llm-wiki

llm-wiki-skills

中文版 | English

Build a durable, compounding personal knowledge base as plain Markdown — works with Claude Code, Copilot, Codex, and any SKILL.md-compatible agent.

Features

• Multi-platform support: Works with Claude Code, GitHub Copilot, Codex, and other SKILL.md-compatible agents
• Canonical single-source: Skill logic maintained in one location, synced across all platform manifests
• Personal knowledge base: Build a durable, compounding wiki as plain Markdown files with cross-links, source tracking, and uncertainty markers

Architecture: Three Layers

LLM Wiki is not software — it's a working paradigm with three-layer separation:

┌─────────────────────────────────────────────────────────────┐
│  Layer 3: Convention Layer (SCHEMA.md)                     │
│  Tells LLM how to maintain the wiki                      │
├─────────────────────────────────────────────────────────────┤
│  Layer 2: Persistent Knowledge Layer                      │
│  (entities/ concepts/ synthesis/ comparisons/ queries/)     │
│  LLM-owned and maintained — primary workspace             │
├─────────────────────────────────────────────────────────────┤
│  Layer 1: Raw Source Layer (raw/)                        │
│  Immutable ground truth — LLM read-only, always retained  │
└─────────────────────────────────────────────────────────────┘

vs. RAG: RAG re-discovers knowledge from scratch every query. LLM Wiki accumulates knowledge over time — the knowledge network grows continuously.

Quick Install

Install by chatting with your agent:

install skill https://github.com/yugasun/llm-wiki-skills

Claude Code

/plugin marketplace add yugasun/llm-wiki-skills
/plugin install llm-wiki-skills@llm-wiki

GitHub Copilot / Other agents

npx skills install yugasun/llm-wiki-skills llm-wiki

Repository Layout

llm-wiki-skills/
├── skills/llm-wiki/           # Canonical skill source
│   ├── SKILL.md               # Main skill definition
│   └── scripts/               # Helper scripts
├── .claude-plugin/            # Claude Code plugin manifests
└── scripts/                   # Maintenance scripts
    ├── sync-skill-layout.sh  # Sync skill to all platform dirs
    └── verify-skill-layout.sh # Validate layout integrity

What You Can Do

Once installed, the skill enables these workflows:

Task	Prompt Example	What Happens
Initialize	initialize a wiki at ~/mywiki	Creates SCHEMA.md, index.md, log.md + directory structure
Ingest	ingest https://example.com/article	Fetches content → saves to raw/ → synthesizes linked page
Query	what does my wiki say about transformers?	Searches accumulated pages, compiles answer with sources
Lint	lint my wiki	Checks for broken links, orphan pages, missing frontmatter
Orient	orient my wiki	Reads schema, index, and recent log before any operation

Designed for Obsidian, VS Code, and plain text tools — no proprietary format lock-in.

Usage

Start a conversation with your agent and describe what you need:

initialize a personal knowledge base in ~/wiki
ingest these URLs into my wiki: https://example.com/article, https://example.com/paper
what does my wiki say about transformer architectures?
lint my wiki for broken links
what's in my wiki about [topic]?

The skill resolves wiki root in order: WIKI_PATH env > ~/.llm-wiki > current directory with required files > ~/wiki

Prerequisites

• Node.js 18+ — only for npx-based install
• Any SKILL.md-compatible agent: Claude Code, GitHub Copilot, Codex, etc.

References

• Karpathy's LLM Wiki — inspiration
• LLM Wiki Architecture — detailed design

License

MIT — see LICENSE