What makes llm-wiki SKILL inspired by Karpathy notable?

2K downloads. Trust evidence available

How should llm-wiki SKILL inspired by Karpathy be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

agentCLAWHUBUnverified

llm-wiki SKILL inspired by Karpathy

Use when an AI Agent (Claude Code, Codex, OpenClaw, or similar) needs to operate an llm-wiki knowledge base: ingest source files into Markdown wiki pages, answer questions from wiki/index.md and linked pages, run agent-bridge status/lint/link/relink/merge/query/index/Zotero relocation tasks, preserv

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

2.0k

Updated

Oct 9, 2026

Version

1.5.3

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 2K downloads reported by the source. Last updated 10/9/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 9, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 9, 2026
Adoption signal: 2K downloadsadoption · observed Oct 9, 2026
Latest release: 1.5.3release · observed Sep 9, 2026
Handshake status: UNKNOWNsecurity

Install and run

Setup complexity: low.

clawhub skill install s173ycbfz08xfrrqb7xrepkwa1842rdz:041-llm-wiki

Install using `clawhub skill install s173ycbfz08xfrrqb7xrepkwa1842rdz:041-llm-wiki` in an isolated environment before connecting it to live workloads.
No published capability contract is available yet, so validate auth and request/response behavior manually.
Review the upstream CLAWHUB listing at https://clawhub.ai/nemo4110/041-llm-wiki before using production credentials.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-nemo4110-041-llm-wiki/snapshot"

Documentation

CLAWHUB

160,000 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: llm-wiki
version: "1.6.0"
description: "Use when an AI Agent (Claude Code, Codex, OpenClaw, or similar) needs to operate an llm-wiki knowledge base: ingest source files into Markdown wiki pages, answer questions from wiki/index.md and linked pages, run agent-bridge status/lint/link/relink/merge/query/index/Zotero relocation tasks, preserve provenance and temporal metadata, or use Zotero as a literature-discovery layer."
---

# LLM-Wiki

## Core Principle

Treat the LLM as the programmer and the wiki as the codebase. The user provides materials and judgment; the Agent extracts durable knowledge, preserves provenance, maintains links, and keeps the Markdown wiki structurally consistent.

Keep this file as the operational skill. Use `README.md` for user-facing overview, `AGENTS.md` / `CLAUDE.md` for the full protocol, and `ROADMAP.md` for project plans.

## Start Every Wiki Task

1. Read `AGENTS.md` or `CLAUDE.md` when the task touches wiki behavior, source handling, or ingest/query protocol.
2. Use the project Python: `.venv\Scripts\python.exe` on Windows, `.venv/bin/python` on Unix, or `uv run python` when configured.
3. Run `<PY> scripts/agent-bridge.py check` before wiki operations. If it reports missing dependencies, state the exact blocker and continue only with tasks that do not require the unavailable runtime.
4. Protect `sources/`: never write Agent-generated summaries, drafts, or speculative content there. Only user-provided files or verified network fetches or Zotero MCP material may be source assets.
5. Check `git status --short` before editing. Do not revert user changes.
6. `wiki/*` is gitignored by default; when using ripgrep for wiki/source discovery, pass `--no-ignore` or read the files directly so ignored knowledge pages are not silently omitted.

## Choose the Work Mode

| Task | Use | Notes |
| --- | --- | --- |
| Status, lint, link discovery, relink, merge, semantic query, embedding index, Zotero sync planning | `scripts/agent-bridge.py` | Algorithmic/read-only tasks. Prefer dry-run before writing. |
| Ingest source material | Protocol mode | Requires LLM judgment: read source, extract metadata, create/update pages. |
| Answer wiki questions | Protocol mode | Read `wiki/index.md`, relevant pages, and link neighbors; synthesize with `[[PageName]]` citations. |
| Apply relation updates | Hybrid | Let `agent-bridge.py` discover candidates, then review and merge only safe changes. |
| Plan/apply Zotero attachment relocation | Hybrid | Use `zotero-relocate`; review dry-run, require local authorization for apply, and never bypass the Phase 0/API gate. Read the managed root from `config.yaml` `zotero_relocation.root` (prefer a cloud-synced path for cross-device use); `--root` is only a one-off override. |

Agent Bridge quick commands:

```bash
<PY> scripts/agent-bridge.py check
<PY> scripts/agent-bridge.py status
<PY> scripts/agent-bridge.py lint
<PY> scripts/agent-bridge.py link --source "PageName" --mode light
<PY> scripts/

README.md

# LLM-Wiki Skill

[简体中文](docs/README.cn.md) | English

Claude Code SKILL implementation of [Karpathy's llm-wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f).

> **Core Philosophy**: LLM as programmer, Wiki as codebase, User as product manager.

## Why SKILL Form?

We chose the SKILL form because it brings these advantages:

- **Zero deployment** — No services to run, no databases to configure; works the moment you clone the repository
- **Native integration** — Direct command execution via Claude Code, no middleware or protocol translation needed
- **Plain-text data** — Pure Markdown files, git-native, with no proprietary formats or vendor lock-in
- **Editor freedom** — Use Obsidian, VS Code, or any text editor you prefer
- **Minimal footprint** — A small Python helper/CLI layer around a plain Markdown wiki, keeping complexity low

## Quick Start

### Option A — One-command install with `uv` (Recommended, no clone needed)

Install the CLI as an isolated tool directly from the repository and scaffold a
knowledge base anywhere — you never run `git clone` yourself:

