Illustrated Tutorial: Install claude-obsidian in three steps, so that every AI note has a source and every sentence can be traced back.图文教程:三步装好 claude-obsidian,让 AI 笔记条条有出处、句句可回溯
Follow along: Prepare the Obsidian vault and Claude Code → Install the plugin and connect the skills → Drop a piece of material into the vault and see it generate notes with citations → Ask questions to verify evidence-based answers. Runs entirely locally, making it suitable for users who want to seriously build a second brain.
跟着做一遍:准备 Obsidian 库和 Claude Code → 安装插件并接入技能 → 丢一篇资料进库看它生成带引用的笔记 → 提问验证证据式回答。全程本地运行,适合想认真搭第二大脑的用户。

Step 1: Prepare the environment第一步:准备环境
Confirm that the following three things are in place: Obsidian is installed (free download from the official website), Claude Code is installed and can be launched normally, and there is at least 1GB of free space on the machine. If you don't have Obsidian, spend five minutes installing it and creating a new, empty Vault.
Download the claude-obsidian plugin package from this site (see the button above) and extract it. Remember the path of the extracted directory, as it will be used in later steps.
It is recommended to create a new, dedicated Vault as a "test bed": Habits for the knowledge system should be cultivated starting with a small vault. Once it has been satisfactorily tested, the official information can be migrated over.
确认三样东西就位:已安装 Obsidian(官网免费下载)、已安装 Claude Code 并能正常启动、机器上有至少 1GB 空闲空间。没有 Obsidian 的先花五分钟装好并新建一个空 Vault。
从本站下载 claude-obsidian 插件包(见上方按钮)并解压,记住解压出的目录路径,后面步骤会用到。
建议新建一个专门的 Vault 当「试验田」:知识系统的习惯要从小库开始养,等验证满意再把正式资料迁进来。

Step 2: Integrate Claude Code第二步:接入 Claude Code
Open the terminal, use cd to enter the extracted claude-obsidian directory, and follow the installation section in the README file included in the package: You can register it as a local plugin marketplace and then install it, or directly copy the skills directory to ~/.claude/skills/.
After installation, restart Claude Code, enter /skills or view the list of skills to confirm that claude-obsidian has been loaded. For Codex users, place it in ~/.agents/skills/.
Key step: Start a Claude Code session in the directory path of the target Vault (cd to the Vault root directory and then claude). The Agent will only follow the "evidence-based" workflow if it can see the repository files.
打开终端,cd 进入解压出的 claude-obsidian 目录,按包内 README 的安装小节操作:可将其注册为本地插件市场后安装,或直接把技能目录复制到 ~/.claude/skills/ 下。
装完重启 Claude Code,输入 /skills 或查看技能列表确认 claude-obsidian 已加载;Codex 用户对应放进 ~/.agents/skills/。
关键一步:在目标 Vault 的目录路径下启动 Claude Code 会话(cd 到 Vault 根目录再 claude)。Agent 只有看到库文件,才会走「基于证据」的工作路径。

Step 3: Drop a source and see what the citation note looks like第三步:丢一篇资料,看引用笔记长什么样
Choose an article that you have saved but haven't read for a long time (either a web link or a PDF path will do), and say to the Agent: "Organise this article into a note with a source link and save it to the library".
Observe the Markdown output: There should be structured sections, with each key conclusion followed by a source link, and bidirectional links should be automatically established between files. Open it in Obsidian, and click on the link to return to the original source.
If you are not satisfied with the format, simply tell the Agent to make adjustments (for example, "Add a summary sentence at the beginning of each note"), and it will incorporate the request into your own note template.
选一篇你收藏了很久没读的文章(网页链接或 PDF 路径都行),对 Agent 说:「把这篇文章整理成带来源链接的笔记,存进库里」。
观察产出的 Markdown:应该有结构化小节、每个关键结论后跟着来源链接,文件之间会自动建立双链。在 Obsidian 里打开它,点链接能回到原始出处。
如果不满意格式,直接告诉 Agent 调整(比如「每篇笔记开头加一句摘要」),它会把要求沉淀成你自己的笔记模板。

Step 4: Validate the "Evidence-Based Response" with the pace of everyday use第四步:验证「证据式回答」与日常使用节奏
Ask a question that is already covered in the library (for example, "In the article about local models that I saved last week, which models did the author recommend?") Check whether the answer references a specific document. Then ask a question that is not in the library to confirm that it explicitly states "not found in the library" rather than making something up on the spot.
Daily routine suggestion: Throw bits of information into the vault for it to organise. Use "Help me summarise the connections between this week's new content" once a week. The value of bidirectional linked notes will compound over time.
Let me emphasise the privacy bottom line once again: The entire process runs locally, and data does not leave the device. However, model API calls to the Agent will still send the content that you explicitly allow it to read. Please use your own judgement regarding confidential files.
问一个库里已覆盖的问题(例如「我上周存的那篇讲本地模型的文章里,作者推荐了哪几个模型?」),检查回答是否引用了具体文件;再问一个库里没有的问题,确认它明说「库里找不到」而不是现编。
日常节奏建议:碎片资料随手丢进库让它整理,每周用一次「帮我总结这周新增内容之间的联系」,双链笔记的价值会随时间复利。
隐私底线再强调一次:全程本地运行,资料不出机器;但给 Agent 的模型 API 调用仍会发送你明确让它读取的内容,涉密文件请自行判断。

Problem Encountered遇到问题
Agent cannot see the library: In nine out of ten cases, the session start directory is incorrect. It must be started in the Vault root directory. Skill not loaded: Confirm that the directory has been placed in skills/ and that the host has been restarted. The plugin package that accompanies this article can be found via the download button above.
Agent 看不到库:十有八九是会话启动目录不对,必须在 Vault 根目录启动;技能没加载:确认目录放进 skills/ 后重启了宿主。本文配套插件包见上方下载按钮。
想马上用起来?去「AI工具」栏挑一个直接下载,或在「AI教程」里跟着图文步骤做一遍。
去 AI工具 → 看 AI教程 →阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。
今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告
评论 文明发言,让讨论更有价值