大家好，我是nature skills的创立者袁一哲，从github以及其他途径联系我的顶尖AI人才数不胜数，所以我成立了TOP AI CREW！今天起人才联盟正式招募！ 这里汇聚经过严格筛选的行业强者，多元思维碰撞，前沿技术共生。 拒绝单打独斗，告别低效内耗，和一群同频的顶尖伙伴并肩前行，深耕AI领域，突破技术边界，一同站上行业前沿。 敢想、敢闯、敢创造，下一个AI传奇，由我们共同书写！

感谢大家持续关注 nature-skill。如果你有任何需求，欢迎提交 issue；如果我们认为该需求有意义且可行，也会尽量推进实现。我们同样欢迎 PR，但请务必按照 README 后面说明的格式提交，以便我们更高效地审核与合并。

个人微信

个人简介

Hello everyone, I’m Yizhe Yuan, founder of Nature Skills. After being contacted by countless top AI talents through GitHub and other channels, I decided to launch TOP AI CREW. Starting today, our talent alliance is officially open for recruitment. This is a community of carefully selected industry leaders: a place where diverse perspectives collide, frontier technologies evolve, and ambitious builders grow together. No more working in isolation. No more wasting energy on inefficient solo battles. Here, you’ll move forward alongside a group of world-class, like-minded peers—deepening your expertise in AI, pushing technical boundaries, and advancing to the forefront of the industry together. Think boldly. Move fearlessly. Create relentlessly. The next AI legend will be written by us—together.

Thank you for your continued interest in nature-skill. If you have any feature requests or suggestions, please feel free to submit an Issue. If we find the proposal meaningful and feasible, we will do our best to implement it. We also welcome Pull Requests (PRs). However, please follow the contribution format described later in this README to help us review and merge submissions more efficiently.

微信群聊 答疑交流群！进群请立刻改备注

还在寻找能够落地的 AI 前沿交叉赛道吗？我们课题组现向对“医学 + AI”充满热情的你发出邀请！ 这里有充足的计算资源，以及深耕医疗大模型（LLM）、视觉预训练、Prompt Engineering 及自动化医疗 AI Agent 的科研团队。我们更看重你的自驱力、学习能力与科研产出追求。 项目信息文档链接：https://iigqjt2m4ia.feishu.cn/wiki/VIvDwHu18iTc6mk411xco8chnJb 密码：664#N926 如果你有相关代码基础或项目经验，渴望在顶级交叉学科中积累成果，请将简历发送至： 📧 sjtu520aimedws@163.com （标题格式：姓名-专业-医学AI科研申请） 期待与你在 AI 赋能医疗的征途中，做出最扎实的科研工作！

nature-skills is a repository of reusable instruction bundles centred on SKILL.md. Each skills/nature-* directory is one installable unit. Copy the whole folder, not only SKILL.md, because many skills depend on references/, static/, assets, scripts, or README context. The skills/_shared/ directory is shared support content used by several skills and should stay next to the nature-* folders when you install skills manually.

Codex plugin marketplace installation

This repository includes Codex plugin packaging at plugins/nature-skills/, so Codex users can install the complete Nature Skills bundle from the plugin marketplace instead of copying each skill folder manually.

CLI installation:

codex plugin marketplace add https://github.com/Yuan1z0825/nature-skills --ref main
codex plugin add nature-skills@nature-skills

Codex Desktop users can add the same repository as a custom plugin marketplace:

• Marketplace source: https://github.com/Yuan1z0825/nature-skills.git
• Branch/ref: main
• Plugin: nature-skills

After installation, all nature-* skills are available through the plugin as a complete bundle, together with the shared support directory used by the newer router-style skills. If the skills do not appear immediately, refresh the plugin page or start a new Codex session.

Manual local-skill installation

Codex can also use these folders directly as local skills.

Clone the repo

git clone https://github.com/Yuan1z0825/nature-skills.git
cd nature-skills

Install one skill

mkdir -p ~/.codex/skills
cp -R skills/_shared ~/.codex/skills/
cp -R skills/nature-reader ~/.codex/skills/

Copying _shared is harmless even for skills that do not use it, and it avoids broken relative references for skills such as nature-reader, nature-writing, nature-polishing, and nature-paper2ppt.

Install all current skills

mkdir -p ~/.codex/skills
cp -R skills/_shared ~/.codex/skills/
for d in skills/nature-*; do
  cp -R "$d" ~/.codex/skills/
done

Update after pulling new changes

git pull
cp -R skills/_shared ~/.codex/skills/
for d in skills/nature-*; do
  cp -R "$d" ~/.codex/skills/
done

Finish

• Restart Codex so newly added skills are picked up.
• Then ask naturally, for example: Translate this paper into a full markdown reader. or Make this paper into a Chinese journal-club PPT.

