Note_Claude

在 Claude 裡做筆記：一句指令，把對話原文一字不改地歸檔進你的 Obsidian / Google Drive。 Take notes inside Claude — one command to archive your conversations, word-for-word, into Obsidian / Google Drive.

---

1. 專案簡介 · About The Project

你在跟 Claude 聊天時，常常會冒出值得留下來的內容。這個工具讓你用一句指令 /note-claud，把那段對話原封不動存進你的筆記庫（Google Drive / Obsidian）——不改字、不摘要、不自作聰明。

When you chat with Claude, valuable ideas often come up. This tool lets you save that conversation into your note vault (Google Drive / Obsidian) with a single command — /note-claud — exactly as it was: no edits, no summaries, no "smart" rewriting.

---

2. 為什麼要用這個專案？ · Why Choose This?

GitHub 上把「AI 對話存成筆記」的工具其實不少，但它們大多走「盡量多抓、進場就自動整理」的路線。這個專案刻意反著走——先把乾淨的原料存好，聰明的事留到之後再做。具體來說：

There are quite a few tools on GitHub that turn AI chats into notes, but most of them follow the "grab everything, auto-organize on arrival" approach. This project deliberately goes the other way — save clean raw material first, and leave the clever work for later. Here's what that means:

✅ 不用匯出，在對話裡直接存 很多同類工具要你先打開整串對話，從瀏覽器匯出或下載 JSON 檔，再跑腳本轉檔。這裡只要在 Claude 裡打一句指令，它就直接從你最近的對話裡把內容撈出來——整段「匯出、下載、轉檔」的前置作業通通省掉。

✅ No export step — save right inside the chat Many similar tools make you open the full thread, export it from the browser or download a JSON file, then run a conversion script. Here, you just type one command inside Claude, and it pulls the content straight from your recent conversations — the whole "export, download, convert" routine is gone.

✅ 以「主題」為單位，不是整串硬塞 它不會把整串對話全部灌進來，也不會把你的歷史紀錄一次倒光。它會先掃描近期對話、找出「值得歸檔的主題」，列成清單問你：「偵測到主題 A / B / C，要存哪些？」你只留下真正想留的。

✅ Saves by topic, not by dumping whole threads It doesn't pour an entire thread into your vault, nor does it bulk-import your whole history. It scans your recent chats, finds "archive-worthy topics," and asks: "Found topics A / B / C — which ones do you want to keep?" You keep only what actually matters.

✅ 逐字保存，零改寫、零幻覺 這是最大的不同：多數工具進場就自動摘要、自動打標籤、自動重組你的內容——但 AI 一動手整理，就可能改了原意、加了它自己腦補的東西。這個工具明文規定什麼都不做：不摘要、不改寫、不猜分類，標籤欄位一律留白。你存下來的，就是對話原本的樣子。

✅ Word-for-word, zero rewriting, zero hallucination This is the biggest difference: most tools auto-summarize, auto-tag, and auto-restructure your content the moment it arrives — but once an AI starts "organizing," it can distort meaning or invent things that were never said. This tool is explicitly forbidden from doing any of that: no summaries, no rewrites, no guessed categories — tag fields are left blank on purpose. What you save is exactly what was said.

✅ 「搬運」和「整理」分成兩步，各司其職 市面工具幾乎都把「抓內容」和「整理內容」壓成同一個自動步驟。這裡刻意拆開：Claude 只負責把原文老實搬進 RAW/ 暫存區，之後再由第二階段的整理流程（如 Cowork）補標籤、精煉成結構化筆記。好處是：你的正式筆記庫永遠不會被半生不熟的自動整理污染。

✅ "Capture" and "organize" are two separate steps Most tools squash capturing and organizing into one automatic step. Here they're deliberately split: Claude only does the honest job of moving raw text into a RAW/ staging folder; a second-stage refinement process (such as Cowork) adds tags and turns it into structured notes later. The benefit: your real vault never gets polluted by half-baked auto-organization.

✅ 寫入前一定先問你 每一次寫入都要經過你點頭。它不是在背景默默運作的黑盒子——什麼被存、存去哪，永遠由你決定。

✅ It always asks before writing Every write needs your approval. It's not a black box running silently in the background — what gets saved, and where, is always your call.

✅ 隱私內建，不是事後補丁 遇到個人資料或不該外流的機敏內容（例如醫療、院內未公開資料），它會整段排除、不寫入，並主動提醒你。這在通用型工具裡幾乎看不到。

✅ Privacy is built in, not bolted on When it encounters personal data or confidential content (e.g., medical or internal unpublished information), it excludes that section entirely, never writes it, and warns you. This is something you'll rarely find in general-purpose tools.

一句話總結：主流工具想當「自動知識引擎」；這個專案想當「乾淨原料的入庫單」。原料不失真，整理才有意義。

In one sentence: mainstream tools try to be an "automatic knowledge engine"; this project is a "clean intake form for raw material." Organization only means something when the raw material is faithful.

