如何构建永不崩溃的 Obsidian 个人操作系统

CyrilXBT 2026-05-20 37.5 KB 828.7K 711 𝕏 文章列表

文章介绍了一套基于 Obsidian 的三层架构个人操作系统，旨在解决传统生产力系统在低谷期容易崩溃的问题。该系统通过 Claude Code (MCP) 实现智能维护，利用 N8N 进行自动化，并结合严格的 Vault 结构（如捕获、活跃、归档等八大文件夹），将笔记软件转变为无需手动维护、能在压力下自我运行的智能 OS。

# How to Turn Obsidian Into a Personal Operating System That Never Breaks Down **作者**: CyrilXBT **日期**: 2026-05-20T02:26:13.000Z **来源**: [https://x.com/cyrilXBT/status/2056924424838815824](https://x.com/cyrilXBT/status/2056924424838815824) --- ![Image](/doc/2098/image/img_dfea2392d5d6.jpg) Most productivity systems break down for the same reason. They are designed for good days. On a good day you have the time and energy to maintain the system. You file notes correctly. You update project statuses. You review your inbox. You process everything that came in. On a bad day the system becomes a source of guilt rather than leverage. You have not processed the inbox in four days. Notes are piled in random folders. Project statuses are weeks out of date. You stop using the system because maintaining it takes more energy than it saves. The Obsidian personal operating system is designed differently. It is not designed for good days. It is designed to survive bad ones. The architecture maintains itself when you are overwhelmed, produces useful output when you are inconsistent, and compounds intelligently whether you are actively curating it or not. This guide is the complete build for an Obsidian system that runs your life and never breaks down regardless of what life throws at you. ## Why Most Obsidian Systems Break Before the build you need to understand why the systems most people build do not last. They break for three reasons. Manual maintenance burden. The system requires you to update it regularly to stay accurate. Project statuses need manual updates. Dashboards need manual refreshing. Tags need manual application. When life gets busy the maintenance falls behind and the system becomes unreliable. An unreliable system gets abandoned. Complexity that compounds over time. People start with a simple structure and add complexity every time they encounter a problem the current system does not handle. Six months later they have a baroque architecture that takes 20 minutes to navigate and requires reading documentation they wrote themselves to understand. No intelligence layer. The vault stores information but does not reason about it. You capture ideas but never find them again. You take meeting notes but never surface the action items when they become relevant. The system is an archive not an intelligence. The operating system architecture solves all three. Manual maintenance disappears because Claude Code connected via MCP maintains the system automatically. Complexity stays controlled because the architecture has a fixed number of components each with a specific function and no component duplicates another. Intelligence gets added through Claude which reads the vault, makes connections, surfaces relevant information, and generates outputs that the vault alone cannot produce. ## The Three-Layer Architecture The personal operating system has three distinct layers. Layer 1: The Storage Layer This is Obsidian itself. Plain text Markdown files organized in a consistent structure. Every piece of information your life generates lives