企业大模型技能库 · Enterprise AI Skill Library
从 CEO 到一线员工，每个岗位都有专属的 AI 使用技能 From CEO to frontline — specialized AI skills for every enterprise role
English · 中文 · Vision 愿景 · Contribute 贡献 · Community 社区
与 data/roadmap.json 中 vision 字段一致 · Same as the vision field in that file:
中文： 打造可复用、可评审的企业大模型技能库，以双语为方向（可先单语并标注翻译协作），覆盖从高管到一线岗位；社区通过 Issue/PR 模板贡献，CI 保障格式与质量。
English: A reusable, reviewable enterprise LLM skill library—bilingual by aspiration, with single-locale first contributions welcome when translation needs are noted—from leadership to frontline roles, with community contributions via Issue/PR templates and CI guarding format and quality.
OpenSkill 是一个开源的企业级大模型技能库，收录了从 CEO 到一线员工各个岗位的高质量 AI 技能（System Prompts）。
每个"技能"（Skill）都是一个即用型的提示词，设计原则：
- 🎯 原子级：一个 Skill 只做一件事，做好做深
- 🌐 双语：目标为每个 Skill 具备中英可用内容；也欢迎先提交单语并标注翻译需求（见 CONTRIBUTING.md）
- ✅ 可评审：每个 Skill 以包含真实输出示例为目标（存量内容持续补齐）
- 🔌 框架无关：兼容 ChatGPT / Claude / Gemini / 开源模型
当前大多数提示词库存在以下问题：
- ❌ 通用化，没有针对特定岗位的深度设计
- ❌ 质量参差不齐，无法评估
- ❌ 无法版本控制，团队无法复用
- ❌ 与 Agent 框架不兼容
OpenSkill 的解法：用**结构化的 Skill 格式（基于 COSTAR 框架）**解决以上所有问题，并建立社区驱动的质量保障体系。
OpenSkill is an open-source enterprise AI skill library containing high-quality AI skills (system prompts) for every role, from CEO to frontline employees.
Each "Skill" is a ready-to-use prompt designed with:
- 🎯 Atomic: One Skill does one thing — deeply and well
- 🌐 Bilingual: We aim for Chinese and English coverage per skill; single-locale first contributions are welcome with a clear translation note (see CONTRIBUTING.md)
- ✅ Reviewable: Every Skill aims to include a real output example (we keep backfilling gaps)
- 🔌 Model-agnostic: Works with ChatGPT, Claude, Gemini, and open-source models
Most prompt libraries today are:
- ❌ Too generic — little depth for specific job roles
- ❌ Uneven quality — hard to judge what “good” looks like
- ❌ Hard to version — teams can’t reuse safely
- ❌ Weak fit for agents — not structured for tooling
OpenSkill addresses this with a structured COSTAR-based skill format and community-driven quality norms.
| 层级 Level | 岗位 Roles | 技能数 Skills |
|---|---|---|
| 🏢 C-Suite | CEO · CTO · CFO · COO · CMO · CHRO |
|
| 👔 Management | Engineering Manager · Product Manager · HR Manager · Sales Manager · Operations Manager · Finance Manager · Marketing Manager |
|
| 💻 Individual Contributor | Software Engineer · Data Analyst · Data Scientist · Designer · Marketing · Sales Rep · Customer Success · Legal · Finance · DevOps · Security |
|
| 🌐 Cross-functional | OKR · Meeting · Email · Data · Weekly · Decision · Retro · Stakeholder · Kickoff · Knowledge transfer · Minutes · Exec summary · Feedback · Escalation · Deck outline · SOP · Change · RACI · Risks · Brainstorm |
|
📈 欢迎共建 · Built with the community — 通过 Issue / PR 提交技能与改进 · Submit skills and improvements via Issues and PRs
直接访问 openagenticos.github.io/OpenSkill — 按岗位筛选、搜索、一键复制系统提示词。支持中英文切换。
English: Visit openagenticos.github.io/OpenSkill — filter by role, search, and one-click copy system prompts. Supports EN/中文 toggle.