```bash
# 1. Install uv once (https://docs.astral.sh/uv/)
#    Windows:  powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
#    macOS/Linux: curl -LsSf https://astral.sh/uv/install.sh | sh

# 2. Install llm-wiki straight from GitHub (uv fetches it for you)
uv tool install git+https://github.com/Nemo4110/llm-wiki.git
# or run without installing:  uvx --from git+https://github.com/Nemo4110/llm-wiki.git llm-wiki --help

# 3. Create a knowledge base in any directory
llm-wiki init my-kb
cd my-kb

# 4. Drop materials into sources/ and let your agent ingest them
llm-wiki status
```

`llm-wiki init` materializes `wiki/`, `sources/`, `AGENTS.md`, `CLAUDE.md`, and
`config.yaml.example` from templates bundled inside the installed package — so
there is no checkout to manage at all. It also creates or non-destructively extends
`.gitignore` so private/local state such as `var/`, `temp/`, `.mcp.json`, and
`config.yaml` cannot be committed accidentally.

> Upgrading later is one command: `uv tool upgrade llm-wiki` (re-fetches the
> latest commit from the default branch).

### Option B — Clone and install from source

For development or to hack on the SKILL itself:

```bash
git clone https://github.com/Nemo4110/llm-wiki.git
cd llm-wiki
```

The CLI tool currently supports Python 3.12-3.13. The actively verified local development matrix is:

| Platform | Python | Status |
|----------|--------|--------|
| Windows | 3.13 | Verified |
| Windows | 3.12 | Supported target |
| Linux/macOS | 3.12-3.13 | Supported target, not the primary local verification platform |

Python 3.8-3.11 are not part of the current support matrix. Choose your preferred installation method:

#### Using uv (Recommended if you have uv)

```bash
# Create virtual environment and install the package (editable)
uv venv
uv pip install -e .

# Activate (Windows)
.venv\Scripts\activate
# Or Linux/macOS
source .v

sources/README.md

# Sources 目录

> 放入你想让 Wiki 吸收的所有原始资料。

> **警告**：`sources/` 目录中的文件必须是原始资料的真实副本。
> 如果你发现某个文件的内容像是 AI 生成的摘要而非原始文档，请立即删除并重新获取。
> Agent 被严格禁止向此目录写入任何 LLM 生成的内容。

## 支持的格式

- Markdown (.md)
- 文本文件 (.txt)
- PDF (.pdf) — **需要安装安全版本的 pdfplumber>=0.11.8 和 pdfminer.six>=20251107**
- 代码文件 (.py, .js, etc.)
- 图片 (.png, .jpg) — 需要 vision 能力
- 网页链接 (.url 或粘贴内容)

### PDF 处理注意事项

**安全要求**：
- 必须使用安全版本：pdfplumber >= 0.11.8，pdfminer.six >= 20251107
- **原因**：CVE-2025-64512 漏洞可导致任意代码执行
- **避免**：直接使用系统工具（如 pdftoppm）读取 PDF，这会触发依赖错误

**依赖安装**：

```bash
# 安装安全版本的 PDF 处理库
pip install pdfplumber>=0.11.8 pdfminer.six>=20251107
```

这些依赖已在 `src/requirements.txt` 中定义，安装项目依赖时会自动安装。

## 使用流程

1. **放入资料**：复制或下载文件到此目录
2. **告诉 Agent**：「请摄入新资料」或 `/wiki-ingest 文件名`
3. **Agent 处理**：
   - 读取内容
   - 提取要点
   - 更新 wiki 页面
   - 记录日志

### 验证 Agent 操作

llm-wiki 协议要求 Agent 在下载文件后、摄入前，**主动验证文件内容是否与你的请求一致**。

Agent 会检查：
1. **文件可读性**：文件是否损坏或无法解析
2. **错误页面检测**：内容是否包含 `404`、`Access Denied`、`Subscribe` 等关键词
3. **格式一致性**：文件头是否与扩展名匹配（如 PDF 应以 `%PDF` 开头）
4. **元数据比对**：提取标题、作者、DOI/arXiv ID，与你提供的描述进行比对

**如果验证失败**，Agent 会：
- **停止摄入**，不创建任何 wiki 页面
- 向你报告提取到的实际元数据 vs 你的期望
- 等待你的确认或提供正确的 URL

**如果你怀疑 Agent 可能伪造了来源文件**，还可以人工检查：

1. **工具调用痕迹**：Agent 的回复中是否包含 `curl`、`wget`、`playwright` 等实际网络请求的命令和输出？
2. **文件内容**：打开文件查看，是否包含原始文档的完整内容（而非摘要）？
3. **文件时间戳**：文件创建时间是否与你提供 URL 的时间一致？