If you prefer not to use the terminal, copying the skills/nature-* folder(s) into ~/.codex/skills/ manually works as well; also copy skills/_shared/ once. For a longer walkthrough, see install.md.

Claude Code plugin marketplace installation

This repository also includes Claude Code plugin metadata at .claude-plugin/, so Claude Code users can install the complete Nature Skills bundle directly as a plugin instead of setting up wrappers first.

CLI installation:

claude plugin marketplace add Yuan1z0825/nature-skills
claude plugin install nature-skills@nature-skills

If the plugin does not appear immediately, refresh the plugin page or start a new Claude Code session.

Alternative: wrapper/subagent installation

Claude Code can also use a thin subagent or slash-command wrapper that points to a stable clone of this repository, so supporting files such as references/, static/, assets, scripts, and skills/_shared/ remain available.

mkdir -p ~/ai-skills
cd ~/ai-skills
git clone https://github.com/Yuan1z0825/nature-skills.git

Create a user-level subagent wrapper:

mkdir -p ~/.claude/agents
cat > ~/.claude/agents/nature-reader.md <<'EOF'
---
name: nature-reader
description: Full-paper bilingual, figure-aware, source-grounded Markdown reader for journal or conference papers. Use proactively when the user asks to translate an entire paper or generate a complete markdown reader.
---

When invoked, first read `~/ai-skills/nature-skills/skills/nature-reader/SKILL.md`.
Treat that file as the governing workflow.
If the skill references supporting files, read only the specific files you need from
`~/ai-skills/nature-skills/skills/nature-reader/` and
`~/ai-skills/nature-skills/skills/_shared/`.
Do not replace the skill with a generic paper-summary response.
EOF

After that, start a new Claude Code session or open /agents, and invoke it naturally or explicitly:

Use the nature-reader subagent to turn this PDF into a full markdown reader.

If you prefer commands instead of subagents, create a project or user command under .claude/commands/ or ~/.claude/commands/ that tells Claude Code to read the real SKILL.md from the cloned repository.

Official Claude Code docs:

• Subagents
• Slash commands

If your agent supports reusable prompt files, system prompts, or agent profiles, the minimum portable unit is the skill directory itself:

skills/
├── _shared/              # keep this when a skill references ../_shared
└── nature-<topic>/
    ├── README.md
    ├── SKILL.md
    ├── manifest.yaml     # present for router-style skills
    ├── static/           # present for router-style skills
    └── references/...

In that case:

1. Copy the whole skill directory into your prompt library or project.
2. Preserve SKILL.md, manifest.yaml, static/, references/, scripts, assets, and any needed skills/_shared/ files together.
3. Adapt the frontmatter and body to the target agent's native format if needed.

Adding a new skill? Follow the contribution guide at the bottom of this file.

What it does — Generates multi-panel matplotlib figures that match Nature journal visual standards: correct typography, semantic colour palette, editable SVG output, and non-redundant panel information architecture.

Example output gallery — Five dense, simulated Nature-style result figures are included in the nature-figure gallery: material/mechanism, spatial imaging, in vivo efficacy, single-cell systems and perturbation validation.

Chart-type atlas — The nature-figure chart atlas classifies 10 supported chart families, including bar, line, heatmap, scatter/bubble, radar/polar, distribution, forest/interval, area/stacked, image-plate and network/matrix layouts.

Built from — Production scripts from papers published in Nature Machine Intelligence and top ML/bioinformatics venues (figures4papers). The figures4papers demo scripts and preview assets are bundled inside skills/nature-figure/assets/figures4papers/, with a routing guide at skills/nature-figure/references/demos.md.

Key rules enforced

• Three mandatory rcParams must always appear first:

  plt.rcParams['font.family'] = 'sans-serif'
  plt.rcParams['font.sans-serif'] = ['Arial', 'DejaVu Sans', 'Liberation Sans']
  plt.rcParams['svg.fonttype'] = 'none'   # text stays as <text> nodes, not paths

• Primary output is always .svg; .png at 300 dpi is a secondary raster preview.
• Multi-panel figures follow a three-level information hierarchy: overview → deviation → relationship. No two panels may answer the same scientific question.

Reference files

skills/nature-figure/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
└── references/
    ├── api.md            PALETTE, helper signatures, validation rules
    ├── design-theory.md  Typography, layout, export policy, anti-redundancy rules
    ├── common-patterns.md Ultra-wide panels, legend axes, print-safe bars
    ├── tutorials.md      End-to-end walkthroughs (bars, trends, heatmaps)
    ├── chart-types.md    Radar, 3D sphere, scatter, fill_between, log-scale
    └── demos.md          Bundled figures4papers scripts and preview routing

Supported chart types — Stacked bar, grouped bar, horizontal ablation bar, trend/line, sequential heatmap, diverging z-score heatmap, bubble scatter, radar/polar, 3D sphere illustration, fill-between area, log-scale bar, GridSpec multi-panel.

