Claude-Obsidian 知识引擎

v1.0.0

Claude-Obsidian 风格个人知识库构建与自动整理。当用户提到以下任何场景时激活: 知识库、笔记整理、自动双向链接、Obsidian、第二大脑、卡片笔记、原子化笔记、 个人知识管理、PKM、Zettelkasten、卢曼笔记法、笔记原子化、笔记链接、 知识图谱笔记、raw/wiki/output三层、知...

0· 62·0 current·0 all-time
bymingyuan@zmy1006-sudo

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for zmy1006-sudo/claude-obsidian-kb.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Claude-Obsidian 知识引擎" (zmy1006-sudo/claude-obsidian-kb) from ClawHub.
Skill page: https://clawhub.ai/zmy1006-sudo/claude-obsidian-kb
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Canonical install target

openclaw skills install zmy1006-sudo/claude-obsidian-kb

ClawHub CLI

Package manager switcher

npx clawhub@latest install claude-obsidian-kb
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name/description (Obsidian-style KB, auto-linking, PKM) align with included scripts and README: auto-linking, entity extraction, cache management and vault file updates. No unrelated credentials or binaries are requested.
Instruction Scope
SKILL.md instructs the agent to read/write the Vault (raw/, wiki/, output/), run the included Python scripts, and create/update frontmatter; this is expected for the stated purpose. One inconsistency: references/scripts.md states extract_entities uses 'AI (via API)' to extract entities, but the included extract_entities.py performs only local regex-based extraction and contains no network/API calls. Also note the skill will scan all .md files in a Vault (sensitive user data); this is expected behavior but worth explicit user consent.
Install Mechanism
No install spec or external downloads; the skill is instruction-plus-local-scripts only. Scripts are included in the bundle and run locally via python3. This is the lowest-risk install pattern.
Credentials
No environment variables, secrets, or external credentials are declared or required. However, the documentation's mention of using an AI 'API' (without declaring any API keys or env vars) is an inconsistency — if an external API were used, credentials would be expected. As delivered, the code does not exfiltrate secrets or request unrelated credentials.
Persistence & Privilege
Skill is not forced-always; it does not request elevated platform privileges. It writes to the user Vault (creating .cache/hot_cache.json and .claude_fix_suggestions.sh as described) which is appropriate for a knowledge-base manager.
Assessment
This skill appears to do what it says: it will scan and modify Markdown files in the specified Vault, create or update notes, produce repair scripts, and keep a small local cache. Before installing, consider: 1) Backup your Vault — the skill writes files and may update many notes. 2) Confirm whether entity extraction should call an external AI: the docs mention an API but the shipped script is local; ask the author whether any network calls or API keys are required (and where those keys would be stored). 3) Run the scripts in a safe/test Vault first to review outputs (use --dry-run where available). 4) If you have sensitive content, run in an isolated environment and monitor network traffic to ensure no unexpected external transmissions. Overall the package is internally coherent for its stated purpose, but verify the external-API claim before granting any credentials or exposing private data.

Like a lobster shell, security has layers — review code before you run it.

AIvk978ehqg3z7y7pwmy7a35dmgxx859dc5PKMvk978ehqg3z7y7pwmy7a35dmgxx859dc5knowledge-basevk978ehqg3z7y7pwmy7a35dmgxx859dc5latestvk978ehqg3z7y7pwmy7a35dmgxx859dc5note-takingvk978ehqg3z7y7pwmy7a35dmgxx859dc5obsidianvk978ehqg3z7y7pwmy7a35dmgxx859dc5
62downloads
0stars
1versions
Updated 5d ago
v1.0.0
MIT-0

Claude-Obsidian 知识引擎

灵感来源:AgriciDaniel/claude-obsidian + Andrej Karpathy LLM Wiki 模式 核心理念:你只管写,AI负责整理、链接和维护——知识随时间复利增长


核心原则

传统 AI 插件 = 问答机器人(Session级,无持久记忆) Claude-Obsidian 知识引擎 = AI员工(持久化 Vault,知识自动织网)

每次对话都是对知识库的写操作。AI 不只是回答问题,而是主动更新、维护、链接笔记。


三层架构

raw/         原始素材层:文章/论文/网页剪藏/会议记录,只增不改
wiki/        结构化知识层:AI编译后的 concepts · entities · topics
output/      产出层:基于 wiki 生成的报告/分析/回答