工作流与 Recipe 是声明式编排与阅读入口：在网站上按步骤跟做，需要模型辅助时打开链接的技能页并复制提示词；不会在 OpenSkill 云端自动执行。若要在 Claude、扣子（Coze）、OpenClaw 等环境中落地，请阅读在线 使用指南（中文） / Usage guide (English)（与工作流、场景详情页中的说明一致）。
English: Workflows and recipes are declarative maps and reading paths: follow steps on the site, open linked skills to copy prompts when you need the model; nothing runs on OpenSkill’s servers. For Claude, Coze, OpenClaw, and similar tools, see the Usage guide (English) / 使用指南（中文） — same story as the callouts on workflow and recipe pages.
- 找到你岗位的技能目录（如 CEO 战略愿景（中文））
- 复制「系统提示词」部分到 ChatGPT/Claude 的 System Prompt 区域
- 用「用户提示词模板」填入你的实际内容，发送即可
每个 GitHub Release 附带：
openskill.json—format_version: 3，按id合并 zh + en（persona_zh/persona_en、system_prompt_zh/system_prompt_en等，见 SKILL_SCHEMA.md）；技能可含可选字段evaluation_rubric、test_cases、enhancers、composable_with（存在则导出）。openskill.zh.json— 每条 skill 一条中文侧记录（来自*.zh.skill.md）openskill.en.json— 每条 skill 一条英文侧记录（来自*.en.skill.md）openskill.workflows.json—format_version: 1，声明式工作流（来自workflows/<category>/*.workflow.md配对，含category流程域）openskill.recipes.json—format_version: 1，角色/场景入口 Recipe（来自recipes/<layer>/*.recipe.md配对，含category岗位路径）openskill.progressive.json/openskill.mcp-resources.json—format_version: 1，渐进分层与实验性skill://描述（Release 构建含此两项；本地需npm run export:all）
维护者推送 v* 标签即可由 Actions 构建并上传；下载地址与 owner/repo 无关，fork 亦适用。
English: Each release ships the skill bundles above plus openskill.progressive.json and openskill.mcp-resources.json from npm run export:all. Locally, default npm run export omits those two; use export:all when you need them. Push a v* tag to build (any fork uses its own Releases URL).
import json
import urllib.request
BASE = "https://github.com/OpenAgenticOS/OpenSkill/releases/latest/download"
for name in ("openskill.json", "openskill.zh.json", "openskill.en.json"):
with urllib.request.urlopen(f"{BASE}/{name}") as r:
bundle = json.loads(r.read().decode())
print(name, "skills:", bundle["skills_count"])
with urllib.request.urlopen(f"{BASE}/openskill.en.json") as r:
bundle = json.loads(r.read().decode())
by_id = {s["id"]: s for s in bundle["skills"]}
skill = by_id["c-suite/ceo/strategic_vision"]
print(skill["name"], len(skill["system_prompt"]))本地生成（需 Node 24）：npm ci && npm run export → 写入主技能与工作流/Recipe JSON；npm run export:all 额外生成 dist/openskill.progressive.json 与 dist/openskill.mcp-resources.json。可选：npm run eval -- skills/.../x.zh.skill.md（需 API Key，见 SKILL_SCHEMA.md）。
English (local): npm ci && npm run export writes the core bundles; npm run export:all adds progressive + MCP resource JSON under dist/ (gitignored). Optional npm run eval for LLM-as-judge runs on skills with test_cases.
# pip install requests pyyaml
import yaml, re, requests
def load_skill(skill_path: str, locale: str = "zh") -> dict:
"""Load a skill from GitHub raw URL (locale: zh | en)."""
suffix = f".{locale}.skill.md"
url = f"https://raw.githubusercontent.com/OpenAgenticOS/OpenSkill/master/skills/{skill_path}{suffix}"
content = requests.get(url).text
match = re.match(r'^---\n(.*?)\n---', content, re.DOTALL)
frontmatter = yaml.safe_load(match.group(1))
# Chinese: ## 系统提示词 — English: ## System Prompt
heading = r'## 系统提示词' if locale == "zh" else r'## System Prompt'
prompt_match = re.search(heading + r'.*?```(?:xml)?\n(.*?)```', content, re.DOTALL)
frontmatter['system_prompt'] = prompt_match.group(1) if prompt_match else ""
return frontmatter
skill = load_skill("c-suite/ceo/strategic_vision", locale="zh")
print(f"Loaded: {skill['name']}")
print(f"System Prompt Preview:\n{skill['system_prompt'][:200]}...")from langchain_openai import ChatOpenAI
from langchain_core.messages import SystemMessage, HumanMessage
skill = load_skill("individual-contributor/software-engineer/code_review", locale="zh")
llm = ChatOpenAI(model="gpt-5.4")
response = llm.invoke([
SystemMessage(content=skill['system_prompt']),
HumanMessage(content=f"请评审以下代码：\n```python\n{your_code}\n```")
])
print(response.content)OpenSkill 的核心是社区。 你不需要是 AI 专家，只需要有某个岗位的实际工作经验！
OpenSkill's core is its community. You don't need to be an AI expert — just real work experience in any role!
