nature-reader Skill for Claude Code: Bilingual Paper Reader

Turns a research paper into a Chinese-English Markdown reader with figures, tables, equations and a source map back to the original.

Installs: 15.4K; on skills.sh
Stars: 51.7K; on GitHub
Files: 19; besides SKILL.md
Updated: Nov 2026; last repo commit

nature-reader makes Claude build a full-paper reader from a PDF, a scanned PDF, publisher HTML, a DOI or arXiv link, or pasted text. Each paragraph is paired with its Chinese translation, figures and tables sit near their first mention, and every block gets an ID such as S001, F001 or E001.

It needs the separate nature-shared folder from the same repo, which it reads for its terminology ledger, so install both. The output is paper.md, source_map.json, translation_notes.md and an assets folder. Before it finishes, the skill runs a bundled Python script that checks equations and links.

Good for

Researchers who read papers in both Chinese and English
Students who want a full translation, not a summary
Anyone who needs answers traced back to page and paragraph

Install nature-reader

Claude Code

Copy this prompt.

Install the nature-reader skill in Claude Code. Run this command:

npx skills add Yuan1z0825/nature-skills --skill nature-reader --skill nature-shared

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js 18+.

Choose both skills when the CLI asks. nature-reader reads ../nature-shared, so the two folders must sit next to each other. The bundled check needs Python 3.

From the project README. Source.

What's in the skill

When it loads

Create source-grounded Chinese-English paper readers with aligned text, figures, tables, and equations. Use for 全文翻译、中英文对照、论文精读 or source-linked questions about a paper; respect a requested excerpt or question without generating a full reader.

Sections in SKILL.md

Routing protocol
1. Load the manifest and the core layer
2. Detect the source format
3. Load the matching fragment(s)
4. Build the reader using the loaded material
5. Reach for references only when needed

Key facts

Author: yuan1z0825 (repo owner)
Repo: yuan1z0825/nature-skills
Installs on skills.sh: 15,404
Repo stars: 51,696
License: Apache-2.0
Last repo commit: 2026-11-01
Files in the skill folder: 20 files including SKILL.md
Skill version: 2.1.1
Also needs: nature-shared (support folder in the same repo)
Bundled script: scripts/validate_reader_math.py (Python)

What it includes

Tells a reader request apart from a question or a short excerpt
For a source-linked question it reads only the relevant source. For an excerpt it translates just that excerpt.
Route the request
Picks pdf-text, scanned-pdf, html, doi-arxiv or pasted-text
States the detected format in one line so you can correct it, then loads only the matching instructions.
Network · Detect the source format
Gives every block a stable ID
S for body text, C for captions, F for figures, T for tables and E for display equations, each with page, original text, translation and confidence.
Build a source map first
Keeps terms, symbols, citations and numbers intact
Translates every substantive block, keeps paragraph order and marks uncertain text instead of guessing.
Translate conservatively
Crops each figure or table and puts it near its first mention
Crops go into the assets folder with the original and Chinese captions attached.
Files · Place figures and tables
Produces paper.md, source_map.json and translation_notes.md
paper.md has Original and Chinese block pairs, a section index, an equation index and a terminology table.
Files · Write the reader
Renders equations as Markdown math, keeps crops for image-only ones
Every display equation gets an E anchor. Low-confidence formulas show the original crop plus a labelled best-effort transcription.
Files · Equation handling
Checks delimiters, bare LaTeX, equation IDs and source-map links
Run as python scripts/validate_reader_math.py paper.md --source-map source_map.json before delivery. Add --strict for a reusable artifact.
Runs code · scripts/validate_reader_math.py
Answers from the paper with block IDs and page numbers
Reuses IDs from an existing reader, or cites verified locations without inventing IDs.
Answer follow-up questions

From the official docs

Permissions

Low risk

The skill reads the paper you give it and writes a reader plus a source map into a folder, and it runs one bundled Python check. Use a folder you chose and keep the output to your own copy of copyrighted papers.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It writes files into your output folder
The reader creates paper.md, source_map.json, translation_notes.md and an assets folder. Point it at an empty folder you chose. github.com
It runs a bundled Python check
The output contract tells Claude to run scripts/validate_reader_math.py on paper.md and source_map.json before it finishes. The script is in the skill folder, so you can read it first. github.com
Copyright is yours to watch
The skill advises keeping chat replies short for copyrighted PDFs and building the bilingual reader only from your own file or lawful open-access text. Follow the same rule when you share the output. github.com

Technical details

Talks to	Publisher or preprint sites, to resolve a DOI or arXiv link (static/fragments/source/doi-arxiv.md)
Sign-in and keys	The skill files name no account, API key or environment variable, and validate_reader_math.py imports only the Python standard library.

Questions

What does the nature-reader skill do?

It builds a full-paper Markdown reader with the English original and a Chinese translation side by side, figures and tables placed near their first mention, rendered equations and a source map of page and block IDs.

Does it work as a summary tool?

No. The principles say to translate for meaning and never degrade the reader to a summary. If constraints stop full processing, it makes a draft reader and labels what is missing.

Do I need to install anything besides the skill?

Yes. Install nature-shared with it, because nature-reader reads the terminology ledger from that folder. The install box above sets up both. A Claude app upload is not offered here, since the skill uses a relative path into nature-shared and the README documents no upload route.

Can it read scanned PDFs?

Yes. The scanned-pdf instructions OCR every page, record a confidence level per block and keep equation crops instead of guessing. For PDF extraction it also tells Claude to load a separate pdf skill, which this page does not install.

What if I only have a DOI?

It resolves the DOI or arXiv ID to the article first, prefers an open-access copy and notes which version it read. If only the abstract is reachable, it builds a draft and marks the rest as not retrieved.

Is it safe to install?

It runs one bundled Python script and writes files into your output folder. It also looks up the paper online when you give a DOI or arXiv link. Read scripts/validate_reader_math.py first if you want to check it.

What is it related to?

The README lists nature-downloader for lawfully getting the full text first, nature-paper2ppt for slides and nature-citation for claims that need citations. They are optional.

Information

updated Nov 2, 2026

Repository: yuan1z0825/nature-skills
Folder: skills/nature-reader

yuan1z0825/nature-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 7

nature-skills README · github.com
nature-reader manifest.yaml · github.com
nature-reader SKILL.md · github.com
nature-reader README_EN · github.com
workflow.md · github.com
output-contract.md · github.com
doi-arxiv.md · github.com