paper2code skill
An agent skill that turns an arxiv paper into a minimal Python implementation, with code anchored to sections and equations and gaps flagged honestly
- 1,542 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Installs packages via pip and downloads papers from the network
- Creates code files and artifacts in the project working directory
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/paper2codeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Installs packages via pip and downloads papers from the network
- Creates code files and artifacts in the project working directory
This is third-party code. Review the repository files before installing.
What it does
The skill walks the agent through a pipeline: it downloads and parses the arxiv paper, identifies the single core contribution, runs an ambiguity audit and only then writes code. Every non-trivial decision carries a reference to the paper's section and equation, and anything the paper does not state is marked with tags such as UNSPECIFIED and PARTIALLY_SPECIFIED plus a list of reasonable alternatives. The output is a project layout with the model, loss functions, a hyperparameter config, reproduction notes and a walkthrough notebook. It offers minimal, full and educational modes and a choice of the pytorch, jax or numpy framework. The skill never fills gaps silently and does not guarantee the paper itself is right.
Who it is for. For developers and researchers who implement machine learning papers and want verifiable code.
Good fit when
- You need a working implementation skeleton from an arxiv link or ID
- You need to see which details come from the paper and which were chosen for it
- You want a teaching walkthrough of an architecture with code anchored to paper sections
Not a fit when
- You need a full production training pipeline with tracking and distributed runs
- You need a correctness guarantee rather than an honest implementation of what the paper describes
Example request
Implement the paper https://arxiv.org/abs/1706.03762 with code anchored to sectionsLimitations
The skill does not guarantee correctness: if the paper is incomplete or wrong, the code reflects that, though it is flagged. It does not download datasets; it provides a Dataset class skeleton with instructions. It does not set up training infrastructure, implement baselines or reimplement standard components. For parsing it installs packages via pip and fetches the paper over the network, and result quality depends on the agent's own model.
How to disable. Remove the skill from the agent's skills directory. If installed via symlink, remove the link; if copied, delete the paper2code folder.
FAQ
What happens with unspecified details?
The skill uses a reasonable default and marks the line with an UNSPECIFIED tag plus alternatives so the choice can be checked against the paper.
What input formats does it accept?
A full https://arxiv.org/abs/XXXX.XXXXX link, a bare ID and a versioned ID, plus mode and framework flags.
Discussion
No questions or reviews yet.
Related
SkillsA skill that finds AI papers from 28 labs on arXiv over any date range and splits led papers from co-authored ones.
Codex skill that picks fresh research papers for a researcher profile and builds a scored summary in Chinese.
A three-stage workflow for math modeling tasks: analysis and model choice, Python or MATLAB computation, and paper assembly in Word or LaTeX
Probabl skills for machine learning experiments: a scikit-learn and skrub pipeline, evaluation, audit and a link to Skore Hub
A template for data scientists: AGENTS.md, CLAUDE.md and skills for Python, SQL, Polars and model diagnostics.
A pipeline that picks remote sensing papers from arXiv daily, reviews each in a GitHub Issue and publishes a daily digest.
README in short
The README frames the problem: machine learning papers often omit hyperparameters, and naive code generation fills gaps silently. paper2code answers this by anchoring every decision to a section and equation, auditing ambiguities before writing code and flagging unspecified choices honestly. Install is via npx skills add with a choice of agents and linking method. The output includes model.py, loss.py, a hyperparameter config, a reproduction notes file and a notebook runnable on CPU with toy dimensions. The repository ships worked examples for the Attention Is All You Need and DDPM papers. MIT licensed.
README badge
Original SKILL.md
---
name: paper2code
description: Converts an arxiv paper into a minimal, citation-anchored Python implementation. Trigger when user runs /paper2code with an arxiv URL or paper ID, says "implement this paper", or pastes an arxiv link asking for implementation. Flags all ambiguities honestly. Never invents implementation details not stated in the paper.
---
# paper2code — Orchestration
You are executing the paper2code skill. This file governs the high-level flow. Each stage dispatches to a detailed reasoning protocol in `pipeline/`. Do NOT skip stages. Do NOT combine stages. Execute them in order.
## Parse arguments
Extract from the user's input:
- `ARXIV_ID`: the arxiv paper ID (e.g., `2106.09685`). Strip any URL prefix.
- `MODE`: one of `minimal` (default), `full`, `educational`.
- `FRAMEWORK`: one of `pytorch` (default), `jax`, `numpy`.
If the user provided a full URL like `https://arxiv.org/abs/2106.09685`, extract the ID `2106.09685`.
If the user provided a versioned ID like `2106.09685v2`, keep the version.
## Set up working directory
Create a temporary working directory: `.paper2code_work/{ARXIV_ID}/`
This is where intermediate artifacts go. The final output goes in the current directory under `{paper_slug}/`.