一個 Claude的skill:把近期對話中值得歸檔的主題,原文匯入 Google Drive,交後續流程整理。
A Claude Code
/note-claudskill that archives note-worthy topics from your recent conversations verbatim into Google Drive, ready for later refinement.
設計理念是「只抓原文、不加工」。Claude 只負責兩件事:找出值得歸檔的主題、把對話原文落地成 .md 檔。它不摘要、不改寫、不重構。這樣可以確保歸檔內容是「乾淨的原始素材」,避免 AI 在歸檔當下就過度詮釋或遺漏。
English
/note-claud follows one principle: capture raw, add nothing. Claude does exactly two things — find archive-worthy topics and write the conversation down verbatim as a .md file. It does not summarize, rewrite, or restructure; all refinement is deferred to a later "Cowork" pass. This keeps every archive as clean source material, free from premature interpretation or loss.
| 功能 · Feature | 說明 · Description |
|---|---|
| 🔍 主題偵測 · Topic detection | 掃描最近幾次對話,找出可歸檔的知識點或討論串。 Scans recent conversations for archive-worthy knowledge or threads. |
| 📋 盤點確認 · Confirm before write | 列出候選主題請你勾選,存哪些由你決定。 Lists candidate topics and lets you choose what to keep. |
| 📝 原文匯入 · Verbatim import | 逐字保留對話,含「使用者/Claude」角色標示,不刪改。 Saves the conversation word-for-word with speaker labels intact. |
| 🗂️ 一主題一檔 · One file per topic | 多主題分檔,單批上限 3 篇,超過先寫 3 再問。 One file per topic; max 3 per batch, then asks before continuing. |
| ☁️ 直寫 Drive · Direct to Drive | 以原生 .md 寫入 Google Drive,Obsidian 可直接讀。Writes native .md to Google Drive, readable by Obsidian. |
| 🔒 隱私邊界 · Privacy guard | 個資/未公開資料整段排除並提醒,不寫入。 Excludes personal/confidential data entirely and warns you. |
中文
- 在 Claude Code 輸入
/note-claud - Claude 掃描近期對話,回報偵測到的候選主題:
近期對話偵測到可歸檔主題:[A] / [B] / [C],要存哪些?
- 你選定要歸檔的主題
- Claude 把每個主題的對話原文寫成
.md,寫入你設定的 Google Drive 資料夾 - 回報檔案連結;之後交由 Cowork 流程補 YAML、精煉成結構化筆記
English
- Type
/note-claudin Claude Code. - Claude scans recent chats and reports the candidates:
Archive-worthy topics detected: [A] / [B] / [C]. Which to save?
- You pick the topics to archive.
- Claude writes each topic's raw conversation as
.mdinto your configured Google Drive folder. - It returns the file links; a later Cowork pass adds YAML and refines them into structured notes.
- 格式 · Format:
YYYY-MM-DD 討論主題.md - 同日多檔第二筆起加
-02/-03。
Same-day duplicates get-02/-03. - 必帶
.md,否則 Obsidian 不認。
Must end in.md, or Obsidian won't recognize it. - 禁用字元 · Forbidden chars:
| / \ :與 emoji。
中文
機敏值(Drive 資料夾 ID、落點路徑)放在 config.local.json,此檔已被 .gitignore 排除,不會進版控。第一次使用時,複製範本並填入你的值:
English
Sensitive values (Drive folder ID, target path) live in config.local.json, which is git-ignored and never committed. On first use, copy the template and fill in your values:
cp config.example.json config.local.json
driveFolderId怎麼找 · How to find it 開啟 Drive 資料夾,網址https://drive.google.com/drive/folders/【這段就是 ID】。 Open the folder; the ID is the last segment of the URL.
中文
把 SKILL.md 放到 Claude Code 能找到 skill 的位置(使用者層級或專案層級),/note-claud 即可使用。config.local.json 需與 SKILL.md 同目錄,skill 會在動作前讀取它。
English
Place SKILL.md where Claude Code discovers skills (user- or project-level), and /note-claud becomes available. Keep config.local.json in the same directory as SKILL.md; the skill reads it before acting.
~/.claude/skills/note-claud/
├── SKILL.md # skill 定義 · skill definition
└── config.local.json # 你的機敏設定 · your local config (git-ignored)
| 檔案 · File | 用途 · Purpose | 版控 · VCS |
|---|---|---|
SKILL.md |
skill 主檔:觸發、流程、格式、邊界。 The skill: trigger, flow, format, boundaries. |
✅ |
config.example.json |
設定範本(僅佔位值)。 Config template (placeholders only). |
✅ |
config.local.json |
你的真實設定。 Your real config. |
🚫 ignored |
.gitignore |
排除機敏/本機檔案。 Excludes sensitive/local files. |
✅ |
README.md |
本文件。 This document. |
✅ |
中文
- 原封不動 ≠ 全部照搬:不刪改對話內容本身,但個資、院內未公開資料整段排除,不寫入並提醒。
- 不替後續流程做整理:不摘要、不重構、不臆測
domain/stage/type,這些欄位留白交給 Cowork。 - 機敏設定一律存在
config.local.json,不進版控。
English
- Verbatim ≠ copy everything: the conversation text is kept unaltered, but personal or confidential data is excluded wholesale and flagged, never written.
- No premature refinement: no summarizing, no restructuring, no guessing
domain/stage/type— those fields stay blank for the Cowork pass. - All sensitive settings stay in
config.local.json, out of version control.