implement-paper-auto

implement-paper-auto is a skill for implementing a research paper in a marimo notebook fully automatically, without asking the user for feedback. Its central premise is that a good implementation tells a story: rather than fully reproducing a paper or rerunning a giant benchmark, it picks the single most compelling idea and explains it clearly in a minimal, CPU-runnable notebook that a reader can actually learn from.

To get reliable source material, the skill fetches structured, LLM-friendly paper content from alphaxiv.org. It parses the paper ID from various arXiv/alphaxiv URL forms, then fetches an AI-generated overview via a read-only curl to the public overview endpoint, falling back to the full extracted paper text endpoint when a specific equation, table, or proof is needed, with documented handling for 404s (direct the user to the PDF as a last resort). No authentication is required for these public endpoints. The guidance then focuses on craft: pick the idea easiest to explain with a minimum code example, favor one or two informative charts and lightweight interactivity (a dropdown or slider), avoid overdone toy examples in favor of a creative example that shows where one approach shines, and keep scrolling to a minimum.

It targets researchers, educators, and developers who want to distill a paper into a clear, runnable marimo notebook automatically. The only shell usage is read-only curl against public alphaxiv endpoints, so it is a benign research-and-education tool.