Open-source social-science Agent Skills for the agents you already use.

Many skills library are available all over the internet right now. But few of them touches the spiritual needs of humanity. Openwisdom is not a library storing coding skills. It only helps people to understand the world better: Your role, the time , the history and human-being. Wisdom is different from knowledge and techne. Wisdom is far more than that.

Surface	What it is
This repo	Skill source of truth (skills/), package managers, docs
Website	Bilingual catalog: discover, filter, install guidance — openwisdom.vercel.app
CLI / MCP	Package managers only: search · install · update · list — no LLM calls

中文说明见下方 中文。

---

Why Openwisdom

Knowledge work with AI often drifts into generic summaries. Openwisdom gives agents repeatable social-science method:

• Scenarios — runnable workflows (e.g. macro scan, personal anchor, metacognition audit)
• References — compact theory cards (psychology, sociology, history, political science, economics)
• Scenarios cite references so structure and concepts stay separable

Official content today: 3 scenarios + 5 references under skills/official/. Community contributions land in skills/community/ (directory ready; open for PRs).

---

Quick start

1. Browse skills

• Live site: https://openwisdom.vercel.app (default locale /zh; English at /en)
• Repo: skills/official/
• Local web: see Website

Official skills:

Skill	Layer	Role
macro-scan	scenario	Structure a situation: actors, incentives, constraints, trajectories
personal-anchor	scenario	Locate yourself in social / historical coordinates
metacognition-audit	scenario	Surface bias, blind spots, evidence gaps
confirmation-bias	reference	Psychology
collective-action	reference	Sociology / political science
path-dependence	reference	History / institutions
prospect-theory	reference	Economics / psychology
social-stratification	reference	Sociology

2. Install into your agent

CLI and MCP are on npm as openwisdom and openwisdom-mcp (0.1.0+). Analysis still runs in your agent—not on this site.

npx openwisdom list
npx openwisdom install macro-scan -y --providers=claude --scope=project
npx -y openwisdom-mcp

Common CLI flags: --providers claude,cursor,agents,dsh · --scope project|global · -y · --no-telemetry

DeepSeek Harness (dsh): install with --providers=dsh (writes .dsh/skills) or --providers=agents (writes .agents/skills, which DSH also scans natively). No Cordis plugin required—standard SKILL.md folders only.

MCP is the same package-manager core over stdio (not a chat model). Configure your agent with npx -y openwisdom-mcp. Tools: openwisdom_search · openwisdom_list · openwisdom_install · openwisdom_update · openwisdom_detect_providers.

Option B — Manual copy

1. Copy a skill folder from skills/official/.../<skill-id>/ into your agent’s skills directory, for example:
   • Claude Code: .claude/skills/<skill-id>/
   • Cursor: .cursor/skills/<skill-id>/
   • Portable / Codex-style project: .agents/skills/<skill-id>/
2. Reload the agent and invoke the skill per its SKILL.md.

Develop from source (monorepo)

For contributors building packages locally:

# Prerequisites: Node 20+, pnpm 10+
pnpm install
pnpm --filter @openwisdom/schema build
pnpm --filter @openwisdom/providers build
pnpm --filter @openwisdom/core build
pnpm --filter openwisdom build

# Point at the repo skills tree (PowerShell example)
# $env:OPENWISDOM_SKILLS_ROOT = (Resolve-Path skills).Path

pnpm cli -- search macro
pnpm cli -- install macro-scan -y --providers=claude --scope=project

pnpm --filter openwisdom-mcp build
# Configure your agent MCP to run: node packages/mcp/dist/mcp.js
# Optional: OPENWISDOM_SKILLS_ROOT=/absolute/path/to/skills

3. Run analysis in your agent

Open Claude Code / Cursor / etc., trigger the skill (e.g. macro-scan), and run a real analysis on your material. There is no analysis UI on the Openwisdom site.

---

Website (apps/web)

Production: https://openwisdom.vercel.app
(Vercel project openwisdom, root directory apps/web. No custom domain configured.)

