"Wer nicht von dreitausend Jahren sich weiß Rechenschaft zu geben, bleibt im Dunkeln unerfahren, mag von Tag zu Tage leben."
不能汲取三千年历史经验的人，没有未来可言。
He who cannot draw on three thousand years of history is living hand to mouth.
— Johann Wolfgang von Goethe
Our mission: Unite the knowledge of the world's top experts across every domain — to accelerate AI-driven scientific discovery.
Call for action: Share your research expertise. Together, we create the AI-era Einstein, Da Vinci, and Kant.
View Interactive Knowledge Tree →
OpenScientist is a curated library of Claude Code Skills — structured Markdown files that give AI agents deep, expert-level reasoning capabilities in specific scientific domains.
Each skill is written by a domain expert and encodes the knowledge, tools, reasoning protocols, and common pitfalls of their field. Point your AI agent at a skill, and it reasons like a domain expert.
Contribute your expertise, or use this repo to supercharge your AI agent's scientific discovery.
OpenScientist is a fully open-source, non-profit initiative. As the project grows, we plan to establish a non-profit organization (NGO) to ensure long-term governance, transparency, and stewardship of all contributed knowledge.
Turning your know-how into AI-reusable knowledge means:
- Boost your own research efficiency — your AI agent gains your expertise and works alongside you
- Boost humanity's research efficiency — every scientist benefits from the collective knowledge
- Survive the Singularity — when ASI takes over, your contribution to this repo might just save your life
Aligned with the arXiv category taxonomy. 8 domains, 155 subcategories.
| Domain | arXiv | Subcategories | Reviewer(s) |
|---|---|---|---|
| ⚛️ Physics | astro-ph, cond-mat, gr-qc, hep, nlin, physics, ... | 51 | Seeking reviewer |
| ➗ Mathematics | math | 32 | Seeking reviewer |
| 💻 Computer Science | cs | 40 | Seeking reviewer |
| 🧬 Quantitative Biology | q-bio | 10 | Seeking reviewer |
| 📊 Statistics | stat | 6 | Seeking reviewer |
| ⚡ Electrical Engineering & Systems Science | eess | 4 | Seeking reviewer |
| 📈 Economics | econ | 3 | Seeking reviewer |
| 💹 Quantitative Finance | q-fin | 9 | Seeking reviewer |
View all 155 subcategories in the interactive knowledge tree →
Each skill is a single .md file. Install it once, invoke it any time in Claude Code:
# 1. Clone
git clone https://github.com/OpenScientists/OpenScientist.git
# 2. Copy a skill (or symlink a whole domain)
cp OpenScientist/skills/physics/quantum-physics/quantum-entanglement.md ~/.claude/skills/
# or: ln -s $(pwd)/OpenScientist/skills/physics ~/.claude/skills/os-physics
# 3. Invoke in Claude Code
/quantum-entanglement → Claude reasons as a quantum physics expert| Section | Purpose |
|---|---|
| YAML frontmatter | Machine-readable metadata: name, domain, author, status |
## Purpose |
When to invoke this skill |
## Domain Knowledge |
Core concepts, equations, established facts |
## Reasoning Protocol |
Step-by-step guide for AI reasoning |
## Tools |
Key software, libraries, databases used in this domain |
## Common Pitfalls |
Mistakes and edge cases to avoid |
## Examples |
Worked examples |
## References |
Key papers and textbooks |
| Status | Meaning |
|---|---|
draft |
Authored, not yet peer-reviewed |
reviewed |
Approved by a domain expert reviewer |
verified |
Tested in real AI-scientist workflows |
Every pull request touching a skill file triggers CI (utils/tools/validate.py) that checks required fields and section structure. A PR cannot be merged if validation fails.
We welcome contributions from domain experts. See CONTRIBUTING.md for the full guide.
Who can contribute? We maintain a high bar for scientific accuracy.
