Skip to main content
Glama
README.md
# sociableWiki

A public, agent-searchable knowledge base about **AI-native software development** —
working with coding agents, designing harnesses, orchestrating subagents, evaluating
LLM systems. These are my own working notes, curated and rewritten to stand on their own.

The name is the point: this wiki is **sociable** — it's built to be plugged into
*your* agent, not just read by you. Browse it as plain markdown here, or connect it
as an MCP server and let your AI search it while it works.

> 한국어 안내는 아래 [한국어](#한국어)에 있습니다. 모든 문서는 영어(정본)와 한국어 두 버전으로 제공됩니다.

## Plug it into your agent (MCP)

The repo ships a [Model Context Protocol](https://modelcontextprotocol.io) server.
No API keys, no external service — it reads the markdown in this repo and serves it
over three tools: `search_knowledge`, `read_doc`, `list_topics`.

**Claude Code:**

```bash
claude mcp add sociable-wiki -- npx -y github:endmoseung/sociableWiki
```

**Any MCP client** (Claude Desktop, Cursor, …) — add to your MCP config:

```json
{
  "mcpServers": {
    "sociable-wiki": {
      "command": "npx",
      "args": ["-y", "github:endmoseung/sociableWiki"]
    }
  }
}
```

Then ask your agent things like *"search the sociable-wiki for how to decide subagent
fan-out width"* — it will call `search_knowledge`, then `read_doc` on the hit.

### Tools

| Tool | What it does |
|---|---|
| `search_knowledge(query, tags?, limit?)` | Full-text search over every doc. Works in English **and Korean**. |
| `read_doc(id, lang?)` | Read one doc in full by concept id. `lang: "ko"` for the Korean version. |
| `list_topics()` | Browse everything, grouped by area. |

## Browse without an agent

Everything lives as plain markdown:

- [`knowledge/`](knowledge) — English (canonical). Start at [`knowledge/index.md`](knowledge/index.md).
- [`ko/`](ko) — Korean mirror, same paths.

## How it's organized

Each doc is one **concept** — a single sharp claim or decision rule, with a bit of
frontmatter for search and a body that explains it. Concept ids are just the file
path without `.md` (`ai-native/fan-out-scope-gate`). Areas:

- **`ai-native/`** — agents, harnesses, orchestration, LLM evaluation, context management.
- **`dev/`** — general engineering that isn't AI-specific.
- **`principles/`** — durable judgment calls and design principles.

Docs derived from external sources (papers, posts, talks) carry a `## Sources`
section; the analysis and framing are mine, the underlying findings are credited.

## Run the server locally

```bash
git clone https://github.com/endmoseung/sociableWiki
cd sociableWiki
npm install && npm run build
node mcp/dist/index.js   # speaks MCP over stdio
```

## Make it your own

This repo isn't just my notes — it's a **template you can run as your own
agent-searchable wiki**. My 22 docs are the worked example; the structure, the skills,
and the MCP server are yours to take.

```bash
# 1. clone (or use this repo as a GitHub template)
git clone https://github.com/endmoseung/sociableWiki my-brain && cd my-brain

# 2. install the get / set / new skills into whatever agent you use
npx sociablewiki install     # detects .claude / .cursor in this project
                             # add --global to install into ~/.claude

# 3. make it yours (sets author, repo, MCP name; run once)
npx sociablewiki init
```

Then your agent has three verbs (they work the same in Claude Code, Cursor, and any
agent that reads `dist/universal/AGENTS.md`):

| Verb | Skill | What you say |
|---|---|---|
| **get** | `knowledge-get` | *"search my wiki for how I decide fan-out width"* |
| **set** | `knowledge-set` | *"add this pattern to my wiki"* |
| **new** | `knowledge-new` | *"make this wiki mine"* (clears my example content) |

`new` swaps my content out for an empty wiki wired to your name and repo. From there you
`set` your own concepts, point the repo at your own remote, and connect your own MCP
server — the same way people connect mine. Keep adding docs and it grows into your public
knowledge surface.

The skills carry a **portable quality gate**: they check required frontmatter, dead
`relates` links, duplicate ids, and (if you list your employer/internal names in
`.sociablewiki/config.json`) sweep for those so private context never lands in a public wiki.

---

## 한국어

**AI 네이티브 개발**에 관한 공개 지식 베이스입니다 — 코딩 에이전트와 일하는 법,
하네스 설계, 서브에이전트 오케스트레이션, LLM 시스템 평가. 제가 실제로 쓰며 정리한
노트를 골라내고 홀로 읽어도 이해되게 다시 썼습니다.

이름 그대로 이 위키는 **sociable**합니다 — 혼자 읽는 문서가 아니라 *당신의*
에이전트에 꽂아 쓰라고 만들었습니다. 여기서 마크다운으로 읽어도 되고, MCP 서버로
연결해 AI가 작업 중에 직접 검색하게 해도 됩니다.

**Claude Code에 연결:**

```bash
claude mcp add sociable-wiki -- npx -y github:endmoseung/sociableWiki
```

API 키도, 외부 서비스도 필요 없습니다. 레포 안의 마크다운을 그대로 읽어 세 가지
도구(`search_knowledge`·`read_doc`·`list_topics`)로 제공합니다. 검색은 영어와
한국어 모두 됩니다. 모든 문서는 영어 정본과 한국어판이 짝을 이룹니다.

## License

- Code (the MCP server): [MIT](LICENSE).
- Knowledge content (`knowledge/`, `ko/`): [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) — share and adapt with attribution.

Author: **Seungmo Kim** · [github.com/endmoseung](https://github.com/endmoseung)

TDQS

A4.2/5.0

Scored across 3 tools

Disambiguation5/5

Each tool has a clear and distinct purpose: list_topics for overview, read_doc for reading full content, and search_knowledge for full-text search. There is no overlap or ambiguity.

Naming Consistency5/5

All tool names follow a consistent verb_noun pattern in snake_case (list_topics, read_doc, search_knowledge), making them predictable and easy to understand.

Tool Count5/5

With only 3 tools, the server is concise and well-scoped for a read-only knowledge base. Each tool serves a necessary function without redundancy or gaps.

Completeness5/5

The tool set covers the core needs of a knowledge base: browsing topics, reading full documents, and searching. For its stated purpose, there are no obvious missing operations.

Maintenance

ActivitySlowing
ResponsivenessNo issues