wiki-knowledge-complier — плагин для Claude Code

<h1 align="center">Wiki Knowledge Compiler</h1>

<p align="center">
  <em>Turn your documents into a personal knowledge base — powered by AI, stored as simple files.</em>
</p>

<p align="center">
  <a href="https://github.com/bradduy/wiki-knowledge-compiler/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License" /></a>
  <a href="https://github.com/bradduy/wiki-knowledge-compiler"><img src="https://img.shields.io/badge/platform-Claude%20Code-blueviolet.svg" alt="Claude Code" /></a>
  <a href="https://obsidian.md"><img src="https://img.shields.io/badge/works%20with-Obsidian-7C3AED.svg" alt="Works with Obsidian" /></a>
  <img src="https://img.shields.io/badge/offline-100%25-brightgreen.svg" alt="100% Offline" />
</p>

<p align="center">
  <img src="./assets/demo.gif" alt="Wiki Knowledge Compiler Demo" width="100%" />
</p>

<p align="center">
  <img src="./assets/logo.png" alt="Wiki Knowledge Compiler Logo" width="100%" />
</p>

## Table of Contents

- [Recent Updates](#-recent-updates)
- [Features](#-features)
- [Localizations](#-localizations)
- [Quick Start](#-quick-start)
- [Usage](#-usage)
  - [Adding Documents](#-adding-documents)
  - [Example: From Document to Knowledge](#-example-from-document-to-knowledge)
- [Knowledge Graph](#-knowledge-graph)
- [Confidence & Smart Tracking](#-confidence--smart-tracking)
- [Browse with Obsidian](#-browse-your-wiki-with-obsidian)
- [How Setup Works](#%EF%B8%8F-how-setup-works)
- [Other Ways to Install](#-other-ways-to-install)
- [Contributing](#-contributing)
- [License](#-license)

---

## 🆕 Recent Updates

### v0.2.0 — Knowledge graph, lifecycle management, and wiki health
- Added knowledge graph that extracts named entities (people, projects, technologies, decisions) during ingestion and connects them with typed relationships in `.data/entities/`
- Added typed relationships between concepts with semantic meaning: `extends`, `contradicts`, `supersedes`, `depends-on`, `generalizes`, `component-of`
- Added confidence scoring to every page (`high`/`medium`/`low`) with `verified` dates and source `authority` tracking (`primary`/`secondary`/`commentary`)
- Added supersession — when new info replaces old, pages are linked with `supersedes`/`superseded_by` instead of being silently overwritten
- Added graph-aware queries to `/wiki` — walks entity relationships alongside keyword/semantic search to find connections that text search misses
- Added `/wiki-audit` command to check wiki health: broken links, orphan pages, stale content, missing frontmatter, citation mismatches, contradictions, entity graph issues — auto-fixes what it can, reports the rest
- Added privacy filter that auto-strips API keys, tokens, passwords, and PII from derived content during ingestion (raw sources are never modified). Pages tagged with `visibility: public/private/internal`
- Added contradiction resolution — when sources disagree, both sides are scored by recency, authority, and evidence count. Stronger claims supersede weaker ones; close calls flagged for human review
- Added retention decay — confidence decays over time unless pages are accessed or reinforced. Each page type has its own half-life (concepts: 180 days, insights: 60 days, drafts: 30 days)
- Added crystallization — when `/wiki` synthesizes 3+ sources, it auto-generates a digest page capturing the question, answer, sources, key connections, and lessons
- Added quality scoring — every page scored 0–100 across 5 dimensions (frontmatter, citations, freshness, relationships, content). Grades A–F. Pages below 60 flagged during `/wiki-update` and `/wiki-audit`
- Improved `/wiki-ingest` to extract entities, apply privacy filtering, and set confidence/authority on summaries
- Improved `/wiki-update` to apply retention decay and compute quality scores after rebuilding indexes
- Improved `/wiki-insights` to read entity graph for structural connections alongside topic and concept pages

> **Upgrading from v0.0.x?** Run `/wiki-setup` again to initialize the new `.data/entities/` directory. Your existing wiki pages will continue to work — new fields (`confidence`, `verified`, `entities`, typed `related`) are added incrementally as you ingest new documents or run `/wiki-update`. No migration needed.

