly. I also want the final manuscript to be ready for arXiv style submission, with figures and citations wired up properly.

If you need to check current docs for any tools along the way, go ahead and look them up online.