What it does — Transforms academic draft text (including Chinese → English translation) into prose matching Nature journal conventions: ≤ 30-word sentences, section-aware tense and hedging, precise vocabulary, correct citation practice, and British English.

Built from — A graduate-level scientific English writing course, Academic Phrasebank, and close reading of curated Nature and Nature Communications research articles across materials, energy systems, construction decarbonization and machine learning.

Key rules enforced

12-step polishing workflow

Sentence split → Section ID → Hourglass check → Tense audit → Sentence edit → Vocabulary upgrade → Template check → Citation audit → House style → Overclaim → Proofreading → Plain-text output

Reference files

skills/nature-polishing/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
└── references/
    ├── latex-layout.md
    ├── published-article-patterns.md
    ├── phrasebank-playbook.md
    ├── section-moves.md
    ├── style-guardrails.md
    └── writing-strategy.md

What it does — Drafts or rebuilds manuscript sections from author-provided claims, results, figures, notes, or Chinese drafts. It is for argument construction: abstracts, introductions, Results narratives, Discussions, Conclusions, titles and full manuscript outlines, method sections, experiment sections and reviewer-facing self-review.

Built from — Close reading of curated Nature and Nature Communications articles, especially how published papers move from field-scale stakes to a narrow gap, then to evidence, interpretation and bounded implication. It also integrates open research-writing notes for paragraph flow, section logic and adversarial paper review.

Key rules enforced

Reference files

skills/nature-writing/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
├── agents/
│   └── openai.yaml
└── references/
    ├── abstract.md
    ├── article-architecture.md
    ├── chinese-author-workflow.md
    ├── conclusion.md
    ├── experiments.md
    ├── introduction.md
    ├── method.md
    ├── paper-review.md
    ├── paragraph-flow.md
    ├── related-work.md
    └── examples/

What it does — Simulates a Nature-style pre-submission reviewer assessment from the referee perspective. It returns three reviewer reports plus a cross-review synthesis, focusing on novelty, significance, technical soundness, presentation, and likely editorial risk.

Key rules enforced

Reference files

skills/nature-reviewer/
├── README.md
├── SKILL.md
└── references/
    ├── source-basis.md
    ├── reviewer-workflow.md
    ├── review-axes.md
    ├── report-structure.md
    ├── role-boundaries.md
    └── qa-checklist.md

What it does — Converts manuscript text or standalone claims into strict Nature / CNS-family citation candidates, then exports one reference-manager-ready file in ENW, RIS, or Zotero RDF. It can also generate an HTML screening page for year filtering, citation selection, and format-specific download.

Built from — Crossref metadata retrieval, DOI record export, and journal-family filtering logic for Nature Portfolio, the AAAS Science family, and Cell Press.

Key rules enforced

Reference files

skills/nature-citation/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
├── references/
│   ├── journal-scope.md
│   ├── ris-endnote.md
│   ├── script-usage.md
│   └── search-strategy.md
└── scripts/
    └── nature_citation.py

Example workflow — Segment a paragraph, search in-scope citations, review candidates in the HTML browser, then download only the selected records as ENW, RIS, or Zotero RDF.

What it does — Prepares and audits Data Availability statements, repository plans, dataset citations, and FAIR metadata checks for Nature-family and Springer Nature submissions. It is bilingual-aware: Chinese author notes such as "data availability statement", "request from corresponding author", "raw data", "restricted data", and "public database" are converted into precise submission-ready English with Chinese action notes.

Built from — Springer Nature research data policy, Nature Portfolio reporting standards, Scientific Data repository and citation practice, the FAIR Guiding Principles, and DataCite metadata conventions.

Key rules enforced

Reference files

skills/nature-data/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
├── agents/
│   └── openai.yaml
└── references/
    ├── chinese-author-alignment.md
    ├── fair-metadata-checklist.md
    ├── policy-principles.md
    ├── repository-and-identifiers.md
    ├── source-basis.md
    └── statement-patterns.md

What it does — Drafts, audits, and revises point-by-point reviewer response letters for Nature-family and high-impact journal manuscript revisions. It treats the response letter as an editor-facing verification document: every reviewer concern is assigned a stable ID, classified, mapped to an action, and tied to manuscript evidence, a revision location, or an unresolved author-input flag.

Built from — Nature editorial process guidance, Nature-family revision-package instructions, Springer Nature rebuttal advice, and transparent peer-review considerations.

