By zhaow-de
LLM-compiled knowledge base. Natural language routing, collector catalogs, topic archive lifecycle, inventory tracking, dataset manifests, collection ingestion for external wikis/spec repos, truth-seeking audits, session resume, concurrent-safe derived indexes, topic-isolated wikis, parallel multi-agent research, thesis-driven investigation, repo assessment, Obsidian dual-linking, retardmax mode, and --min-time sustained research.
Archive or restore whole topic wikis so old interests stay preserved but out of default context. Archived topics move under HUB/topics/.archive and are hidden from normal semantic and maintenance workflows.
Assess a local repo against the active wiki's research body and the broader market. Gap analysis, opportunities, competitive landscape.
Truth-seeking umbrella audit for llm-wiki. Combines active wiki maintenance, output drift checks, provenance review, and fresh research when trust is in doubt.
Find, catalog, deduplicate, and optionally inventory bounded sets of artifacts, examples, resources, memes, tools, entities, media, or other collectible things.
Compile raw sources into wiki articles. Synthesizes, cross-references, and organizes active knowledge.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
An opinionated implementation of Karpathy's LLM Wiki concept — a Claude Code plugin that turns Claude into the compiler and query engine for a personal, LLM-maintained knowledge base. Ready for using Obsidian as an optional content viewer.
This README has two audiences, and it is split to match:
For wiki users — install the plugin and drive it with /pkb commands to research, ingest, compile, and query a personal knowledge base. Installing means cloning this repo once; after that, nothing inside the repo needs to be read or changed.
For plugin developers — change the prompts, command specs, scripts, and tests that are this repo.
Bookmarks rot. Notes go stale. "I read something about this once" is not knowledge.
pkb turns Claude Code into a compiler for the things worth knowing. Point it at the world — a question, a URL, a pile of PDFs — and it researches, reads, cross-checks, and writes the result into a wiki that stays queryable forever. The articles are never written by hand; Claude writes them. The job here is to ask.
The mental model is one sentence: raw sources are the source code, Claude is the compiler, the wiki is the executable. Everything else follows from that.
raw/ ← sources, ingested once, never edited (the source code)
wiki/ ← articles Claude synthesizes from raw/ (the executable)
output/ ← reports, slides, plans on request (what ships)
The rule: never hand-edit wiki/. It is compiled — Claude rebuilds it from raw/. When an article is wrong, fix the source (ingest a better one, /pkb-retract a bad one) and recompile. Editing the compiled article is pointless: the next compile overwrites it.
Each topic is its own isolated wiki (crypto-quant, market-microstructure, defi…). Topics sit side by side under one hub but never bleed into each other — research stays scoped and context stays clean:
~/llm-wiki-data/ # Hub — registry only, no content
├── wikis.json # Registry of all topic wikis
├── _index.md # Lists topic wikis with stats
├── log.md # Global activity log
└── topics/
└── <topic>/ # One isolated wiki per topic
├── raw/ # Immutable sources
├── wiki/ # Compiled, cross-linked articles
├── output/ # Generated artifacts + projects
├── inventory/ # Lazy: durable tracking records
├── datasets/ # Lazy: manifests for large/external data
├── _index.md
├── config.md
└── log.md
_index.md files are a derived cache rebuilt from frontmatter — also never hand-edited. For the full design — components, flows, and the rationale behind them — see docs/llm-agent-architecture.md.
pkb installs into Claude Code from a local checkout of this repo. Cloning it is a one-time install step — afterward the repo never needs opening again:
git clone https://github.com/zhaow-de/llm-wiki.git
claude plugin marketplace add /absolute/path/to/llm-wiki
claude plugin install pkb@llm-wiki
# restart Claude Code or run /reload-plugins in Claude Code
Then, in Claude Code, run once to register the /pkb-* slash commands:
/pkb:install-commands
After this, invoke wiki commands as /pkb, /pkb-research, /pkb-ingest, etc. Re-run /pkb-install-commands after plugin updates to sync any new or removed commands.
Confirm it loaded with /pkb status. Claude Desktop picks up the same plugin once Claude Code has installed it.
The wiki lives at ~/llm-wiki-data/ by default; change the hub path with /pkb config hub-path "<new-path>". To move to a newer version later, run git -C /path/to/llm-wiki pull then claude plugin update pkb@llm-wiki (restart or /reload-plugins).
npx claudepluginhub zhaow-de/llm-wiki --plugin pkbBuild and maintain an LLM-curated personal knowledge base in your project — Andrej Karpathy's LLM Wiki pattern, designed to scale to thousands of pages without becoming a context bottleneck. Now with an optional compiled graph layer for typed, provenance-backed relationships.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Intelligent draw.io diagramming plugin with AI-powered diagram generation, multi-platform embedding (GitHub, Confluence, Azure DevOps, Notion, Teams, Harness), conditional formatting, live data binding, and MCP server integration for programmatic diagram creation and management.
Persistent file-based planning for AI coding agents. Crash-proof markdown plans (task_plan.md, findings.md, progress.md) that survive context loss and /clear, with an opt-in completion gate and multi-agent shared state. Manus-style. Works with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and 60+ agents via the SKILL.md standard. Includes Arabic, German, Spanish, and Chinese (Simplified and Traditional).
Complete creative writing suite with 10 specialized agents covering the full writing process: research gathering, character development, story architecture, world-building, dialogue coaching, editing/review, outlining, content strategy, believability auditing, and prose style/voice analysis. Includes genre-specific guides, templates, and quality checklists.