AlexChanshuo

LLM WIKI TO AGENT Template

Name: LLM WIKI TO AGENT Template
Author: AlexChanshuo

AI community

SOURCE https://github.com/AlexChanshuo/LLM-WIKI-TO-AGENT-Template

Description

Meta-template: bootstrap one Obsidian vault + one Hermes Telegram librarian + paranoid GitHub backup, following Karpathy's LLM Wiki pattern. Give this repo to Claude Code (or Codex via AGENTS.md) and it walks you through filling in the placeholders.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

LLM-WIKI-TO-AGENT-Template

A reproducible blueprint for building one private Obsidian knowledge base, one Telegram-native capture agent, and one paranoid GitHub backup — end to end, on a single Mac.

**Provided by Alex Ma (馬驊)** — CC10 member, founder of [PainPoint Tech (痛點科技)](https://www.painpoint-ai.com). Contact: [alexma@painpoint-ai.com](mailto:alexma@painpoint-ai.com) · Web: [www.painpoint-ai.com](https://www.painpoint-ai.com)

中文簡介 (zh-TW)

這是一份 **meta-template**:讓你在一台 Mac 上,從零蓋出「Obsidian 知識庫 + Telegram 對話 Agent + GitHub 偏執備份」的完整堆疊。所有架構決策都來自 Andrej Karpathy 2026 年 4 月 3 日的 **「LLM Wiki」** gist。

**做法很簡單:**

git clone 這個 repo 到你的本機
在這個資料夾打開 Claude Code(或 OpenAI Codex,本 template 兩邊都支援)
跟它說:「walk me through AGENT.md」
回答它問的 6 個問題,剩下 45–90 分鐘它會自己跑完

**蓋出來的是什麼?** 一個你自己擁有、會自己更新、可以用 Telegram 對話的私人知識庫。新訊息 / 文章 / 語音 / 照片進來 → Python hook 確定性存檔 → Claude 自動分類、抽重點、跨頁更新 wiki → 每天 03:33 偏執備份 → 你週末問它「上週重點?」秒答。

**詳細中文文件:**

📘 `docs/zh-TW-overview.md` — 繁體中文完整總覽 (架構、四個操作、安全決策、常見問題)
🤖 `AGENT.zh-TW.md` — AGENT.md 中文導讀 (Claude / Codex 在執行什麼)
🎁 配套公開導讀: `AlexChanshuo/Gift-for-Event` — 馬驊 AI 知識庫分享會的全中文實戰指南

**作者:** 馬驊 (Alex Ma),CC10 成員、[痛點科技](https://www.painpoint-ai.com) 創辦人。聯絡: [alexma@painpoint-ai.com](mailto:alexma@painpoint-ai.com)。

Think of it as a private library

Easiest way to grok the architecture: **stop thinking "knowledge base" and start thinking "library."** Every piece has a real-world equivalent.

You walk past your private library every morning. A **Librarian** (a Telegram bot named `{{AGENT_NAME}}`) stands at the front desk. You toss a voice memo, an article, a half-formed thought into the drop box — they smile, confirm receipt, and put your contribution on the **Acquisitions cart**. Overnight, while you sleep, the library runs its cataloging shift: every item on the cart gets read, cross-referenced, filed onto the ri