mimimaster/piwin/llm-wiki

Claude Code and agent skill by mimimaster/piwin

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Knowledge management, document synthesis, and concept graph maintenance following Andrej Karpathy's LLM-Wiki pattern. Use when ingesting articles/docs/papers, creating or querying wiki concepts, maintaining cross-links, or performing wiki health checks.

Install

npx vskill@latest install mimimaster/piwin/llm-wiki

SKILL.md

Andrej Karpathy LLM-Wiki Pattern

"The central idea of LLM-Wiki is that the LLM acts as the encyclopedist and curator of a personal, compounding knowledge base formatted in plain Markdown with [[wikilinks]]."

A structured knowledge architecture where LLMs ingest raw information, distill it into interconnected concept pages, maintain a master index, record an append-only audit trail, and continuously verify graph health.


Architecture Overview

~/.piwin/wiki/
├── INDEX.md # Master catalog, domain taxonomy, concept lookup table
├── LOG.md # Append-only chronological audit log of all updates & ingests
├── raw/ # Immutable raw sources (papers, notes, transcripts, clips)
└── concepts/ # Interconnected Markdown notes with YAML frontmatter + [[wikilinks]]
├── transformer-architecture.md
├── attention-mechanism.md
└── ...
The Three Layers
  1. Raw Sources (`raw/`): Read-only raw documents, articles, transcripts, or web clips.
  2. Distilled Wiki (`concepts/`): High-signal, synthesized encyclopedia pages in Markdown.
  3. Control & Audit (`INDEX.md` + `LOG.md`): Table of contents and journal of all changes.

The 3 Core Operations

┌─────────────────────────────────────────────────────────────┐
│ LLM-Wiki Lifecycle │
├──────────────────────────────┬──────────────────────────────┤
│ 1. Ingest (消化与提炼) │ 2. Query (检索与概念联想) │
│ • Read raw source • Read INDEX.md first │
│ • Cross-check INDEX.md • Retrieve target concepts │
│ • Update 1–5 concept notes│ • Follow [[wikilinks]] │
│ • Append to LOG.md • Synthesize cited answer │
├──────────────────────────────┴──────────────────────────────┤
│ 3. Lint (巡检与维护) │
│ • Scan for broken [[wikilinks]] │
│ • Identify orphan concepts with zero in/out links │
│ • Ensure every concept is registered in INDEX.md │
└─────────────────────────────────────────────────────────────┘

1. Ingest Workflow (消化提炼)

When given new documents, text, notes, or technical materials:

  1. Read `INDEX.md` First:
  • Inspect the current catalog to see what concepts already exist.
  • Avoid creating duplicates or conflicting definitions.
  1. Decompose & Synthesize:
  • Do not simply dump the entire raw file into a page.
  • Break down ideas into atomic, durable concepts (e.g. [[Direct Preference Optimization]], not Paper Summary March 2024).
  • Create or update 1 to 5 interconnected concept pages.
  1. Format Each Concept Page:

```markdown


title: Concept Name aliases: [Acronym, Alt Name] tags: [domain, sub-domain] summary: One-sentence concise definition. updated: YYYY-MM-DD


# Concept Name

## Definition & Core Intuition Clear, concise explanation of the concept.

## Key Mechanisms & Components Structured breakdown with cross-references like [[Related Concept]].

## Relationships & Connections

  • Extends: [[Prior Concept]]
  • Applied in: [[Application Concept]]
  • Contrasted with: [[Alternative Concept]]

```

  1. Maintain Cross-links (`[[wikilinks]]`):
  • Use [[Concept Name]] or [[Target Slug|Display Label]].
  • Every concept page must link to at least 1 parent or related concept.
  1. Update Index & Log:
  • Use tool wiki_write to save concepts, which automatically records entries into INDEX.md and LOG.md.

2. Query Workflow (检索与联想)

When answering questions using the knowledge base:

  1. Check `INDEX.md`:
  • Call wiki_read(name: "INDEX") to locate matching concepts.
  1. Read Target Concepts:
  • Call wiki_read(name: "...") for the most relevant concepts.
  1. Traverse Connected Outlinks:
  • If deeper context is required, read the outlinks discovered in the target concept.
  1. Synthesize & Cite:
  • Formulate the response incorporating explicit [[wikilinks]] so the user can explore the concept network.

3. Lint Workflow (巡检与健康维护)

Periodically or after bulk ingestions:

  1. Run wiki_lint:
  • Inspects ~/.piwin/wiki/concepts/.
  • Identifies broken links ([[Target]] pointing to non-existent concept files).
  • Identifies orphan concepts (zero backlinks and omitted from INDEX.md).
  1. Fix detected issues:
  • For broken links: create the stub concept or correct the typo.
  • For orphans: link them from INDEX.md and related parent concepts.

Tools Reference

  • wiki_read: Read a concept page, INDEX, or LOG.
  • wiki_write: Create or update a concept page with automatic INDEX.md and LOG.md sync.
  • wiki_lint: Run comprehensive graph health check.
  • /wiki: Slash command for terminal / chat operations (/wiki lint, /wiki log).

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/mimimaster/piwin/llm-wiki/badge)](https://verified-skill.com/skills/mimimaster/piwin/llm-wiki)

Something wrong with this skill or its page? Tell us.