在 GitHub 网页上直接添加新技能（无需本地环境）：
- 进入
skills/找到你的岗位目录 - 点击右上角 「Add file」→「Create new file」
- 文件名格式：
你的技能名.zh.skill.md与你的技能名.en.skill.md（同一技能、同一id） - 复制 最简贡献模板
- 填写内容，点击 「Propose new file」 提交 PR
就这么简单！ 维护者会在 72 小时内帮你完善格式。
语言： 可先只写中文或只写英文；若另一侧尚未补齐，请在 PR 里注明 Translation needed: en 或 Translation needed: zh（或请维护者打 needs-translation 标签）。详见 CONTRIBUTING.md。
English (same flow): Open skills/, use Add file → Create new file, add something.zh.skill.md and something.en.skill.md (same id), paste the minimal template, then Propose new file to open a PR. Maintainers typically respond within ~72 hours.
Language: You may start in Chinese or English only; if the other locale is not ready, note Translation needed: en or Translation needed: zh in the PR (or ask for needs-translation). See CONTRIBUTING.md.
| 贡献方式 · How to contribute | 难度 · Level | 时间 · Time |
|---|---|---|
| 🌟 新技能 / New skill | ⭐ 简单 · Easy | 15 min |
| 🔧 改进技能 / Improve skill | ⭐ 简单 · Easy | 10 min |
| 💬 测试反馈 / Test & feedback | ⭐ 简单 · Easy | 5 min |
| 📖 文档 / Docs | ⭐⭐ 中等 · Medium | 30 min |
| 🛠️ 工具链 / Tooling | ⭐⭐⭐ 进阶 · Advanced | 1+ h |
- Issues — 缺陷、技能提案、用法交流、集成与想法（可先搜索是否已有主题）· Bugs, skill proposals, Q&A, integrations, and ideas（本仓库未启用 GitHub Discussions，相关交流请用 Issues）
- Code of Conduct — 参与前请阅读 · Please read before participating
- Security — 漏洞请按此私下报告，勿发公开 Issue · Report vulnerabilities privately, not in public issues
本地运行 npm run validate / npm run build-index / npm run export 请使用 Node.js 24（与 package.json 的 engines 及 CI 一致）。
English: Use Node.js 24 locally for npm run validate / npm run build-index / npm run export (matches package.json engines and CI).
发版 · Cutting a release： 打标签并推送即可触发 Release · dist bundle 工作流，上传 openskill.json（format_version: 3）、openskill.zh.json、openskill.en.json 与 zip：git tag v2.1.0 && git push origin v2.1.0（需仓库 Actions → General → Workflow permissions 允许 Read and write）。Fork 上同样按各自 owner/repo 生成 Release。
English: Tag and push to run Release · dist bundle: uploads three JSON files plus zip. git tag v1.0.0 && git push origin v1.0.0 (enable Workflow permissions → Read and write). Forks publish under their own owner.
| 文档 · Doc | 中文 | English |
|---|---|---|
| CONTRIBUTING.md | 贡献流程、单语/翻译标注、本地校验、合并预期 | Contributor flow, single-locale & translation notes, local validation, merge expectations |
| docs/GOVERNANCE.md | 响应时间、标签、Issues、分支保护说明 | Response SLA, labels, Issues, branch protection notes |
| docs/SKILL_INDEX.md | 机器生成技能索引（npm run build-index） |
Generated skill index (npm run build-index) |
| SKILL_SCHEMA.md | Skill 完整格式与字段说明 | Full skill format and frontmatter reference |
| CODE_OF_CONDUCT.md | 社区行为准则（正文英 + 顶部中文概要） | Code of Conduct (EN body, CN summary on top) |
| SECURITY.md | 安全披露方式 | How to report security issues |
赞助 · Sponsors: 若组织已开通 GitHub Sponsors，可通过页顶 Sponsor 徽章 支持；配置见 .github/FUNDING.yml。
Sponsors: If the org has GitHub Sponsors enabled, use the badge above — settings live in .github/FUNDING.yml.