here in a format that is human-readable, machine-readable, and permanent. Layer 2: The Intelligence Layer This is Claude Code connected to your vault via the Filesystem MCP. Claude reads your files, makes connections across them, generates outputs from them, and updates them based on what happens in your life. Layer 3: The Automation Layer This is N8N running on a $5 server. It schedules workflows, fires triggers, calls the Claude API, and passes information between systems without you initiating anything. These three layers are what transform a collection of notes into an operating system. Remove any one layer and you have something less. Storage without intelligence is an archive. Storage with intelligence but without automation is a tool you use. All three together is a system that operates. ## The Vault Structure That Never Breaks The vault structure is the foundation of everything. Get this right and the system stays coherent as it grows. Get it wrong and it becomes a maze. The structure has eight folders. Every note in your entire vault belongs in exactly one of them. No folder is optional. No folder overlaps with another. 00 - CAPTURE/ [Everything unprocessed lands here] 01 - ACTIVE/ projects/ [project-name]/ overview.md tasks/ notes/ outputs/ areas/ health/ finances/ relationships/ learning/ career/ daily/ [YYYY-MM-DD].md 02 - RESOURCES/ research/ references/ templates/ bookmarks/ 03 - SYSTEM/ CLAUDE.md skills/ workflows/ logs/ 04 - GENERATED/ briefings/ summaries/ analyses/ drafts/ 05 - QUEUE/ [pending tasks for Claude] 06 - CALENDAR/ events/ reviews/ 07 - ARCHIVE/ [completed projects and outdated content] The logic behind every folder: 00 - CAPTURE exists because capture and processing are separate activities. You capture at the speed of thought. You process at the speed of attention. The capture folder absorbs everything without judgment so nothing gets lost because processing was too slow. 01 - ACTIVE contains only what is alive right now. Projects you are working on. Areas of your life you are responsible for. Daily notes. Nothing historical. Nothing aspirational. Only current. 02 - RESOURCES is the reference library. Things you might need someday. Not things you are working on now. 03 - SYSTEM is the operating system itself. CLAUDE.md. Skill files. Workflow definitions. Logs. The infrastructure layer. 04 - GENERATED is where Claude deposits outputs. Never manually edit files here. They are system outputs. 05 - QUEUE is the inbox for Claude tasks. You drop files here describing what you need. Claude processes them. Outputs land in GENERATED. 06 - CALENDAR tracks time-based information. Upcoming events. Periodic reviews. Things organized by when rather than what. 07 - ARCHIVE is where things go when they are complete. Projects you finished. References that are outdated. Never delete. Archive. ## The CLAUDE.md That Makes It Intelligent The CLAUDE.md is the document that tells Claude everything about your life before any workflow runs. Without it every Claude session starts from zero. With it every session starts from complete context about who you are, what you are working on, and what matters right now. # Personal Operating System — CLAUDE.md ## Identity Name: [YOUR NAME] Role: [YOUR PRIMARY ROLE] Location: [YOUR CITY] ## Life Areas and Current