- Academic credential — PhD degree or equivalent research position (postdoc, research scientist, professor, etc.) is required
- Real-name identity — Contributors must use their real name and institutional affiliation in the
authorfield (e.g.,"Dr. Albert Einstein (ETH Zürich Physics)") - Domain expertise — You may only contribute skills within your area of professional expertise
- Fork this repo
- Copy the template into the right domain folder:
cp skills/_template.md skills/<domain>/<subdomain>/<your-skill-name>.md
- Fill in every section — Purpose, Domain Knowledge, Reasoning Protocol, Tools, Common Pitfalls
- Validate locally (optional but recommended):
python utils/tools/validate.py skills/<domain>/<subdomain>/<your-skill-name>.md
- Open a pull request — title format:
[physics/quantum-physics] Add quantum-entanglement skill
A domain reviewer listed in CODEOWNERS will be automatically assigned to review your PR for scientific accuracy.
Reviewers are domain experts who ensure the scientific quality of skills in their subdomain.
- Meet all requirements for contributors (i.e. be a qualified contributor first)
- Have substantial peer-review experience in the relevant subdomain
- Review skill PRs in your subdomain for scientific accuracy and completeness
- Provide constructive feedback to contributors
- Promote skill status from
draft→reviewedafter verification
- Approve or request changes on PRs touching your subdomain
- Self-approve and merge your own PRs within your subdomain
- Auto-assigned as reviewer via CODEOWNERS when a PR touches your subdomain
- Apply to become a category reviewer →
- View all reviewers in the Knowledge Tree →
- View all reviewers list →
| Template | When to use |
|---|---|
| Skill Request | Need a skill but can't write it yourself |
| Reviewer Application | Apply to become a subdomain reviewer |
| Propose New Area | Propose a new top-level domain |
- Contributor opens a skill PR
- CI automatically runs
validate.pyto check required fields and structure - CODEOWNERS assigns the subdomain reviewer
- Reviewer approves or requests changes
- Merge → skill status starts as
draft
Status lifecycle: draft → reviewed (reviewer approves) → verified (tested in a real AI workflow)
When a Reviewer Application issue receives the approved label, the onboard-maintainer workflow automatically:
- Adds the reviewer to
.github/CODEOWNERS - Adds their name to the Knowledge Tree and Reviewers page
- Closes the issue with a welcome comment
Each skills/<domain>/<subdomain>/ path maps to a reviewer in .github/CODEOWNERS. PRs touching that path automatically request the assigned reviewer. Unclaimed subdomains fall back to @HHHHHejia.
Want to help with CI, documentation, community management, or project operations? Reach out: hejia@tapntell.ai
CC BY 4.0 — free to share and adapt, with attribution.
我们的使命： 汇集全人类各领域顶尖专家的知识，加速 AI 驱动的科学进步。
行动号召： 共享你的研究知识，创造 AI 时代的爱因斯坦、达芬奇与康德。
OpenScientist 是一个精心策划的 Claude Code Skills 库 —— 每个 Skill 是一个结构化的 Markdown 文件，赋予 AI 智能体特定科学领域的专家级推理能力。
