Guide · Agent skill
How to use the Nature Paper Card agent skill: install and run it
Updated · written with AI from the project's own documentation and checked against our catalogue data
nature-paper-card is an agent skill that turns one scientific paper into a structured deep-reading Paper Card with 16 numbered sections. It is aimed at researchers, graduate students and scientists who want to check whether a paper's claims are backed by its experiments, not just read a summary. After this guide you will know what the skill does, how to install it, how to phrase a request that triggers it, and what it needs to run.
Nature Paper CardYuan1z0825/nature-skills · Apache-2.0 Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it. Keep the folder name nature-paper-card.
- 2Place the folder. Put it in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Check Python and the paper. Make sure Python is available and have the paper ready as a PDF or nature-reader source-map JSON.
- 4Make a natural request. Ask for a Paper Card for the paper, naming the skill and the parts you care about, such as the evidence chain or limitations.
- 5Review the card. Read paper-card.md, check the Not assessable sections, and review audit-report.json for errors and warnings.
What is an agent skill, and what does this one do?
An agent skill is a folder of instructions and optional scripts that an AI agent loads when a task matches the skill's description. This one is called nature-paper-card. Its description says it builds a structured deep-reading Paper Card for one scientific paper, covering methods, experiment-to-claim evidence, limitations and research ideas.
Instead of a plain summary, the agent produces a card that breaks down the methods, links each central claim to the experiments or figures that support it, lists limitations (both those the authors state and those the agent identifies), and proposes research ideas framed as hypotheses to test. The skill runs bundled scripts to prepare the paper text and to audit the finished card against the source for groundedness.
- Produces a fixed 16-section card (Sections 01 to 16) for a single paper.
- Separates author statements from the agent's own analysis.
- Marks sections as not assessable when the supplied material cannot support them.
- Does not do full bilingual translation or formal peer review; the description excludes both.
Install the skill from the ZIP file
Download the ZIP of the skill from this page and unzip it. For a personal install, place the unzipped skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead.
If you use the Claude apps rather than a local folder, you can upload the skill instead. Open Settings, go to Capabilities, then Skills, and upload it there.
- Personal install: ~/.claude/skills/nature-paper-card/
- Project install: .claude/skills/nature-paper-card/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the skill
Check what you need before you start
The skill needs Python, because its bundled scripts are run with python. You also need the paper itself as a PDF or as a nature-reader source-map JSON file, or equivalent text. If page extraction fails, the skill can use environment PDF or OCR capability if your setup has it.
The skill runs scripts and does not use the network, according to its listing. Its quality depends on the input: if you only supply an abstract or metadata, many sections will be marked not assessable.
- Python installed and available on the command line
- The paper as a PDF or nature-reader source-map JSON, or equivalent text
- Optional: PDF or OCR capability if page extraction fails
Trigger the skill with a natural request
You do not need a special command. Describe the task in plain language and attach or point to the paper. The skill's description lists Chinese trigger phrases such as 论文精读卡, 方法拆解 and 证据链分析, and the README gives English examples of requests that work.
Name the skill in your request if you want to be sure it is used. Say what you want checked, such as the methods, the evidence chain, or the limits of the claims.
- "Use nature-paper-card to deep-read this PDF and generate a complete Paper Card."
- "Analyze the method modules, essential equations, and experiment-to-claim evidence of this paper."
- "What does this paper actually prove, and what does it not prove?"
- "Based on the limitations in the paper, propose a few follow-up research directions I can test."
Read the output and the audit
The main deliverable is a Markdown file, paper-card.md by default, with all 16 sections in order. Where the material is missing, the card says Not applicable or Not assessable rather than inventing content. The skill also writes a source bundle, source_bundle.json, and an audit report, audit-report.json, when it runs its scripts.
The audit script checks the finished card against the source. Treat its errors as blockers that must be fixed before you rely on the card. Warnings are for your own scientific judgement.
- paper-card.md: the 16-section deep-reading card
- source_bundle.json: the normalised source package from the paper
- audit-report.json: the groundedness audit result
Safety notes on scripts and network use
This skill runs bundled scripts: prepare_paper.py to prepare the paper, and audit_paper_card.py to audit the card. Run them as the skill directs and do not expect them to fetch anything online, since the listing states the skill does not use the network.
The skill's own rules forbid the agent from writing replacement scripts, such as an extra PDF extractor, during a normal run, or from patching the bundled scripts unless you explicitly ask to develop the skill. Keep it that way by reviewing the files before you install if you work with confidential papers.
- Scripts are run with python from the skill folder.
- The skill is documented as not using the network.
- Confidential manuscripts stay local under the skill's documented workflow.
What people use it for
- Checking whether a paper's central claim is supported by its ablations and baseline comparisons before citing it.
- Preparing journal-club or literature-note material from one paper, organised into a consistent structure.
- Finding follow-up research questions from a paper's stated limitations, framed as hypotheses to test.
- Separating what the authors state from what the agent infers when reviewing a clinical or methods paper.
Frequently asked questions
Is nature-paper-card free to use?
The skill is licensed under Apache-2.0, as listed on this page. Check the licence file in the downloaded ZIP for the full terms before redistributing it.
How do I install the nature-paper-card skill?
Download the ZIP, unzip it, and place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What does the nature-paper-card skill require?
It requires Python, because its bundled scripts run with python, and a PDF or nature-reader source-map JSON of the paper or equivalent text. PDF or OCR capability is optional and only needed if page extraction fails.
Can nature-paper-card translate a whole paper?
No. The skill is scoped to deep reading and explicitly excludes full-paper bilingual translation. Use the nature-reader skill for full-text bilingual reading artifacts, as the skill's own notes suggest.
Can it do peer review of my manuscript?
No. Formal peer review and publication-ready articles are outside its scope. It analyses an existing paper's evidence and limitations, which is a different job.
Why are some sections marked Not assessable?
If you supply only an abstract or metadata, the skill cannot verify experiments or methods, so it marks those sections Not assessable instead of guessing. Supplying the full paper gives a complete card.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use nature-paper-card if you need to check whether a single paper's claims are backed by its experiments, or want a consistent deep-reading card for journal club or literature notes. Pick something else if you need a full bilingual translation, a formal peer review, or a card built from an abstract alone.
Download the skill