nature-skills：学术论文全流程 18 个 AI Skill，两个月 3.4 万星

by Mycelium Protocol

---

学术论文写作的痛点不是某一个环节，而是整个链条：从文献检索、图件制作、英文润色、统计审查、审稿回复，到最后的专利转化——每个环节都有自己的规范和陷阱，任何一环出问题都可能推迟投稿。

nature-skills 从一个技能出发（nature-polishing），逐渐扩展到 nature-figure，再到今天的 18 个技能，覆盖了学术论文从开始写到发表后的几乎所有操作节点。两个月时间，GitHub 收获 3.4 万星。

GitHub: https://github.com/Yuan1z0825/nature-skills | ⭐ 34,065 | Apache 2.0

---

安装

npx skills add Yuan1z0825/nature-skills

支持：Claude Code、Codex、OpenClaw、OpenCode、Hermes。安装后触发词直接在对话里说即可，不需要单独配置。

---

18 个技能总览

技能	状态	用途	触发词示例
nature-polishing	Stable	学术文本润色/重构/翻译为 Nature 风格英文，扫描术语、单位、数值精度和声称漂移	”Nature style”, “润色”, “论文英文”
nature-figure	Stable	投稿级科研图工作流（Python/R），含 GPT Image 2 论文示意图草稿	”Nature figure”, “投稿级图片”, “scientific figure”
nature-ref-verifier	Stable	参考文献多源交叉验证：逐字段对比作者/标题/年份/卷期/页码	”verify refs”, “校验文献”, “文献验证”
nature-literature-pipeline	Stable	自动化文献发现管线：多源检索、六维评分、精读推送和本地归档	”literature pipeline”, “每日文献”, “文献推送”
nature-citation	Beta	严格限定在 Nature/CNS 系列的支撑文献检索，导出 ENW/RIS/Zotero RDF	”Nature citation”, “CNS citation”, “支撑文献”
nature-reader	Beta	全文 Markdown reader，带来源锚点、图文对应、公式渲染和中英文对照	”nature reader”, “全文 Markdown”, “原文对照”
nature-paper-card	Beta	精读单篇论文，生成有来源约束的 16 节 Paper Card，含证据链和可检验研究想法	”nature paper card”, “论文精读”, “证据链”
nature-response	Beta	解析返修邮件，为互盲审稿人分别生成独立回复 + cover letter + LaTeX 模板	”response to reviewers”, “rebuttal letter”, “返修邮件”
nature-paper2ppt	Beta	从科研论文生成中文 PPTX 文献汇报 deck	”paper PPT”, “journal club”, “论文汇报”
nature-paper-to-patent	Beta	从论文生成有证据约束的中国发明专利草稿，支持专利点挖掘和查新	”paper to patent”, “论文转专利”, “权利要求书”
nature-academic-search	Beta	多源文献检索、引用核验、严格他引审计、文章引用指标表	”search papers”, “查文献”, “verify DOI”
nature-downloader	Beta	通过图书馆资源入口和开放获取路径合法获取学术全文/PDF	”download papers”, “图书馆下载文献”, “CARSI”
nature-writing	Draft	起草 Nature 风格手稿章节，重建论文论证	”Nature writing”, “写摘要”, “写引言”
nature-reviewer	Draft	模拟 Nature 风格预投稿评审，输出三份互盲 reviewer reports	”Nature reviewer”, “预投稿评审”, “reviewer report”
nature-data	Draft	Data Availability statement、数据仓储方案和 FAIR 检查	”Data Availability”, “数据可用性”, “FAIR metadata”
nature-statistics	Draft	审查/改写统计报告，覆盖 p 值、多重比较、效应量、置信区间	”Nature statistics”, “统计审查”, “p value”
nature-experiment-log	Draft	标准化记录实验图片、语音和文字材料，生成 Obsidian 实验日志	”实验日志”, “记录实验”, “experiment log”
nature-proposal-writer	Beta	proposal-first 科研写作状态机，先建立证据/论证/章节契约，再起草文本	”proposal”, “开题报告”, “科研写作 QA”

Stable = 规则已稳定可生产使用；Beta = 功能完整但边界还在打磨；Draft = 可用但规则还在迭代。

---

这套技能怎么用

每个技能都自包含：自己的 SKILL.md（触发后由 Agent 加载）、README（面向人的说明）和可选的 references/ 目录（规则库）。触发方式是说触发词，Agent 加载对应的 SKILL.md 后按规则执行，返回直接可用的产物——可粘贴文本、.svg、.pptx、.docx，不是建议列表。

