# arxiv2agent
**English** | [简体中文](README.zh-CN.md)
A few small download tools that let an agent pull any arXiv paper onto disk as **LaTeX source**, so it reads the real paper instead of search snippets or a lossy PDF.
```bash
arxiv2agent find "Attention Is All You Need" # title → arXiv ID (arXiv API)
arxiv2agent fetch 1706.03762 -o papers/ # raw source tree + one flattened .tex
arxiv2agent digest 1706.03762 -o papers/ # structured: paper.json, sections/, tables/, …
```
`fetch` and `digest` take a list of IDs. No LLM inside, no API key; arXiv requests are rate-limited politely and cached.
## What it looks like
```console
$ arxiv2agent find "Attention Is All You Need" -n 3
1706.03762 2017 Attention Is All You Need Ashish Vaswani et al.
2104.04692 2021 Not All Attention Is All You Need Hongqiu Wu et al.
2501.06425 2025 Tensor Product Attention Is All You Need Yifan Zhang et al.
$ arxiv2agent fetch 1706.03762 -o papers/
$ ls papers/1706.03762/source
Figures/ background.tex introduction.tex ms.tex nips_2017.sty results.tex …
```
**"How did they typeset Table 1?"** The agent greps `papers/1706.03762/1706.03762.tex` and gets the real thing: `booktabs`, `\toprule`, a `\rule{0pt}{2.0ex}` strut, the commented-out `\scalebox` the authors tried:
```latex
\begin{tabular}{lccc}
\toprule
Layer Type & Complexity per Layer & Sequential & Maximum Path Length \\
\hline
\rule{0pt}{2.0ex}Self-Attention & $O(n^2 \cdot d)$ & $O(1)$ & $O(1)$ \\
```
**"What does Section 3.2 of this paper actually claim?"** The agent runs `digest`, reads `sections/` or `paper.json`, and answers from the full text, with exact equations (`equations/*.tex`) and resolved BibTeX (`references.json`).
**"Compare the Introductions of these ten papers."** Run one `digest` with ten IDs, then loop over `paper.json` (the schema is the same for every paper).
## Install
```bash
uv tool install git+https://github.com/wuyoscar/arxiv2agent
```
For an agent (Claude Code, Codex, …), tell it:
> Install https://github.com/wuyoscar/arxiv2agent with `uv tool install`, then read its `SKILL.md` and register it as a skill.
[`SKILL.md`](SKILL.md) covers the commands, the digest layout, and when to use which.
## Acknowledgements
The LaTeX download/flatten pipeline is vendored from **arxiv-to-prompt** (see [NOTICE.md](NOTICE.md)). Also inspired by **DeepXiv**.
## License
MIT.