Status Health: [BRIEF STATUS — e.g., "Training for half marathon, running 4x per week"] Finances: [BRIEF STATUS — e.g., "Saving for house deposit, target by December 2026"] Relationships: [BRIEF STATUS — e.g., "Partner: [name]. Key friendships to maintain: [names]"] Learning: [BRIEF STATUS — e.g., "Currently studying: [topic]. Goal: [outcome]"] Career: [BRIEF STATUS — e.g., "Current role: [title]. Working toward: [goal]"] ## Active Projects [PROJECT NAME]: [ONE-SENTENCE DESCRIPTION] | Status: [STATUS] | Next: [SPECIFIC NEXT ACTION] [Repeat for every active project] ## Current Priorities 1. [MOST IMPORTANT THING THIS WEEK] 2. [SECOND MOST IMPORTANT] 3. [THIRD MOST IMPORTANT] ## Standards for Generated Content Voice: [HOW YOU WRITE AND COMMUNICATE] Format preferences: [YOUR PREFERENCES] What you never want: [SPECIFIC THINGS TO AVOID] ## Operating Rules - Never delete files. Move to ARCHIVE with timestamp. - Never send communications without human review. - Always date-stamp generated files as YYYY-MM-DD. - Log every significant action in SYSTEM/logs/operations.md. - When uncertain about placement, use GENERATED and flag. ## Update Schedule This file is reviewed and updated: [YOUR SCHEDULE — e.g., every Monday morning] Update the Current Priorities section every Monday morning. This is the single most important maintenance habit in the entire system. Five minutes of honest reflection on what actually matters this week dramatically improves the relevance of everything the system generates. ## The Five Workflows That Make It an Operating System Five workflows transform the vault from a storage system into an operating system. Each one runs automatically. Each one outputs to GENERATED. Each one maintains itself without your manual intervention. Workflow 1: The Daily Morning Briefing Every morning at 6AM this workflow reads your vault and generates a briefing. The Claude prompt: Read CLAUDE.md for full life context. Generate a morning briefing covering: MOST IMPORTANT TODAY: The single highest-leverage action I can take today based on my current priorities. SCHEDULE: Any events in CALENDAR/events/ dated today. Prepare a one-sentence brief for each one. OPEN LOOPS: Any items prefixed with OPEN: in yesterday's daily note. PROJECT PULSE: For each active project in CLAUDE.md: one sentence on status, one sentence on next action. WEEKLY FOCUS: If today is Monday, read the previous week's daily notes and identify the most important thing to accomplish this week. Format for under 300 words. Start with MOST IMPORTANT TODAY. Save to: GENERATED/briefings/[DATE]-morning.md This briefing runs before you open your laptop. You read it in three minutes. You know what matters before you see a single notification. Workflow 2: The Capture Processor Every evening at 8PM this workflow processes everything in 00 - CAPTURE. The Claude prompt: Read all files in the CAPTURE folder created today. For each captured item: 1. Identify what type it is: TASK: Something that requires action IDEA: Something to develop or research later REFERENCE: Information to store for later use NOTE: Context or observation about something active EVENT: Something time-bound to calendar 2. File it in the correct location: TASK → Create a task note in the relevant project or area IDEA → File in RESOURCES/research/ with the date REFERENCE → File in RESOURCES/references/ by topic NOTE → Append to the relevant project or daily note EVENT → Create an event file in CALENDAR/events/ 3. For TASKS: extract the specific action, assign a due date if mentioned, and add to the relevant project task list. 4. After processing: move the original capture file to ARCHIVE. Log all processing actions in SYSTEM/logs/capture-log.md. This workflow means 00 - CAPTURE empties itself every evening. Nothing accumulates. Nothing gets lost in the inbox. Every capture ends up exactly where it belongs without you doing the filing. Workflow 3: The Weekly Review Generator Every Sunday at 7PM this workflow generates your weekly review. The Claude prompt: Read all daily notes from the past 7 days. Read all project notes modified this week. Read CLAUDE.md for current life context and priorities. Generate the weekly review: WHAT MOVED FORWARD: Specific wins from the week. What caused each one. Be specific. Name the projects, the actions, the results. WHAT DID NOT MOVE: Honest assessment of what stalled. The most likely reason for each item that did not progress. THE WEEK'S PATTERN: One theme or insight that appeared repeatedly. Whether it signals something to change. NEXT WEEK'S PRIORITIES: Three specific priorities for next week ranked by impact. For each: why it matters, what the specific next action is. ONE DECISION: The most important decision sitting open right now. What information you have to make it. What the cost of delaying it another week is. Save to: GENERATED/summaries/[DATE]-weekly-review.md Update CLAUDE.md: Current Priorities section with next week's top 3. The weekly review used to take 45 minutes of staring at your notes trying to remember what happened. Now it takes 10 minutes to read what the system generated and 2 minutes to add anything it missed. Workflow 4: The Queue Processor Every 2 hours this workflow checks 05 - QUEUE for any files you dropped there. The convention is simple. Name the file with a verb and a topic: RESEARCH-stoic-philosophy-applications.md SUMMARIZE-project-meeting-notes.md DRAFT-email-to-landlord-about-repairs.md PLAN-trip-to-portugal-october.md DECIDE-whether-to-accept-new-client.md The workflow reads the filename to understand the task type and the file content for specific instructions. The Claude prompt: Check the QUEUE folder for any unprocessed files. For each file found: 1. Read the filename to identify the task type 2. Read the file content for specific instructions 3. Execute the task using the relevant skill from SYSTEM/skills/ if one exists 4. Save output to GENERATED/[task-type]/[DATE]-[topic].md 5. Move the queue file to ARCHIVE/queue-processed/ Log all queue processing in SYSTEM/logs/queue-log.md. If the task requires information not available in the vault: Flag in the output as NEEDS HUMAN INPUT: [what is needed] The queue processor is the most powerful workflow in the system because it handles anything. You think of something you want at midnight. You drop a file in the queue. It is processed by 2AM. The output is waiting when you wake up. Workflow 5: The Project Health Monitor Every Monday at 7AM this workflow checks all active projects. The Claude prompt: Read CLAUDE.md for the list of active projects. Read the overview.md file in