五条共同设计原则：

1. 优先使用一手来源：规则基于已发表 Nature 内容、官方期刊指南或明确的本地来源，不是泛泛审美偏好
2. 显式胜过隐式：每条规则都说明理由，而不是只给断言
3. 感知章节与任务上下文：写作、图件、引用和回复依赖论文的不同位置
4. 输出优先：每个技能返回能直接使用的产物
5. 可扩展：每个技能自包含，新增不影响既有技能

---

从 nature-polishing 到 18 个技能

这套合集从单个润色技能出发，逐步扩展到图件制作，再到今天覆盖论文全链路的 18 个工具。整个过程是开源社区协作的结果——从 DeepMind 的 Science Skills 得到启发，发展出一套完整的 Nature 级论文辅助体系。

---

Mycelium Protocol — 追踪 AI 系统的底层演化

---

关于 Mycelium

菌丝协议。持续追踪 AI 工具、系统和实验的内容节点。

---

nature-skills: 18 AI Skills for the Full Academic Paper Workflow — 34k Stars in Two Months

by Mycelium Protocol

---

The pain of academic paper writing is not one single step — it is the whole chain: literature search, figure production, English polishing, statistical review, reviewer response, patent conversion. Any link that goes wrong delays submission.

nature-skills started with one skill (nature-polishing), expanded to nature-figure, and has grown to 18 skills covering nearly every operational node from starting to write to post-publication. In two months, the project has reached 34,000 GitHub stars.

GitHub: https://github.com/Yuan1z0825/nature-skills | ⭐ 34,065 | Apache 2.0

---

Install

npx skills add Yuan1z0825/nature-skills

Works with Claude Code, Codex, OpenClaw, OpenCode, and Hermes. After install, just say the trigger phrase in conversation — no extra configuration.

---

18 Skills at a Glance

Skill	Status	Purpose
nature-polishing	Stable	Polish, rewrite, or translate academic text to Nature-style English; scan for terminology, units, precision, and claim drift
nature-figure	Stable	Publication-ready figure workflow (Python/R); includes GPT Image 2 paper schematic drafts
nature-ref-verifier	Stable	Multi-source cross-verification of references: author/title/year/volume/page field-by-field comparison
nature-literature-pipeline	Stable	Automated literature discovery: multi-source retrieval, six-dimension scoring, deep-read push, local archive
nature-citation	Beta	Retrieve supporting references strictly scoped to Nature/CNS series; export ENW/RIS/Zotero RDF
nature-reader	Beta	Full-text Markdown reader with source anchors, figure–text alignment, equation rendering, bilingual parallel
nature-paper-card	Beta	Deep-read a single paper: 16-section Paper Card with evidence chains, argument logic, and testable research ideas
nature-response	Beta	Parse revision emails; generate independent replies for each blind reviewer + cover letter + LaTeX template
nature-paper2ppt	Beta	Generate a Chinese-language PPTX journal-club deck from a research paper
nature-paper-to-patent	Beta	Generate evidence-constrained Chinese invention patent drafts from papers; patent mining and novelty search
nature-academic-search	Beta	Multi-source literature search, citation verification, strict citation audit, citation impact tables
nature-downloader	Beta	Legally obtain full-text PDFs via library portals, CARSI, and open-access paths
nature-writing	Draft	Draft Nature-style manuscript sections; rebuild paper argumentation
nature-reviewer	Draft	Simulate Nature-style pre-submission review; output three blind reviewer reports with Major/Minor issues
nature-data	Draft	Data Availability statement, data repository plan, FAIR metadata check
nature-statistics	Draft	Review/rewrite statistical reporting: p-values, multiple comparisons, effect sizes, confidence intervals
nature-experiment-log	Draft	Standardize recording of experiment images, voice, and text; generate Obsidian experiment log
nature-proposal-writer	Beta	Proposal-first research writing state machine: establish evidence/argument/section contract before drafting

Stable = production-ready rules. Beta = functionally complete, boundaries still being refined. Draft = usable but rules still iterating.

---

Design Principles

All 18 skills share five design principles:

1. First-source priority: rules are based on published Nature content, official journal guidelines, or explicit local sources — not general aesthetic preference
2. Explicit over implicit: every rule explains its reasoning rather than just asserting it
3. Context-aware: writing, figures, citations, and reviewer responses each depend on where in the paper you are
4. Output-first: every skill returns something directly usable — pasteable text, .svg, .pptx, .docx — not a list of suggestions
5. Self-contained and extensible: each skill lives in its own directory; adding a new skill does not require modifying any existing one

---

From nature-polishing to 18 Skills

The collection started from a single polishing skill, expanded to figure production, and has grown into a full-coverage Nature-paper toolset through open-source community collaboration — inspired by DeepMind’s Science Skills, developed into something broader.

---

Mycelium Protocol — tracking the deep evolution of AI systems

© 2026 Mycelium Protocol. All rights reserved.