How to use the Nature Paper X-Ray agent skill

What is an agent skill, and what does this one make the agent do?

An agent skill is a set of instructions and reference files that an AI agent loads when a request matches the skill's description. You do not run it like a normal program; you ask for the task in plain language and the agent uses the skill's method.

The nature-paper-xray skill is for one supplied paper at a time. Its stated goal is to reconstruct how the authors got to their result rather than restate what the abstract claims. The agent is asked to read the body, appendix, footnotes, captions and supplementary material if you supply them, then produce a long-read that leads with what the paper actually claims, followed by the research path, the load-bearing design judgments, worked examples and the evidence boundary.

• Recovers the concrete failure the work reacts to and the bet the authors placed on their strongest evidence.
• Separates design choices that the result depends on from ones that merely accompany it, citing an ablation or controlled comparison where one exists.
• Gives each key formula its symbols, a purpose and a hand-computable example.
• Flags what the paper does not report instead of filling gaps with plausible numbers.
• Explicitly not a summary, not a translation and not a peer review.

Install nature-paper-xray: ZIP download or Claude app upload

Download the ZIP of the skill from our page. The skill folder is nature-paper-xray, taken from skills/nature-paper-xray in the Yuan1z0825/nature-skills repository, and it is licensed under Apache-2.0.

For a personal install across projects, unzip the folder into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/nature-paper-xray/
• Project install: .claude/skills/nature-paper-xray/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the skill

Check the folder contents after installing

After unzipping, the skill folder should contain SKILL.md, which holds the instructions the agent reads, plus two reference files in references/: deep-reading-protocol.md and evidence-boundary.md. There are also agents/openai.yaml, manifest.yaml and README files in English and Chinese.

The SKILL.md file is the core. If the folder is missing it or its reference files, the skill will not behave as described.

• SKILL.md
• references/deep-reading-protocol.md
• references/evidence-boundary.md
• agents/openai.yaml
• manifest.yaml
• README_EN.md and README.md

How do you trigger the skill with a natural request?

You do not need a special command. The skill's description lists trigger phrases, including Chinese ones such as 精读论文 and 讲透这篇论文, and English phrases such as paper x-ray and deep single-paper comprehension. Ask for a close reading of a paper you supply and the agent should pick the skill up.

Give the agent the paper itself (PDF, arXiv link, publisher HTML page or pasted text). Adding the appendix or supplementary material helps, because the skill notes that appendices often contain ablations and real search ranges that the main text leaves out. Say what you need the reading for, since that changes the level of detail.

• "Read this paper closely and tell me why the authors chose this design."
• "Walk me through this formula with a small example I can compute by hand."
• "What does the first figure want me to believe, and does it hold up?"
• "Were the baselines tuned as hard as the proposed method?"
• "Which parts of this paper are load-bearing, and which can I remove without changing the conclusion?"

What to expect from the output

The output is a Markdown long-read in the order the skill defines: what the paper claims, the reconstructed path, the load-bearing judgment, worked formula examples, the evidence boundary, and what to carry away and what remains open. The agent is instructed to state its scope, such as body only, appendix included or a stated subset.

The skill is designed to answer in the language you use. Formulas are written with $...$ and $$...$$ rather than in code blocks.

• Expect a long document, not a few paragraphs.
• Expect explicit statements such as "the paper does not report this" where evidence is missing.
• If the reading is incomplete, the agent should say where it stopped.

Safety notes: scripts and network access

The skill's listing states that it runs no scripts and uses no network access, and that it requires nothing listed. Its files are instructions and reference documents. Still, review any skill you install before trusting it with sensitive papers, and keep unpublished manuscripts in mind when you choose what to upload.

• Runs scripts: no
• Uses network: no
• Requires: nothing listed

Tips for getting a useful reading

Supply the full paper and its appendix. Name your purpose, such as presenting, citing or reviewing. If the paper has an official code repository, you can provide it: the skill's own material notes that code can reveal normalization, warm-up or data filtering details the paper omits.

Know the boundaries. The skill sets no verdict on publishability or novelty, does not search literature and does not verify citations. Those tasks belong to other skills the file names, such as nature-reviewer, nature-academic-search, nature-citation and nature-ref-verifier.

• Provide the appendix and supplementary material when they exist.
• State the purpose of the reading to set the detail level.
• Add the official code repository if one exists.
• Use nature-reviewer for judgments on publishability or novelty.
• Use nature-academic-search or nature-citation for literature and citation work.