each active project folder. Read any project files modified in the last 7 days. For each project generate a health assessment: STATUS: On track / At risk / Stalled / Blocked EVIDENCE: What specifically indicates this status. Be concrete. Name files, dates, and specific observations. NEXT ACTION: The single most important thing this project needs in the next 48 hours. FLAG: Any project that has had no activity for more than 7 days. These need human attention. Save health report to: GENERATED/briefings/[DATE]-project-health.md For any FLAGGED projects: create a file in QUEUE named: REVIEW-[project-name].md with a summary of the situation. The project health monitor surfaces problems before they become crises. A project that has been stalled for seven days gets flagged on Monday morning rather than discovered in a panic when a deadline arrives. ## The Anti-Breakdown Features The operating system is designed with three specific anti-breakdown mechanisms. The Capture Safety Net Everything goes to 00 - CAPTURE first. No decision is required at capture time. You never skip capturing something because you do not know where to file it. The capture folder absorbs everything and the processor files it later. This means the system survives your busy days intact. On a day when you have no time for anything you still capture to 00 - CAPTURE. The processor handles the rest that evening automatically. The Archive Never Delete Rule Nothing in this system ever gets deleted. Completed projects move to ARCHIVE. Outdated references move to ARCHIVE. Processed captures move to ARCHIVE. This means the system never loses information. You cannot break it by archiving too aggressively. The cost of keeping everything is zero because storage is infinite and the system never requires you to browse the archive for daily operation. The CLAUDE.md Single Source of Truth Everything about your life that Claude needs to operate exists in one file. One file to update. One file to review. One file that governs every workflow. When your priorities change you update CLAUDE.md and every subsequent workflow reflects the new reality automatically. You do not have to update five different documents or re-explain your situation in five different sessions. ## The Build Order That Gets You Running in One Weekend Build the system in this exact order. Resist the urge to build everything simultaneously. Saturday Morning: Storage Layer Create the eight folders. Set up the CLAUDE.md with your actual current information. Write one project overview for your most active project with consistent properties. Time required: 2 hours. Saturday Afternoon: Intelligence Layer Install Claude Desktop. Configure the Filesystem MCP. Point it at your vault. Run the first morning briefing manually and verify it produces accurate output based on your CLAUDE.md. Time required: 1 hour. Saturday Evening: First Queue Task Drop one file in your QUEUE folder with a real task. A research topic you have been meaning to explore. A decision you have been postponing. A summary of notes you took this week. Let Claude process it. Review the output. Note what to improve in your CLAUDE.md. Time required: 30 minutes. Sunday Morning: Automation Layer Set up N8N. Build the morning briefing workflow as a cron job. Let it run at its scheduled time