Skip to content
← Tools / Skills
Skill · .md packagenausedaikestutis-heiyan/10 trustMetrics-checked
paper2code logo

paper2code

Turn any arXiv paper into a citation-anchored working implementation

Visit official site →

About

A skill that converts a research paper URL into a working Python implementation where every code decision cites the exact paper section and equation behind it, and every gap is flagged as SPECIFIED, PARTIALLY_SPECIFIED, or UNSPECIFIED instead of being silently filled in. It produces a full project layout — model, loss, training loop, configs, and a walkthrough notebook — plus an ambiguity audit in REPRODUCTION_NOTES.md. It is honest about its limits: it will not download datasets, set up training infrastructure, or guarantee the paper itself is correct.

What it can do

  • Convert arXiv papers into citation-anchored implementations
  • Classify ambiguity per choice
  • Generate model, loss, training, data, and evaluation code
  • Produce a pedagogical walkthrough notebook
  • Flag unspecified hyperparameters with [UNSPECIFIED]

Setup

  1. Install with the skills CLI:

npx skills add PrathamLearnsToCode/paper2code/skills/paper2code

  1. Choose your agents, scope (global recommended), and method (symlink recommended).
  2. Launch your agent (e.g. claude) and run:

/paper2code https://arxiv.org/abs/1706.03762

Prerequisites

A coding agent that reads SKILL.md files; a Python environment for running the output.

Use cases

reproducing an ML paper's architecture in PyTorch or JAXauditing which hyperparameters a paper actually specifiesteaching a model architecture with a cited walkthrough notebook

Works with

Claude CodeAny SKILL.md-compatible agent

Trust score — 5/10

/10 trustMetrics-checked

Scored 5/10 (metrics tier, metrics-v1). Evidence: 0 GitHub stars, 5 commits; fork of PrathamLearnsToCode/paper2code.; README install command points at PrathamLearnsToCode/paper2code, not this fork; flagged in setup.; Worked examples (Attention Is All You Need, DDPM) with honest review.md included; design principles documented.; MIT license confirmed on GitHub license field.. License: MIT License, verified via GitHub license field..

★ 0 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
1.0/10
Maintenance · recency, releases, issue load
4.0/10
Openness · license + docs
8.0/10
Manual review · sanity read + red-flag check
5.0/10
  • · 0 GitHub stars, 5 commits; fork of PrathamLearnsToCode/paper2code.
  • · README install command points at PrathamLearnsToCode/paper2code, not this fork; flagged in setup.
  • · Worked examples (Attention Is All You Need, DDPM) with honest review.md included; design principles documented.
  • · MIT license confirmed on GitHub license field.

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.