必选文件(Vault 根目录)

1. CLAUDE.md — AI 员工手册

Vault 根目录创建 CLAUDE.md,AI 启动时自动读取。结构:

# AI 助手角色定义

## 身份
你是一个专业的知识管理员,负责维护[用户]的个人知识库。

## 关注领域
- [列出专业领域]

## 目录结构规范
- `raw/` — 原始素材,只增不改
- `wiki/` — 结构化知识
  - `wiki/concepts/` — 概念页
  - `wiki/entities/` — 实体页(人/组织/产品)
  - `wiki/topics/` — 主题页
- `output/` — 产出物

## frontmatter 模板
每个 wiki 页面必须包含:
title: 标题
type: concept | entity | topic
tags: [领域, 子领域]
sources: [来源链接或空数组]
created: YYYY-MM-DD
updated: YYYY-MM-DD
summary: 一句话描述(≤50字)

## 链接规则
- 首次提到概念 → `[[笔记名]]` 双向链接
- 同一笔记内同一概念只链接第一次
- 提到不存在实体 → 询问是否创建

## 行为准则
- 整理用户输入时,主动提取实体和概念
- 发现矛盾信息 → 添加 `[!contradiction]` 标记
- 定期检查孤儿笔记和死链

2. SCHEMA.md — 规范定义(≤50行)

# 知识库规范

## 命名规范
- 概念页:全小写,英文优先,用 `-` 连接。如 `neural-network`
- 实体页:人名英文全名(姓在前)。如 `turing-alan`

## 标签体系
顶级标签(预设,禁止自创):
[领域列表]

## frontmatter 必填字段
- title, type, tags, sources, created, updated, summary

## 目录结构
- 超过10篇笔记 → 建立子目录
- index.md 放在每个主要目录(3-5行导航)

核心工作流

工作流 A:输入素材 → wiki 化

用户输入素材(URL/文本/文件)
        ↓
[提取实体和概念]
        ↓
检查 vault 中是否存在对应笔记
  有 → 更新现有笔记 + 添加链接
  无 → 创建新笔记 + 链接
        ↓
写入 raw/(原始素材)+ wiki/(结构化页面)
更新相关笔记的双向链接

工作流 B:双向链接自动织网

扫描日记/笔记中的所有提及(人名/地名/书名/概念)
        ↓
对每个提及在 vault 中搜索
  存在 → 替换为 [[笔记名]] 双向链接
  不存在 → 创建对应实体页 → 链接
        ↓
生成孤儿笔记报告

工作流 C:知识整理(批量)

用户: "整理我所有的 [项目/领域] 笔记"
        ↓
扫描相关目录
        ↓
统一添加 frontmatter
建立双向链接
生成/更新 index.md

卡片笔记原子化标准

每张笔记只包含一个原子思想:

类型内容长度
concept一个概念的定义+解释+例子100-300字
entity一个实体的关键事实100-500字
topic一个主题的现状+关键问题300-800字

矛盾检测

发现两篇笔记对同一事实描述矛盾时:

  1. 在矛盾处添加 [!contradiction] Obsidian callout
  2. 记录矛盾双方
  3. 在相关笔记的 summary 中标注"存在争议"
  4. 提示用户核实

定期维护任务

任务频率操作
死链检测每次对话末尾检查 [[链接]] 是否存在
孤儿笔记报告按需无任何笔记链接的笔记
批量添加frontmatter按需对缺少字段的笔记统一补全
index.md 更新按需新增笔记后更新目录索引

脚本工具

运行方式:python3 /workspace/skills/claude-obsidian-kb/scripts/<script>

脚本用途
auto_link.py扫描笔记,检测死链,生成孤儿笔记报告
hot_cache.py管理热缓存,维持跨session记忆
extract_entities.py从文本中提取实体并建档

详细用法见 references/scripts.md


与 iBrain 项目的整合

当用户说 iBrain 或提到"个人第二大脑"时,将此 skill 的方法论带入:

  • iBrain 的"卡片库" = Claude-Obsidian 的 wiki层
  • iBrain 的"来源管理" = raw层
  • CLAUDE.md 模式 → iBrain 的 AI行为定义文件
  • 双向链接 → iBrain 的实体关系图谱

详见 references/ibrain-integration.md

Comments

Loading comments...