and verify the output appears in GENERATED automatically. Time required: 2 hours. Sunday Afternoon: Remaining Workflows Add the capture processor, weekly review generator, and queue processor as additional N8N workflows. Time required: 2 hours. By Sunday evening you have all five workflows running. The operating system is operational. ## What Never Breaking Down Actually Looks Like Three months after building this system you will notice something. The system has been running the entire time. Not because you maintained it perfectly. You did not. There were weeks where you barely touched Obsidian. There were days where captures piled up in 00 - CAPTURE for three days before you thought about them. But the morning briefing kept running every morning. The captures kept getting processed every evening. The weekly review kept generating every Sunday. The project health monitor kept flagging stalled projects every Monday. The system kept operating whether you were operating it or not. That is the difference between a productivity tool and an operating system. A tool requires you to use it. An operating system runs. Build the foundation this weekend. The operating system runs from the first morning it is live. Follow @cyrilXBT for the exact CLAUDE.md templates, N8N workflow configurations, and Filesystem MCP setup that makes this entire architecture run. ## 相关链接 - [CyrilXBT reposted](https://x.com/cyrilXBT) - [CyrilXBT](https://x.com/cyrilXBT) - [@cyrilXBT](https://x.com/cyrilXBT) - [16K](https://x.com/cyrilXBT/status/2056924424838815824/analytics) - [@cyrilXBT](https://x.com/@cyrilXBT) - [Upgrade to Premium](https://x.com/i/premium_sign_up) - [10:26 AM · May 20, 2026](https://x.com/cyrilXBT/status/2056924424838815824) - [16.1K Views](https://x.com/cyrilXBT/status/2056924424838815824/analytics) - [View quotes](https://x.com/cyrilXBT/status/2056924424838815824/quotes) --- *导出时间: 2026/5/20 14:40:32* --- ## 中文翻译 # 如何把 Obsidian 变成一个永不出故障的个人操作系统 **作者**： CyrilXBT **日期**： 2026-05-20T02:26:13.000Z **来源**： [https://x.com/cyrilXBT/status/2056924424838815824](https://x.com/cyrilXBT/status/2056924424838815824) --- ![Image](/doc/2098/image/img_dfea2392d5d6.jpg) 大多数生产力系统崩溃的原因都是同一个。它们是为好日子设计的。在状态好的日子里，你有时间和精力去维护系统。你正确地归档笔记，更新项目状态，清空收件箱，处理所有进来的信息。在糟糕的日子里，系统就变成内疚的来源，而不是助力。你已经四天没有处理收件箱了。笔记堆积在随意的文件夹里。项目状态已经过时了好几周。你不再使用这个系统，因为维护它耗费的精力比它省下的还多。 Obsidian 个人操作系统的设计理念则完全不同。它不是为好日子设计的。它是为熬过糟糕的日子而设计的。当你不堪重负时，这套架构依然能自我维持；当你三心二意时，它依然能产出有用的结果；无论你是否在积极打理它，它都会智能地不断增值。这份指南是一个完整的 Obsidian 系统搭建方案，它能打理你的生活，并且无论生活向你抛出什么，它都不会崩溃。 ## 为什么大多数 Obsidian 系统会崩溃在搭建之前，你需要理解为什么大多数人搭建的系统无法长久。它们崩溃的原因有三个。手动维护负担。系统要求你定期更新才能保持准确。项目状态需要手动更新。仪表盘需要手动刷新。标签需要手动添加。当生活忙碌起来时，维护就落后了，系统变得不可靠。一个不可靠的系统最终会被抛弃。复杂度随时间不断累积。人们从一个简单的结构开始，每次遇到现有系统无法解决的问题就添加一层复杂度。六个月后，他们拥有的是一个巴洛克式的复杂架构，导航需要 20 分钟，还需要阅读自己写的文档才能看懂。没有智能层。知识库只存储信息，但不对信息进行推理。你记录了想法，却再也找不到它们。你做了会议笔记，但当行动事项变得相关时却从不浮现出来。这个系统只是一个档案库，而不是一个智能体。操作系统架构可以同时解决这三个问题。手动维护消失了，因为通过 MCP 连接的 Claude Code 会自动维护系统。复杂度保持在可控范围内，因为架构中组件数量固定，每个组件都有特定功能，且没有任何组件相互重复。智能通过 Claude 加入，它会读取知识库、建立关联、呈现相关信息，并生成知识库自身无法产出的成果。 ## 三层架构个人操作系统由三个不同的层组成。第一层：存储层这就是 Obsidian 本身。以一致的结构组织的纯文本 Markdown 文件。你的生活产生的每一条信息都以人类可读、机器可读且永久保存的格式存放在这里。第二层：智能层这是通过 Filesystem MCP 连接到你知识库的 Claude Code。Claude 读取你的文件，在它们之间建立关联，从中生成成果，并根据你生活中的变化更新它们。第三层：自动化层这是运行在 5 美元服务器上的 N8N。它调度工作流、触发任务、调用 Claude API，并在各系统之间传递信息，无需你手动发起任何操作。