**如果发现伪造内容**：
- 立即删除伪造文件
- 从原始出处重新获取
- 在 `log.md` 记录问题，以便改进协议

## 文件命名建议

```
YYYY-MM-DD-描述.扩展名
# 例如：
2026-04-10-karpathy-llm-wiki-gist.md
2026-04-09-transformer-paper.pdf
```

## Git 管理说明

**默认情况下，`sources/` 中的文件不会被 Git 追踪**（已加入 `.gitignore`）。

原因：
- 原始资料通常很大（PDF、视频、归档文件）
- wiki 已经提取了关键信息到 `wiki/` 目录
- 原始文件可通过其他方式管理（网盘、Zotero、云存储）

### 如果你想追踪某些文件

编辑 `.gitignore`，添加例外规则：

```gitignore
# 追踪 Markdown 笔记
!sources/*.md

# 追踪特定重要文件
!sources/2026-04-10-key-paper.pdf
```

或使用 `git add -f` 强制添加：

```bash
git add -f sources/important-notes.md
```

## Zotero 私有接入层

`sources/zotero/` 专用于 Zotero 管理文件的本机接入：

- `sources/zotero/metadata.yaml`：私有 Zotero 绑定账本，可记录 item key、attachment key、当前机器本地路径和 source alias。
- `sources/zotero/**`：由 `scripts/zotero_sources.py` 生成的本机 symlink cache。

这整个目录默认不进 Git。metadata 可以包含当前机器的绝对路径，因为它不是公共项目状态。不要把 Agent 生成的摘要、wiki 草稿或综合知识写入这里。

推荐流程：

```bash
python scripts/zotero_sources.py --dry-run
python scripts/zotero_sources.py
```

## 使用注意

- 此目录只由**用户管理**（添加、删除、重命名）
- Agent **只读**，不会修改或删除这里的文件
- **新增**：Agent 仅在用户明确提供 URL/DOI 时，可通过网络工具（curl、playwright 等）获取文件写入此处；禁止写入任何 LLM 生成的内容
- 大型文件建议先压缩或提取关键部分

_meta.json

{
  "ownerId": "kn777gbj4abkdkaryavqq4522982d0zm",
  "slug": "041-llm-wiki",
  "version": "1.5.3",
  "publishedAt": 1788966448152
}

AGENTS.md

# LLM-Wiki Unified Agent Protocol

> This document guides Claude Code, OpenClaw, and other AI Agents on how to use llm-wiki.
> `AGENTS.md` is the canonical protocol file. `CLAUDE.md` must be a relative symbolic link to `AGENTS.md`; edit `AGENTS.md` only.
> **All Agents operating in this project directory MUST read and understand `SKILL.md` before performing any task.**

## Design Philosophy

- **LLM as programmer, Wiki as codebase**
- **User is responsible for**: placing materials, asking good questions, judging significance
- **Agent is responsible for**: summarizing, cross-referencing, indexing, logging, and maintaining structural consistency
- **Accumulation over retrieval**: interactions should leave lasting value when they produce reusable knowledge
- **Zotero as literature layer, wiki as knowledge layer**: Zotero can manage bibliographic metadata, PDFs, annotations, collections, tags, and citation keys; llm-wiki should distill concepts, relationships, time ordering, and synthesis into Markdown

## Core Work Protocol

### Ingest

Use Protocol mode for ingest because it requires LLM judgment. The Agent must:

1. Verify the source exists in `sources/` or was fetched through a real network/Zotero operation.
2. Run the post-fetch verification gate before using network-fetched files.
3. Extract source metadata, including title, authors/creator, URL/DOI/arXiv/citation key where available.
4. Extract source time metadata:
   - `published`: paper publication, arXiv date, article/post date, release date, or documentation date.
   - `updated_at_source`: source-side update time if available.
   - `collected`: user collection, Zotero import, or saved time if available.
   - `ingested`: llm-wiki processing date.
   - `date_precision`: `day`, `month`, `year`, or `unknown`.
5. Build a source content map before page drafting. Enumerate major topic units and source-supported mechanisms, equations, numerical evidence, comparisons, procedures, failure modes, trade-offs, decision rules, and open questions.
6. Allocate every important unit to the target page, another existing page, a new independently reusable page, or an explicit omission reason. A fixed summary budget is not an omission reason.
7. Select a page archetype and source-dependent headings. Required page invariants do not imply a fixed knowledge-body outline.
8. Draft at the depth required to preserve the source's defining reasoning. Keep central formulas, causal chains, implementation details, version differences, and evidence context when they determine the source's value.
9. Run coverage, depth, and batch-template-collapse reviews before setting `status: active`.
10. Preserve `sources_meta`, add temporal structure when useful, run dynamic linking, update `wiki/index.md`, and append `log.md` only after the content reviews pass.

The coverage review must account for each important source unit as included, allocated elsewhere, or intentionally omitted. The depth review must reject abstract-only para

Github ReposUpdated 5h agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.