### v0.0.x — Foundation
- 📅 **Scheduled updates** — `/wiki-schedule` auto-runs `/wiki-update` on a cron schedule via remote agents.
- 🔀 **Mixed ingest** — `/wiki-ingest` accepts files + URLs in a single command.
- 🔄 **Auto-ingest on update** — `/wiki-update` detects and ingests new files dropped into `raw/`.
- 📝 **Simpler commands** — `/wiki-query` renamed to `/wiki`.
- 📁 **Flat structure** — Removed `knowledge-base/` wrapper; `raw/` and `wiki/` live at project root.
- 📥 **Batch ingestion** — Folders, multiple files, and glob patterns.
- 🔍 **Search tiers** — Grep for small projects, qmd for medium/large.
- 🔮 **Obsidian integration** — Auto-install and setup during `/wiki-setup`.
- ⚙️ **Cross-platform auto-install** — Node.js, qmd, Obsidian on macOS, Linux, and Windows.

---

## ✨ Features

| Feature | Description |
|---------|-------------|
| **Knowledge graph** | Extracts people, projects, technologies, decisions as connected entities |
| **Typed relationships** | Concepts and entities linked with semantic meaning (`extends`, `contradicts`, `depends-on`) |
| **Confidence tracking** | Every page rated `high`/`medium`/`low` with verified dates |
| **Retention decay** | Confidence decays over time unless reinforced — each page type has its own half-life |
| **Quality scoring** | Every page scored 0–100 across 5 dimensions, grades A–F |
| **Crystallization** | Multi-source answers auto-distilled into reusable digest pages |
| **Wiki audit** | `/wiki-audit` finds broken links, stale content, contradictions and auto-fixes them |
| **Privacy filter** | Auto-strips API keys, tokens, passwords, PII from derived content |
| **Contradiction resolution** | Conflicting claims scored by recency/authority/evidence, resolved or flagged |
| **Supersession** | Old claims linked to replacements, never silently deleted |
| **Graph-aware queries** | `/wiki` walks entity relationships, not just keywords |
| **Multi-format ingest** | Files, folders, URLs, glob patterns, pasted text |
| **Auto summaries** | One-page summary per source with key takeaways |
| **Concept extraction** | 3-10 atomic concepts per source, auto-deduplicated |
| **Smart search** | Grep (small), qmd hybrid search (medium), qmd MCP (large) |
| **Scheduled updates** | Cron-based auto-ingestion via remote agents |
| **Obsidian integration** | Visual graph browsing, backlinks, live sync |
| **100% offline** | Plain markdown files, no cloud, no database |
| **Cross-platform** | macOS, Linux, Windows with auto-install |

---

## 🌐 Localizations

<p align="center">
  🇬🇧 <strong>English</strong> &nbsp;·&nbsp;
  <a href="./README_VN.md">🇻🇳 Tiếng Việt</a> &nbsp;·&nbsp;
  <a href="./README_CN.md">🇨🇳 简体中文</a> &nbsp;·&nbsp;
  <a href="./README_KR.md">🇰🇷 한국어</a> &nbsp;·&nbsp;
  <a href="./README_JP.md">🇯🇵 日本語</a> &nbsp;·&nbsp;
  <a href="./README_DE.md">🇩🇪 Deutsch</a> &nbsp;·&nbsp;
  <a href="./README_FR.md">🇫🇷 Français</a> &nbsp;·&nbsp;
  <a href="./README_RU.md">🇷🇺 Русский</a>
</p>

---

## 🚀 Quick Start

Three steps. Everything installs automatically — you don't need to set up anything yourself.

**Step 1 — Install the plugin:**
```
/plugin marketplace add bradduy/wiki-knowledge-compiler
```
After adding, configure it in the marketplace:
1. Run `/plugin` → select **Marketplace** tab
2. Find **wiki-knowledge-compiler** and open it
3. Click **Enable** and choose the scope — pick the project folder where you want your wiki to live
4. Confirm — this grants the plugin permission to create files and run commands in that folder

<p align="center">
  <img src="./assets/user_scope_setting.png" alt="Enable scope setting" width="100%" />
</p>

**Step 2 — Run setup:**
```
/wiki-setup
```
Pick your project size (small, medium, or large). The plugin handles the rest — it installs any tools you need, builds your wiki folder, and even offers to set up [Obsidian](https://obsidian.md) so you can browse your wiki visually.