正是这三个层把一堆笔记变成了一个操作系统。去掉任何一层，你得到的东西都会逊色。没有智能的存储只是一个档案库。有智能但没有自动化的存储只是你使用的一个工具。三者合在一起，才是一个能够自行运转的系统。 ## 永不崩溃的知识库结构知识库结构是这一切的基础。把它做对了，系统在增长过程中就能保持条理清晰。做错了，它就会变成一座迷宫。这个结构有八个文件夹。你整个知识库中的每一条笔记都只属于其中一个。没有可选项，没有文件夹之间相互重叠。 00 - CAPTURE/ [所有未处理的内容都落在这里] 01 - ACTIVE/ projects/ [project-name]/ overview.md tasks/ notes/ outputs/ areas/ health/ finances/ relationships/ learning/ career/ daily/ [YYYY-MM-DD].md 02 - RESOURCES/ research/ references/ templates/ bookmarks/ 03 - SYSTEM/ CLAUDE.md skills/ workflows/ logs/ 04 - GENERATED/ briefings/ summaries/ analyses/ drafts/ 05 - QUEUE/ [pending tasks for Claude] 06 - CALENDAR/ events/ reviews/ 07 - ARCHIVE/ [completed projects and outdated content] 每个文件夹背后的逻辑： 00 - CAPTURE 的存在是因为记录和处理是两个独立的活动。你以思维的速度记录，却以注意力的速度处理。捕获文件夹不加评判地吸纳一切，这样就不会因为处理太慢而丢失任何东西。 01 - ACTIVE 只包含当下活跃的内容。你正在进行的项目。你负责的生活领域。每日笔记。没有任何历史内容，没有任何愿景规划，只有当前。 02 - RESOURCES 是参考图书馆。那些你将来可能需要的东西，而不是你现在正在处理的东西。 03 - SYSTEM 是操作系统本身。CLAUDE.md、技能文件、工作流定义、日志。这是基础设施层。 04 - GENERATED 是 Claude 存放输出的地方。永远不要手动编辑这里的文件。它们是系统输出。 05 - QUEUE 是 Claude 任务的收件箱。你把描述需求的文件放在这里，Claude 处理它们，输出落到 GENERATED 中。 06 - 日历（CALENDAR）用于追踪基于时间的信息。即将到来的活动。定期回顾。按“何时”而非“什么”来组织的事物。 07 - 归档（ARCHIVE）是完成的事项的去处。你已完成的项目。已过时的参考资料。绝不删除。归档。 ## 让它变得智能的 CLAUDE.md CLAUDE.md 是在任何工作流运行之前，告诉 Claude 关于你生活的一切的文档。没有它，每次 Claude 会话都从零开始。有了它，每次会话都从完整的上下文开始——你是谁、你正在做什么、以及当下什么最重要。 # 个人操作系统 — CLAUDE.md ## 身份姓名：[你的姓名] 角色：[你的主要角色] 所在地：[你的城市] ## 生活领域及当前状态健康：[简要状态 — 例如：“正在为半程马拉松训练，每周跑步 4 次”] 财务：[简要状态 — 例如：“正在为购房首付储蓄，目标 2026 年 12 月前达成”] 人际关系：[简要状态 — 例如：“伴侣：[姓名]。需要维护的重要友谊：[姓名列表]”] 学习：[简要状态 — 例如：“正在学习： [主题]。目标：[成果]”] 职业：[简要状态 — 例如：“当前职位：[头衔]。努力方向：[目标]”] ## 活跃项目 [项目名称]：[一句话描述] | 状态：[状态] | 下一步：[具体的下一步行动] [对每个活跃项目重复以上格式] ## 当前优先事项 1. [本周最重要的事] 2. [第二重要的事] 3. [第三重要的事] ## 生成内容的标准语气风格：[你的写作与沟通方式] 格式偏好：[你的偏好] 绝对不要的内容：[需要避免的具体事项] ## 运行规则 - 永远不删除文件。移动到 ARCHIVE 并加上时间戳。 - 未经人工审阅，绝不发送任何通信内容。 - 生成的文件始终以 YYYY-MM-DD 格式标注日期。 - 将每项重要操作记录在 SYSTEM/logs/operations.md。 - 拿不准该放哪里时，放入 GENERATED 并标记。 ## 更新计划此文件的审阅与更新时间：[你的计划 — 例如，每周一早上] 每周一早上更新“当前优先事项”部分。这是整个系统中最重要的单一维护习惯。每周花五分钟诚实地反思本周真正重要的事情，会显著提升系统所生成内容的相关性。 ## 让它成为操作系统的五个工作流五个工作流将知识库从存储系统转变为操作系统。每个都自动运行。每个都输出到 GENERATED。每个都无需人工干预即可自我维护。工作流 1：每日晨间简报每天早上 6 点，此工作流读取你的知识库并生成一份简报。 Claude 提示词：阅读 CLAUDE.md 获取完整的生活上下文。生成一份涵盖以下内容的晨间简报：今日最重要之事：基于我当前的优先事项，今天我能采取的杠杆作用最大的单一行动。日程：CALENDAR/events/ 中日期为今天的任何活动。为每项活动准备一句话简介。未闭环事项：昨日每日笔记中任何以 OPEN: 为前缀的条目。项目动态：CLAUDE.md 中每个活跃项目：一句话说明状态，一句话说明下一步行动。本周焦点：如果今天是周一，阅读上周的每日笔记，并找出本周最重要需要完成的事项。格式控制在 300 字以内。以“今日最重要之事”开头。保存至：GENERATED/briefings/[日期]-morning.md 这份简报在你打开笔记本电脑之前就已生成。你花三分钟阅读它。在看到任何一条通知之前，你就已经知道什么最重要。工作流 2：捕获处理器每天晚上 8 点，此工作流处理 00 - 捕获（CAPTURE）中的所有内容。 Claude 提示词：读取今天在 CAPTURE 文件夹中创建的所有文件。对每个捕获的项目： 1. 判断其类型：任务（TASK）：需要采取行动的事项想法（IDEA）：稍后需要发展或研究的事项参考资料（REFERENCE）：为日后使用而存储的信息笔记（NOTE）：关于某个活跃事项的背景或观察活动（EVENT）：有具体时间、需要放入日历的事项 2. 将其归档到正确的位置：任务 → 在相关项目或领域创建任务笔记想法 → 带上日期归档到 RESOURCES/research/ 参考资料 → 按主题归档到 RESOURCES/references/ 笔记 → 附加到相关的项目或每日笔记中活动 → 在 CALENDAR/events/ 中创建活动文件 3. 对于任务：提取具体行动，如提及则分配截止日期，并添加到相关项目的任务列表中。 4. 处理完成后：将原始捕获文件移至 ARCHIVE。将所有处理操作记录在 SYSTEM/logs/capture-log.md 中。这个工作流意味着 00 - 捕获每天晚上都会自动清空。不会堆积。不会在收件箱中丢失任何东西。每个捕获的内容都最终落在它该在的地方，而你无需动手整理。工作流 3：每周回顾生成器每周日晚上 7 点，此工作流生成你的每周回顾。 Claude 提示词：读取过去 7 天的所有每日笔记。读取本周修改过的所有项目笔记。读取 CLAUDE.md 以获取当前生活上下文和优先事项。生成每周回顾：推进了什么：本周的具体成果。每个成果的原因。要具体。指出项目、行动、结果。没有推进什么：对停滞事项的诚实评估。每个未取得进展的事项最可能的原因。本周的模式：反复出现的一个主题或洞见。它是否暗示了需要改变什么。