中文： 可选安装 Dev Containers，在 VS Code 中 Reopen in Container，容器内已配置 Node 24 与 npm ci。
English: Optional: install Dev Containers and use Reopen in Container in VS Code; the image uses Node 24 and runs npm ci on create.
OpenSkill/
├── skills/ # 技能库主体（原子 Prompt + 可选评估/组合元数据）
│ ├── c-suite/ # 高管层 (CEO/CTO/CFO/COO/CMO/CHRO)
│ ├── management/ # 中层管理
│ ├── individual-contributor/ # 一线（工程/数据/营销/销售等，见表格）
│ └── cross-functional/ # 跨职能通用
├── workflows/ # 声明式工作流：workflows/<流程域>/*.zh|en.workflow.md
├── recipes/ # 场景指南：recipes/<层级>/*.zh|en.recipe.md（与 skills 层级对齐）
├── data/ # roadmap.json（愿景与阶段计划，维护者参考）
├── docs/ # 治理、技能索引
├── schema/
│ ├── skill.schema.json # Skill frontmatter JSON Schema
│ ├── workflow.schema.json # Workflow frontmatter JSON Schema
│ └── recipe.schema.json # Recipe frontmatter JSON Schema
├── tools/
│ ├── validate.js # Schema + 章节校验（可传 skills/ 下路径；npm run validate -- …）
│ ├── export_skills.js # 生成 dist/openskill*.json（发版用）
│ ├── count_skills.js # 统计输出
│ └── build_index.js # 生成 docs/SKILL_INDEX.md
├── .github/
│ ├── workflows/ # CI、发版 dist、Dependabot 等
│ ├── ISSUE_TEMPLATE/ # 结构化 Issue 模板
│ ├── dependabot.yml
│ └── PULL_REQUEST_TEMPLATE.md
├── .devcontainer/ # Dev Containers（可选）
├── SKILL_SCHEMA.md # 完整格式文档
├── CONTRIBUTING.md # 贡献指南
├── CODE_OF_CONDUCT.md
└── SECURITY.md
各层向下兼容：只用 Prompt（.skill.md）即可；多用的层越多，越利于评审、组合与编排。
| 层 | 内容 | 位置 |
|---|---|---|
| L0 Prompt | COSTAR 正文 + 基础 frontmatter | skills/**/*.skill.md |
| L1 Evaluation | evaluation_rubric、test_cases（可选） |
Skill frontmatter |
| L2 Context | enhancers、composable_with（可选） |
Skill frontmatter |
| L3 Workflow | 多步编排（skill / human / tool / condition）；category = 流程域 |
workflows/<category>/*.workflow.md |
| L4 Recipe | 角色/场景入口指南；category = 岗位层级 |
recipes/<layer>/*.recipe.md |
导出：npm run export 会生成 dist/openskill.workflows.json 与 dist/openskill.recipes.json（见 SKILL_SCHEMA.md）。
基于 COSTAR 框架（企业 LLM 2025 年最佳实践）设计：
| 维度 | 作用 |
|---|---|
| Persona | AI 扮演什么角色 |
| Objective | 要完成什么任务 |
| Style | 期望的输出风格 |
| Tone | 期望的语气 |
| Audience | 面向什么受众 |
| Response Format | 期望的输出格式 |
延伸阅读 · Further reading: PersonaHub (arXiv:2406.20094)；COSTAR 类结构化提示在企业落地中较常见。
Further reading: PersonaHub (arXiv:2406.20094); structured prompts like COSTAR are widely used in enterprise practice.
想加入这个列表？ 提交你的第一个 Skill！
MIT © OpenAgenticOS · Contributors 贡献者 · Code of Conduct · Security
如果 OpenSkill 对你有帮助，请 ⭐ Star 这个仓库！ If OpenSkill helps you, please ⭐ Star this repo!