<p align="center">
  <img src="./assets/obisidian.gif" alt="Obsidian visual browsing" width="100%" />
</p>

**Step 3 — Add your first document:**
```
/wiki-ingest ~/Documents/my-article.md
```

Done. Your wiki is live.

---

## 🎯 Usage

Each command tells you what to try next, so you always know the next step.

| Command | What it does |
|---------|-------------|
| `/wiki-setup` | ⚙️ One-time setup (installs everything for you) |
| `/wiki-ingest` | 📥 Add documents — one file, a whole folder, or a URL |
| `/wiki` | 🔍 Ask a question, get an answer with sources |
| `/wiki-insights` | ✨ Find patterns and connections across your sources |
| `/wiki-update` | 🔄 Sync wiki (ingest new raw files + refresh indexes) |
| `/wiki-schedule` | 📅 Schedule /wiki-update to run automatically |
| `/wiki-audit` | 🩺 Check wiki health — fix broken links, detect contradictions, score quality |

**Recommended order:** `/wiki-setup` → `/wiki-ingest` → `/wiki` → `/wiki-insights` → `/wiki-update` → `/wiki-audit` → `/wiki-schedule`

### 📥 Adding Documents

You can add documents in many ways:

```bash
# One file
/wiki-ingest ~/Documents/article.md

# A whole folder (scans everything inside)
/wiki-ingest ~/Documents/research/

# Multiple files at once
/wiki-ingest ~/notes/meeting1.md ~/notes/meeting2.md

# All PDFs in a folder
/wiki-ingest ~/papers/*.pdf

# A URL
/wiki-ingest https://example.com/interesting-article

# Or just paste text directly after the command
```

For folders and multiple files, the plugin shows progress as it works through each one.

### 🧪 Example: From Document to Knowledge

Say you add an article about climate change:

```
/wiki-ingest ~/Articles/ipcc-summary-2023.md
```

The plugin creates:

| What | Description |
|------|------------|
| **Summary** | A one-page overview of the article with confidence rating |
| **Concepts** | Pages for key ideas like "carbon budget" and "tipping points" |
| **Entities** | Nodes for real things — "IPCC", "Paris Agreement", "Dr. Smith" |
| **Topic page** | A "climate science" page linking everything together |

Now add a second article about energy policy. The plugin notices both sources discuss carbon budgets and links them automatically — concepts get typed relationships (`extends`, `contradicts`, `depends-on`), and entities connect across sources.

When you ask:
```
/wiki How do carbon budgets affect energy policy?
```

The plugin doesn't just search by keywords. It walks the **knowledge graph** — starting from the "carbon budget" concept, following its relationships to connected entities and topics — and finds connections that keyword search would miss. You get an answer that draws from **both** sources, with links to exactly where each fact came from.

---

## 🧠 Knowledge Graph

Every document you add doesn't just create pages — it builds a **knowledge graph**. The plugin extracts named entities (people, projects, technologies, decisions) and connects them with typed relationships.

```
Redis ──uses──→ Auth Service ──maintained-by──→ Brad
  │                                              │
  └──depends-on──→ PostgreSQL           owns ←───┘
                        │                Auth Migration
                        └──replaces──→ MySQL
```

This means when you ask "what's the impact of upgrading Redis?", the plugin walks outward from Redis through `uses`, `depends-on`, and `maintained-by` edges — finding every connected service, person, and decision. Not just pages that mention "Redis" by name.

**What gets extracted automatically:**

| Entity type | Examples |
|-------------|---------|
| People | "Brad Duy", "Andrej Karpathy" |
| Projects | "Auth Migration", "API Redesign" |
| Technologies | "Redis", "PostgreSQL", "Kubernetes" |
| Libraries | "React", "PyTorch", "Express" |
| Decisions | "Switch to PostgreSQL", "Use microservices" |
| Organizations | "Platform Team", "Anthropic" |

---

## 📊 Confidence & Smart Tracking

Not all knowledge is equally reliable. Every page in your wiki carries metadata that helps you trust it:

- **Confidence** — `high`, `medium`, or `low` based on how many sources support it
- **Verified date** — when the content was last confirmed accurate
- **Authority** — whether a source is `primary` (original research), `secondary` (review), or `commentary` (opinion)
- **Typed relationships** — connections between concepts aren't just links, they have meaning: `extends`, `contradicts`, `supersedes`, `depends-on`

When two sources disagree, the plugin flags the **contradiction** and links both sides — so you see the conflict instead of getting silently wrong answers.