Key rules enforced

Reference files

skills/nature-response/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
├── references/
│   ├── action-mapping.md
│   ├── chinese-author-alignment.md
│   ├── comment-taxonomy.md
│   ├── difficult-cases.md
│   ├── intake-and-routing.md
│   ├── qa-checklist.md
│   ├── response-structure.md
│   ├── source-basis.md
│   └── tone-and-stance.md
├── tests/
    ├── conflicting-reviewers.md
    ├── defensive-draft-audit.md
    ├── evaluation-summary.md
    ├── impossible-experiment.md
    ├── major-revision-missing-evidence.md
    ├── minor-revision.md
    └── rubric.md
└── examples/
    ├── conflicting-reviewers.md
    ├── major-revision-with-missing-evidence.md
    └── minor-revision.md

What it does — Turns a scientific paper, preprint, PDF, article text, abstract, figure legends, or reading notes into a concise Chinese .pptx presentation for journal club, group meeting, lab meeting, paper sharing, or thesis seminar.

The skill identifies the paper type and central argument, selects only figures and tables that support the evidence chain, writes Chinese slide titles, bullets, captions, takeaways and speaker notes, creates the actual PPTX deck, and runs lightweight package QA.

Key rules enforced

Reference files

skills/nature-paper2ppt/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
└── references/
    ├── design-and-layout.md
    ├── figure-assets.md
    └── self-review.md

What it does — Provides a multi-source academic search and reference-management workflow backed by a local MCP server. It searches PubMed, CrossRef and arXiv in parallel, fetches records by DOI, PMID or arXiv ID, formats citations, looks up MeSH terms, verifies bibliographic identifiers, and supports .nbib, .ris, .bib and .enw reference-file workflows.

Built from — A unified MCP server with source adapters for PubMed E-utilities, CrossRef REST metadata and arXiv Atom metadata, plus reusable workflow notes for source-tier routing, search strategy, citation parsing, deduplication, RIS/BibTeX field mapping and reference-file conversion.

Setup note — For Claude Code MCP use, run bash skills/nature-academic-search/install.sh your-email@example.com, restart Claude Code, and optionally set NCBI_API_KEY for higher PubMed rate limits. For plain prompt use, copy the whole skills/nature-academic-search/ directory like the other skills.

Key rules enforced

MCP tools

Reference files

skills/nature-academic-search/
├── README.md
├── SKILL.md
├── manifest.yaml
├── static/
├── install.sh
├── config/
│   ├── mcp-snippet.json
│   ├── settings-snippet.json
│   └── triggers-academic-search.toml
├── mcp-server/
│   ├── academic_search_server.py
│   ├── sources/
│   ├── tests/
│   └── utils/
├── references/
│   ├── citation-parser.md
│   ├── dedup-engine.md
│   ├── ris-bibtex-format.md
│   ├── search-strategy.md
│   ├── source-tiers.md
│   └── workflows/
└── scripts/
    ├── converters.py
    ├── format-converter.py
    └── preflight.py

Example workflow — Search the same topic across PubMed, CrossRef and arXiv, merge and deduplicate candidate papers, verify key identifiers, look up MeSH terms for the biomedical subset, then export or convert the selected references for Zotero, EndNote or BibTeX.

All skills in this collection adhere to the following:

1. Primary sources only — rules are grounded in published Nature content or official journal guidelines, not general style preference.
2. Explicit over implicit — every rule is stated with a rationale, not just asserted.
3. Section-aware — academic writing and figures both require context-sensitivity; each skill applies different logic depending on which part of a paper is being handled.
4. Output-first — every skill returns something immediately usable: copy-paste prose, a .svg file, a .pptx deck, or a concrete recommendation. No intermediate planning documents.
5. Extensible by design — each skill is self-contained in its own directory; adding a new skill requires no changes to existing ones.

To add a skill to this collection:

1. Create a directory

skills/nature-<topic>/

2. Minimum required files

3. SKILL.md frontmatter template

---
name: nature-<topic>
description: >-
  One-sentence description of what the skill does and when to trigger it.
  Include the output format and the primary use case.
---

4. Update this index

Add a row to the Skill index table above:

| [`nature-<topic>`](skills/nature-<topic>/README.md) | Draft / Stable | One-line purpose | trigger keywords |

5. Status labels

The following are documented gaps. Contributions welcome.

微信赞赏	来自粉丝的小感动

项目会长期免费维护，大家正常使用就好。
很多朋友还是学生党，所以完全不需要有任何“必须打赏”的压力。

如果这些内容刚好帮你节省了一些时间、解决了一点问题，
又恰好想请作者喝杯咖啡，那我会非常开心 ☕

你的支持会用于：

• 持续更新内容
• 服务器与工具费用
• 熬夜写文档时的续命奶茶（认真）

无论是否赞赏，都非常感谢你的关注与支持 ❤️