下周的优先事项：下周的三个具体优先事项，按影响力排序。对每一项：为什么重要，具体的下一步行动是什么。一个决策：当下悬而未决的最重要决策。你已掌握哪些可用于决策的信息。再拖延一周的代价是什么。保存至：GENERATED/summaries/[日期]-weekly-review.md 更新 CLAUDE.md：用下周的前三大事项更新“当前优先事项”部分。每周回顾过去需要花 45 分钟盯着笔记，努力回想发生了什么。现在阅读系统生成的内容只需10分钟，再花2分钟补充它遗漏的内容。工作流4：队列处理器这个工作流每2小时检查一次 05 - QUEUE 文件夹，看是否有你放入的文件。命名约定很简单。用动词加主题来命名文件： RESEARCH-stoic-philosophy-applications.md SUMMARIZE-project-meeting-notes.md DRAFT-email-to-landlord-about-repairs.md PLAN-trip-to-portugal-october.md DECIDE-whether-to-accept-new-client.md 工作流通过读取文件名来了解任务类型，通过读取文件内容获取具体指令。 Claude 提示词：检查 QUEUE 文件夹中是否有未处理的文件。对找到的每个文件： 1. 读取文件名以确定任务类型 2. 读取文件内容获取具体指令 3. 使用 SYSTEM/skills/ 中相关的技能（如果存在）执行任务 4. 将输出保存到 GENERATED/[task-type]/[DATE]-[topic].md 5. 将队列文件移动到 ARCHIVE/queue-processed/ 在 SYSTEM/logs/queue-log.md 中记录所有队列处理情况。如果任务需要知识库中不可用的信息：在输出中标记 NEEDS HUMAN INPUT: [需要什么] 队列处理器是整个系统中最强大的工作流，因为它可以处理任何任务。你在午夜想到想要处理某件事。你把一个文件放进队列。凌晨2点它就被处理完了。等你醒来时，输出已经在等着你了。工作流5：项目健康监控器这个工作流每周一早上7点检查所有活跃项目。 Claude 提示词：读取 CLAUDE.md 获取活跃项目列表。读取每个活跃项目文件夹中的 overview.md 文件。读取过去7天内修改过的所有项目文件。对每个项目生成健康评估： STATUS: On track / At risk / Stalled / Blocked（状态：进展顺利 / 有风险 / 停滞 / 受阻） EVIDENCE: 具体表明此状态的证据。要具体。列出文件名、日期和具体观察结果。 NEXT ACTION: 该项目在未来48小时内最需要做的一件最重要的事。 FLAG: 任何超过7天没有活动的项目。这些需要人工关注。将健康报告保存至： GENERATED/briefings/[DATE]-project-health.md 对于任何被 FLAG 的项目：在 QUEUE 中创建一个名为 REVIEW-[project-name].md 的文件，附上情况的摘要。项目健康监控器能在问题演变成危机之前将其暴露出来。一个停滞了七天的项目会在周一早上被标记出来，而不是在截止日期到来时才在慌乱中被发现。 ## 防崩溃特性这个操作系统设计了三个特定的防崩溃机制。捕获安全网所有内容首先进入 00 - CAPTURE。捕获时不需要做任何决定。你永远不会因为不知道该把东西归档到哪里而跳过捕获。捕获文件夹会吸收一切，处理器稍后再进行归档。这意味着即使你忙得不可开交，系统依然完好无损地运转。在你完全没有时间处理任何事情的日子里，你仍然可以将内容捕获到 00 - CAPTURE。处理器会在当晚自动处理其余工作。永不删除归档规则这个系统中的任何东西都不会被删除。已完成的项目移动到 ARCHIVE。过时的参考资料移动到 ARCHIVE。已处理的捕获内容移动到 ARCHIVE。这意味着系统永远不会丢失信息。你不可能因为归档得过于激进而弄坏系统。保留所有内容的成本为零，因为存储空间是无限的，而且系统在日常运作中从不需要你浏览归档。 CLAUDE.md 单一信息源 Claude 运作所需的关于你生活的一切信息都存在于一个文件中。一个需要更新的文件。一个需要审查的文件。一个 governing 所有工作流的文件。当你的优先事项发生变化时，你更新 CLAUDE.md，之后的每个工作流都会自动反映新的情况。你不需要更新五个不同的文档，也不需要在五个不同的会话中重新解释你的情况。 ## 让你在一个周末内运行起来的构建顺序按照这个确切的顺序构建系统。抵制同时构建所有内容的冲动。周六上午：存储层创建八个文件夹。用你当前的真实信息设置 CLAUDE.md。为你最活跃的一个项目编写项目概述，并使用一致的属性。所需时间：2小时。周六下午：智能层安装 Claude Desktop。配置 Filesystem MCP。将其指向你的知识库。手动运行第一次晨间简报，并验证它能根据你的 CLAUDE.md 生成准确的输出。所需时间：1小时。周六晚上：第一个队列任务将一个包含真实任务的文件放入你的 QUEUE 文件夹。一个你一直想探索的研究主题。一个你一直在拖延的决定。一份你这周所做笔记的摘要。让 Claude 处理它。审查输出。在 CLAUDE.md 中记下需要改进的地方。所需时间：30分钟。周日上午：自动化层设置 N8N。将晨间简报工作流构建为定时任务。让它在计划时间运行，并验证输出自动出现在 GENERATED 中。所需时间：2小时。周日下午：其余工作流将捕获处理器、每周回顾生成器和队列处理器添加为额外的 N8N 工作流。所需时间：2小时。到周日晚上，所有五个工作流都已运行。操作系统已投入使用。 ## 永不崩溃实际上是什么样子构建这个系统三个月后，你会注意到一件事。系统一直在运行。不是因为你在完美地维护它。你并没有。有几周你几乎没碰过 Obsidian。有几天，捕获的内容在 00 - CAPTURE 中堆积了三天你才想起去处理。但晨间简报每天早上照常运行。捕获内容每天晚上照常被处理。每周回顾每个周日照常生成。项目健康监控器每周一照常标记停滞的项目。无论你是否在操作它，系统都在持续运转。这就是生产力工具与操作系统之间的区别。工具需要你去使用它。而操作系统会自动运行。这个周末就把基础搭建起来吧。操作系统从上线后的第一个早晨就开始运转。关注 @cyrilXBT，获取精确的 CLAUDE.md 模板、N8N 工作流配置，以及让整个架构得以运行的 Filesystem MCP 设置。 ## 相关链接 - [CyrilXBT 转发了](https://x.com/cyrilXBT) - [CyrilXBT](https://x.com/cyrilXBT) - [@cyrilXBT](https://x.com/cyrilXBT) - [16K](https://x.com/cyrilXBT/status/2056924424838815824/analytics) - [@cyrilXBT](https://x.com/@cyrilXBT) - [升级到 Premium](https://x.com/i/premium_sign_up) - [2026年5月20日上午10:26](https://x.com/cyrilXBT/status/2056924424838815824) - [16.1K 次浏览](https://x.com/cyrilXBT/status/2056924424838815824/analytics) - [查看引用](https://x.com/cyrilXBT/status/2056924424838815824/quotes) --- *导出时间: 2026/5/20 14:40:32*

相关文档