When newer information replaces older claims, the old page is marked as **superseded** with a link to what replaced it. Nothing is deleted — you can always trace the history.

---

## 🔮 Browse Your Wiki with Obsidian

During setup, you'll be asked if you want [Obsidian](https://obsidian.md) — a free app for browsing your wiki visually. It's installed automatically if you say yes.

Open your project folder in Obsidian and you get:

- 🕸️ **Graph View** — a visual map of all your pages and how they connect
- 🔙 **Backlinks** — click any page to see everything that links to it
- 🔍 **Search** — find anything across your entire wiki
- ⚡ **Live sync** — new pages from `/wiki-ingest` appear in Obsidian right away

This is optional. Your wiki works perfectly with just the commands.

---

## 📁 What Gets Created

```
your-project/
  📄 raw/           Your original documents (never modified)
  📚 wiki/          All generated pages (summaries, concepts, topics, insights)
  .data/            Internal working data (hidden)
```

No wrapper folder. `raw/` and `wiki/` live right in your project. If `raw/` already exists, setup just adds `wiki/` next to it.

All files are plain Markdown. Open them with any text editor, Obsidian, or VS Code.

---

## 💡 Tips

- 🌱 **Start small.** Add 2–3 sources and try `/wiki` before adding more.
- 🎯 **Be specific.** "What does source X say about Y?" works better than vague questions.
- ✨ **Try `/wiki-insights`** after adding several sources on a topic — it finds patterns you might miss.
- 🔒 **Your sources are safe.** The plugin never changes your original files.

---

## ⚙️ How Setup Works

`/wiki-setup` handles everything automatically based on your project size:

| Size | Search | What gets installed |
|------|--------|-------------------|
| **Small** | Built-in | Nothing extra needed |
| **Medium** | Smart search ([qmd](https://github.com/tobi/qmd)) | Node.js + qmd (auto-installed) |
| **Large** | Fastest search | Node.js + qmd + MCP server (auto-installed and configured) |

The plugin detects your operating system (macOS, Linux, Windows) and installs everything using the right method for your machine. If anything fails, it falls back to built-in search — you're never stuck.

---

## 📦 Other Ways to Install

### From the Claude Code Marketplace (easiest)

```
/plugin marketplace add bradduy/wiki-knowledge-compiler
```
After adding, configure it in the marketplace:
1. Run `/plugin` → select **Marketplace** tab
2. Find **wiki-knowledge-compiler** and open it
3. Click **Enable** and choose the scope — pick the project folder where you want your wiki to live
4. Confirm — this grants the plugin permission to create files and run commands in that folder

<p align="center">
  <img src="./assets/user_scope_setting.png" alt="Enable scope setting" width="100%" />
</p>

### Clone as a standalone project

```bash
git clone https://github.com/bradduy/wiki-knowledge-compiler.git
cd wiki-knowledge-compiler
```

Open in Claude Code and run `/wiki-setup`.

### Add to an existing project

```bash
git clone https://github.com/bradduy/wiki-knowledge-compiler.git /tmp/wkc
cp -r /tmp/wkc/.claude/ your-project/.claude/
cp -r /tmp/wkc/.claude-plugin/ your-project/.claude-plugin/
cp -r /tmp/wkc/{agents,skills,templates} your-project/
cp /tmp/wkc/CLAUDE.md your-project/CLAUDE.md
cp /tmp/wkc/.data/wiki.config.md your-project/.data/wiki.config.md
bash /tmp/wkc/scripts/init-kb.sh your-project/knowledge-base
rm -rf /tmp/wkc
```

Then run `/wiki-setup`.

---

## 📌 Good to Know

- 📡 **Works offline.** Everything is on your computer.
- 📄 **Plain text files.** No database or special software needed to read your wiki.
- 📑 **PDFs supported.** Complex layouts (tables, multi-column) may not extract perfectly.
- 👤 **Designed for personal use.** Single user.
- 🕐 **Want version history?** Run `git init` in your knowledge base folder.

---

## 🤝 Contributing

Want to improve the plugin? See the [contributing guide](.github/CONTRIBUTING.md) or open an issue.

## 📄 License

MIT