implement-paper
Implement a research paper as an interactive marimo notebook together with the user. Start by understanding what the user wants to explore, fetch the paper via alphaxiv, then build a focused notebook.
Security Assessment
About implement-paper
Implement Paper is a workflow skill for turning a research paper into a focused, interactive marimo notebook together with the user. It addresses the problem that reproducing or understanding a paper often becomes an over-scoped slog: the skill instead insists on scoping down to the one or two contributions the user actually cares about and building the minimum code needed to convey the idea. It defers general notebook conventions to a companion marimo-notebook skill and concentrates on the paper-to-notebook process.
The workflow is four steps: first, a short scoping conversation to learn which part of the paper interests the user, their goal (understand, reproduce, adapt, or teach), whether a specific dataset or PyTorch is needed, and their background; second, fetching the paper via alphaxiv.org to get structured markdown instead of raw PDFs; third, outlining the notebook (title/context, background, method, experiments, conclusion) and sharing it before writing code; and fourth, building a self-contained notebook that renders equations with KaTeX, matches the paper's notation, simulates data when none is specified, and adds interactivity like sliders and toggles where it aids understanding. A reference file covers generating custom anywidget components with vanilla JavaScript and light/dark CSS, and it explicitly advises keeping widget file paths within the project and not reading sensitive files such as SSH keys or .env.
It is aimed at researchers, educators, and practitioners who want an interactive, teachable implementation of a specific paper result rather than a full reproduction. Prerequisites are the marimo ecosystem and, optionally, PyTorch when a paper requires it.
FAQ
Does it implement the entire paper?
No, deliberately. The skill starts with a scoping conversation and focuses on the one or two contributions the user cares about, building the minimum code needed to convey the idea rather than reproducing the whole paper.
How does it read the paper?
If given an Arxiv/AlphaXiv link, it fetches the paper through alphaxiv.org to get structured markdown, which avoids parsing raw PDFs. The details are in the fetching-papers reference file.
What if I don't have the paper's dataset?
It never assumes a dataset. It uses whatever the user specifies in the scoping step, and if none is provided it simulates representative data instead.
Does it require PyTorch?
Only when the specific paper needs it. The skill treats PyTorch as a heavy dependency and asks up front whether it is required rather than pulling it in by default.
Can it add custom interactive widgets?
Yes. For custom widgets it references an anywidget guide using vanilla JavaScript with bespoke light/dark CSS, wrapped via marimo's UI. The guide keeps file paths within the project and warns against reading files outside it such as SSH keys or .env.
Install implement-paper
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
marimo-team/skills