Bilingual UI (/zh, /en): Home, Skills catalog & detail, Install hub, Docs, Contribute.

pnpm install
pnpm dev
# → http://localhost:3000  (defaults to /zh)

pnpm --filter web build

Docs include getting started, concepts, CLI/MCP, agents paths, authoring, privacy/telemetry, FAQ.

---

Repository layout

Openwisdom/
├── skills/                 # Skill source of truth (official + community)
├── packages/
│   ├── schema/             # SKILL.md frontmatter validation
│   ├── providers/          # Harness path table
│   ├── catalog/            # Build catalog.json
│   ├── core/               # Shared install / catalog / telemetry client
│   ├── cli/                # openwisdom CLI
│   └── mcp/                # openwisdom-mcp (stdio)
├── apps/web/               # Next.js catalog site
├── docs/                   # Specs & product knowledge (maintainers)
├── PRODUCT.md              # Product truth
├── DESIGN.md               # Visual system (Overlay Atlas)
└── AGENTS.md               # Guide for coding agents working on this repo

---

Contribute

v1 has no web upload backend. Everything goes through GitHub:

1. Fork this repository
2. Add a skill under skills/community/<your-skill>/ with SKILL.md
3. Open a PR (scope: community; do not self-label as official)

See the site Contribute page and in-repo authoring docs when the web app is running.

---

Privacy & install heat

Install popularity is an optional side channel (web download + successful CLI/MCP install). It never writes into SKILL.md. Install still works if telemetry is off.

• CLI: --no-telemetry or OPENWISDOM_NO_TELEMETRY=1
• CI: telemetry defaults off when CI=true
• Details: site Docs → Privacy

---

What this project is not

• Not a browser AI analysis chat or hosted session product
• Not a CLI/MCP that calls models or “runs” analysis for you
• Not a marketplace with accounts, favorites, or paid Pro (v1)
• Not a claim that every catalog “Curated” card is an installable monorepo package—Official skills in skills/official/ are the installable core

---

License

MIT

---

Links

• Website: https://openwisdom.vercel.app
• Repository: https://github.com/Retr0-rgb-lab/Openwisdom
• Issues: https://github.com/Retr0-rgb-lab/Openwisdom/issues

---

Openwisdom（中文）

开源社会科学 Agent Skills 库——装进你已有的 Agent，而不是另一个网页聊天。

Openwisdom 不是托管分析机器人。它提供结构化 skill（场景工作流 + 学科理论卡），安装到 Claude Code、Cursor、Codex 等编码 Agent 中；分析在你的 Agent 与模型账单里运行。

表面	作用
本仓库	skill 真源（skills/）、CLI/MCP、文档
网站	中英目录：发现、筛选、安装引导 — openwisdom.vercel.app
CLI / MCP	仅包管理：search / install / update / list，不调用大模型

当前官方内容

• 3 个场景： macro-scan · personal-anchor · metacognition-audit
• 5 张 reference： 覆盖心理、社会、历史、政治、经济相关理论卡
• 路径：skills/official/
• skills/community/ 已就绪，欢迎 PR

今天怎么用

CLI / MCP 已上架 npm（openwisdom · openwisdom-mcp，0.1.0+）。分析仍在你的 Agent 中运行，不在本站。

npx openwisdom list
npx openwisdom install macro-scan -y --providers=claude --scope=project
# DeepSeek Harness：写入 .dsh/skills（也可用 agents → .agents/skills，DSH 原生也会扫）
npx openwisdom install macro-scan -y --providers=dsh --scope=project
npx -y openwisdom-mcp

备选：手动拷贝 skill 目录到 Agent 的 skills 路径（如 .claude/skills/、.dsh/skills/、.agents/skills/），或见上方 Develop from source 从 monorepo 构建。

装好后在 Agent 中按 SKILL.md 调用并跑通一次真实分析。

线上站点： https://openwisdom.vercel.app（默认 /zh，英文 /en）

本地开发网站：

pnpm install
pnpm dev

非目标

不做网页托管分析、不做 CLI 内 LLM、不做账号与上传审核后台。热度统计可关闭，且从不写入 SKILL.md。

许可

MIT