每个 Skill 由该领域的专家撰写，编码了领域知识、工具、推理协议和常见陷阱。让 AI 调用一个 Skill，就能像领域专家一样思考。
贡献你的专业知识，或使用本仓库加速你 AI agent 的科学发现。
OpenScientist 是一个完全开源、非盈利的项目。随着项目的发展，我们计划成立一个非盈利组织（NGO），以确保所有贡献知识的长期治理、透明性和可持续管理。
将你的 know-how 变成 AI 可复用的知识意味着：
- 提升你自己的科研效率 —— 你的 AI agent 获得你的专业知识，成为你的研究搭档
- 提升全人类的科研效率 —— 每位科学家都能从集体知识中受益
- 在奇点中存活 —— 当 ASI 统治人类以后，看到这个仓库里你的贡献，没准可以饶你一命
对齐 arXiv 分类体系。8 个顶层领域，155 个子领域。
| 领域 | arXiv | 子领域数 | 审稿人 |
|---|---|---|---|
| ⚛️ Physics 物理 | astro-ph, cond-mat, gr-qc, hep, nlin, physics, ... | 51 | 招募中 |
| ➗ Mathematics 数学 | math | 32 | 招募中 |
| 💻 Computer Science 计算机科学 | cs | 40 | 招募中 |
| 🧬 Quantitative Biology 定量生物学 | q-bio | 10 | 招募中 |
| 📊 Statistics 统计学 | stat | 6 | 招募中 |
| ⚡ Electrical Engineering & Systems Science 电气工程与系统科学 | eess | 4 | 招募中 |
| 📈 Economics 经济学 | econ | 3 | 招募中 |
| 💹 Quantitative Finance 定量金融 | q-fin | 9 | 招募中 |
每个 Skill 是一个 .md 文件，安装一次，在 Claude Code 中随时调用：
# 1. 克隆仓库
git clone https://github.com/OpenScientists/OpenScientist.git
# 2. 复制 Skill（或符号链接整个领域）
cp OpenScientist/skills/physics/quantum-physics/quantum-entanglement.md ~/.claude/skills/
# 或：ln -s $(pwd)/OpenScientist/skills/physics ~/.claude/skills/os-physics
# 3. 在 Claude Code 中调用
/quantum-entanglement → Claude 以量子物理专家身份推理| 部分 | 作用 |
|---|---|
| YAML frontmatter | 机器可读的元数据：name、domain、author、status |
## Purpose |
何时调用此 Skill |
## Domain Knowledge |
核心概念、公式、既定事实 |
## Reasoning Protocol |
AI 推理的分步指南 |
## Tools |
该领域常用的软件、库、数据库 |
## Common Pitfalls |
常见错误和边界情况 |
## Examples |
示范性例题 |
## References |
关键论文和教材 |
| 状态 | 含义 |
|---|---|
draft |
已撰写，尚未同行评审 |
reviewed |
已由领域专家审核通过 |
verified |
已在真实 AI 科学家工作流中验证 |
每次 PR 修改 Skill 文件时，CI 会自动运行 utils/tools/validate.py 检查必填字段和章节结构。校验不通过则无法合并。
我们欢迎各领域专家贡献知识。请参阅 CONTRIBUTING.md 了解完整流程。
谁可以贡献？ 我们对科学准确性有严格要求。
- 学术资质 — 必须持有博士学位或同等研究岗位（博士后、研究员、教授等）
- 实名认证 — 贡献者必须在
author字段使用真实姓名和所属机构（如"Dr. Albert Einstein (ETH Zürich Physics)") - 领域专长 — 只能在自己的专业领域内贡献 Skill
- Fork 本仓库
- 复制模板 到对应领域文件夹：
cp skills/_template.md skills/<领域>/<子领域>/<你的skill名称>.md
- 填写每个章节 —— Purpose、Domain Knowledge、Reasoning Protocol、Tools、Common Pitfalls
- 本地验证（推荐）：
python tools/validate.py skills/<领域>/<子领域>/<你的skill名称>.md
- 提交 Pull Request —— 标题格式：
[physics/quantum-physics] Add quantum-entanglement skill
CODEOWNERS 中的领域审稿人会自动收到 review 请求，负责审核科学内容的准确性。
审稿人是负责其子领域 Skill 科学质量的领域专家。
- 满足贡献者的所有要求（即首先是合格的贡献者）
- 在相关子领域有充分的同行评审经验
- 审核所属子领域的 Skill PR，确保科学准确性和完整性
- 为贡献者提供建设性反馈
- 验证后将 Skill 状态从
draft提升为reviewed
- 对所属子领域的 PR 进行审批或提出修改意见
- 在自己的子领域内可以自审自批、合并自己的 PR
- 通过 CODEOWNERS 自动分配为审稿人
| 模板 | 使用场景 |
|---|---|
| Skill Request | 需要某个 Skill 但自己写不了 |
| Reviewer Application | 申请成为子领域审稿人 |
| Propose New Area | 提议新的顶层领域 |
- 贡献者提交 Skill PR
- CI 自动运行
validate.py检查必填字段和结构 - CODEOWNERS 自动分配子领域审稿人
- 审稿人审批或提出修改意见
- 合并 → Skill 状态初始为
draft
状态生命周期： draft → reviewed（审稿人审核通过）→ verified（在真实 AI 工作流中验证）
当 Reviewer Application issue 被打上 approved 标签后，onboard-maintainer 工作流自动：
每个 skills/<domain>/<subdomain>/ 路径在 .github/CODEOWNERS 中映射到对应审稿人。涉及该路径的 PR 会自动请求对应审稿人 review。尚未认领的子领域由 @HHHHHejia 负责。
想参与 CI 维护、文档完善、社区管理或项目运营？联系我们：hejia@tapntell.ai