---

3. 最新更新 · What's New

以下是最近一次（2026 年 7 月）的更新重點，這個專案持續維護中：

Here are the highlights from the latest update (July 2026) — the project is actively maintained:

• 全面改版中英雙語 README，重寫「為什麼要用這個專案」，說清楚與同類工具的差異。
  Fully revamped bilingual README, with a rewritten "Why Choose This?" that explains how it differs from similar tools.
• 加入 MIT 授權條款，任何人都能自由使用與修改。
  Added the MIT License — anyone can freely use and modify it.
• 正式封裝成可安裝的 Claude Code skill，指令為 /note-claud。
  Packaged as an installable Claude Code skill, triggered by /note-claud.
• 機敏設定分離：個人的 Google Drive 資料夾資訊獨立存放在本機，不會上傳到網路。
  Sensitive settings separated: your private Google Drive info stays local and is never uploaded.

---

4. 快速開始 · Getting Started

只要三個步驟就能裝好。請打開你的終端機（Terminal），依序執行下面的指令：

You can set it up in three steps. Open your terminal and run the commands below in order:

# 步驟 1：把專案複製到 Claude Code 的 skills 資料夾
# Step 1: Clone the project into your Claude Code skills folder
git clone https://github.com/kau10082/Note_Claude.git ~/.claude/skills/note-claud

# 步驟 2：進入資料夾，用範本建立你自己的設定檔
# Step 2: Enter the folder and create your own config from the template
cd ~/.claude/skills/note-claud
cp config.example.json config.local.json

# 步驟 3：打開 config.local.json，填入你的 Google Drive 資料夾 ID
# Step 3: Open config.local.json and fill in your Google Drive folder ID

Windows 使用者請注意：第 1 步的路徑請改用 "$env:USERPROFILE\.claude\skills\note-claud"。
Windows note: for Step 1, use the path "$env:USERPROFILE\.claude\skills\note-claud" instead.

裝好後，重新啟動 Claude Code，就完成了！

Once installed, restart Claude Code — and you're done!

---

5. 基本使用方式 · Usage

在 Claude Code 的對話框裡輸入指令 /note-claud，接著它會帶你走完整個流程：

Type the command /note-claud in Claude Code, and it will walk you through the whole process:

1. 偵測主題：它會掃描你最近的對話，找出值得存檔的主題。
   Detect topics: it scans your recent chats and finds archive-worthy topics.
2. 讓你挑選：它會列出候選清單，由你決定要存哪些。
   Let you choose: it lists the candidates and lets you pick which to keep.
3. 原文寫入：把你選的對話原封不動存成 .md 檔，放進你的 Google Drive。
   Save verbatim: it writes your chosen conversation as a .md file into your Google Drive.

小提醒：指令名稱是 /note-claud（不是 /note）。輸入 /note 時，系統通常會自動補全到它。
Tip: the command is /note-claud (not /note). Typing /note will usually auto-complete to it.

---

6. 目錄結構 · Repository Structure

專案很單純，主要就是這幾個檔案：

The project is simple — just a few files:

Note_Claude/
├── SKILL.md              # 主角：指令的定義與運作流程
├── config.example.json   # 設定檔範本（給大家參考用）
├── config.local.json     # 你的私人設定（本機專屬，不會上傳）
├── README.md             # 說明文件（就是你正在看的這份）
├── LICENSE               # 授權條款（MIT）
└── .gitignore            # 指定哪些檔案不要上傳

各檔案的白話說明如下：

Here's what each file does, in plain words:

• SKILL.md — 這個專案的核心，告訴 Claude 收到 /note-claud 後該做什麼。
  The heart of the project — it tells Claude what to do when it receives /note-claud.
• config.example.json — 設定的「空白範本」，複製它來建立你自己的設定。
  A blank template for settings — copy it to create your own config.
• config.local.json — 你的私人設定，存放 Google Drive 資料夾資訊，只留在你電腦裡、不會上傳。
  Your private settings (Google Drive folder info) — stays on your computer, never uploaded.
• README.md — 說明文件。
  This documentation file.
• LICENSE — 授權條款。
  The license terms.
• .gitignore — 一份清單，指定哪些檔案（例如你的私人設定）不要被上傳。
  A list telling Git which files (like your private settings) to keep off the internet.

---

7. 貢獻指南 · Contributing

非常歡迎你一起讓這個專案變得更好！無論是修正錯字、改善功能，還是回報問題，都幫得上忙。

You're very welcome to help make this project better! Whether it's fixing a typo, improving a feature, or reporting a bug — every bit helps.

想回報問題或提出建議？ 到專案的 Issues 頁面，開一則新的 issue 說明你的想法就好。

Want to report a bug or suggest an idea? Just open a new issue on the project's Issues page and describe it.

想動手改程式？ 流程很簡單：

Want to make changes yourself? The steps are simple:

1. Fork 這個專案（複製一份到你自己的帳號）。
   Fork this project (make a copy under your own account).
2. 建立一個新分支來放你的修改。
   Create a new branch for your changes.
3. 完成後，發一個 Pull Request (PR)，說明你改了什麼。
   When done, open a Pull Request (PR) and describe what you changed.

---

8. 授權條款 · License

本專案採用 MIT License。簡單來說：你可以自由使用、修改、散布，甚至用於商業用途，唯一的條件是保留原始的版權與授權聲明（也就是標示出處）。

This project uses the MIT License. In short: you're free to use, modify, distribute, and even use it commercially — the only condition is to keep the original copyright and license notice (i.e., give credit).