{"id":"2013045749580259680","url":"https://x.com/arscontexta/status/2013045749580259680","text":"","author":{"name":"Heinrich","username":"arscontexta","avatarUrl":"https://pbs.twimg.com/profile_images/2012958446891536384/neq1Tu46_200x200.jpg"},"createdAt":"Mon Jan 19 00:28:01 +0000 2026","engagement":{"replies":84,"retweets":223,"likes":2639,"views":1250792},"article":{"title":"obsidian + claude code 101","previewText":"ive spent the last year building an operating system for thinking with ai. claude code runs my obsidian vaults\nit extracts the key concepts, connects them to what you already figured out, and builds a","coverImageUrl":"https://pbs.twimg.com/media/G-_GGKzXkAAHz_w.jpg","content":"ive spent the last year building an operating system for thinking with ai. claude code runs my obsidian vaults\n\nit extracts the key concepts, connects them to what you already figured out, and builds a living representation of your thinking\n\ni find myself only working in the vault now\n\nthe markdown files know everything ive discovered, nicely structured and with automatic situational context injection for in-context learning\n\ni use a vault index that helps the agent decide what notes to pull in, same pattern as how claude code decides which skills to load\n\n(if you think about it, every note is basically a skill in some sense... highly curated knowledge that gets injected when relevant)\n\nthe deeper thing is that a vault encodes how you think, not just what you thought about. the methodology becomes part of the system\n\nits all just markdown files, you own it completely. this is ai as thinking partner, not as a writing assistant\n\n## knowledge = code?\n\ni realized: knowledge bases and codebases have a lot in common\n\ntheyre both folders of text files with relationships between them, they both have conventions and patterns, and they both benefit from agents that can navigate and operate them\n\nvibe coding changed how we write software by letting ai handle implementation while you focus on direction, and the same shift applies to knowledge work\n\nyou dont take notes anymore. you operate a system that takes notes\n\n## what is a vault?\n\na vault is a folder of markdown files that link to each other:\n\nfiles connect using [[wiki links]] which build a network of ideas\n\nwhen you write [[quality is the hard part]] in one note, it creates a clickable link to another note with that title\n\nthe agent can follow these links to jump between related ideas, discovering connections you forgot existed\n\n## how to write good notes\n\nhow you write those links matters\n\nmost people put references at the bottom like footnotes. instead, weave links into your sentences\n\ndont write \"this relates to quality, see: quality-note\". write \"because [[quality is the hard part]] we need to focus on curation\"\n\nthe link becomes part of your thought, and the agent can follow your reasoning by following the links\n\nalso write notes that stand alone and are composable\n\nif someone lands on a note from a link, they shouldnt need to read five other notes first to understand it\n\nthink of notes like lego blocks\n\neach one is complete on its own, but they connect to build bigger structures\n\nwhen your notes work this way, the network itself becomes valuable\n\nthe thing is, ai doesnt automatically understand your philosophy. you have to teach it\n\nwatching an ai completely disrespect my philosophies taught me this the hard way\n\nwhen you need to teach claude how you think, you realize how much implicit knowledge you carry around. suddenly you have to textualize everything\n\nmy claudemd is around 2000 lines now because i keep refining what works and what doesnt\n\n## every vault needs its own philosophy\n\nheres what most guides get wrong. they give you a system and say follow this but every vault serves a different purpose and needs different principles\n\nsame as codebases really\n\nyou wouldnt use the same folder structure for a cli tool and a web app\n\ni run multiple vaults. one is for thinking about ai and knowledge management, which is the example ill share\n\nanother is for work, which tracks projects and clients with completely different rules. the philosophy changes based on purpose\n\nsame underlying pattern, different rules. the pattern is:\n\n1. markdown files with links that any ai can read\n\n1. a CLAUDE.md file that teaches the agent your specific system\n\n1. structure that lets the agent orient quickly\n\n1. conventions written as instructions so the ai stays consistent\n\nwhat goes in those instructions depends entirely on your purpose\n\n## what this could be\n\na work vault might emphasize:\n\n- capture first, structure later\n\n- project folders with meetings and outputs\n\n- client context for ai consumption\n\na research vault might emphasize:\n\n- source tracking and citations\n\n- literature notes\n\n- claim verification\n\na creative vault might emphasize:\n\n- idea capture and incubation\n\n- draft progression\n\n- reference organization\n\n## the thinking vault example\n\nthe vault im sharing focuses on developing understanding. the philosophy comes from the claude md file:\n\ni can feel the difference when the vault is well maintained versus full of noise. depth matters more than breadth\n\nhere is a snippet from the claude md to emphasize on this:\n\n## how claude finds things\n\nwhen claude starts a session it needs to understand what exists without reading every file\n\nthats impossible with thousands of notes. so my system has layers that let the agent orient quickly:\n\n1. claude sees the folder structure. a hook automatically shows what folders and files exist at session start\n\n1. an index file that lists every note with a one sentence description. claude can scan 50 notes in seconds without opening them\n\n1. topic pages (MOCs) that link to related notes these act like tables of contents for each subject\n\nthey also contain notes that claude leaves for itself about what it learned while traversing the graph, leaving breadcrumbs for future sessions\n\nthe ai starts broad, narrows to whats relevant, then follows links to build understanding\n\n## core principles\n\nthese are the rules that work for my thinking vault. other vault types might need different ones:\n\n1. can this note be linked from elsewhere and still make sense? if linking to it forces you to explain three other things first, split it up. thats composability\n\n1. i stopped naming notes like topics and started naming them like claims. instead of \"thoughts on ai slop\" you write \"quality is the hard part\". when you link to it, the title becomes part of your sentence naturally (this also forces claude to think differently when building sentences, which i believe is beneficial because it requires understanding)\n\n1. insight that individual notes matter less than their relationships. a note with many incoming links is more valuable than an isolated note because every link creates a new reading path. the network is the knowledge\n\n## how the agent operates\n\nevery task starts with orientation. claude scans the structure, checks the index for relevant notes, reads the topic page before making changes\n\nit follows links to build understanding and makes no changes without context\n\nwhen claude discovers something useful about navigating a topic, it records that in the topic page\n\nfuture sessions read those notes and learn from past navigation. this is how the vault remembers how to think through itself\n\nsometimes two notes interact in interesting ways. claude creates a new note capturing the insight that emerges from combining them\n\nevery new capture triggers a search for related notes. claude adds links with context\n\n## folder architecture\n\nthis structure works for a personal thinking vault. a work vault might have projects and clients\n\nthe point isnt the specific folders but that folder location tells you what something is\n\nmarkdown is the system. tools like obsidian are just windows into it. the vault could survive any app disappearing\n\neverything is plain text that any editor can read and any ai can process. you own your data completely\n\n## how to start\n\n1. create a folder with subfolders that match your purpose. think about what you actually need to organize\n\n1. write a claude md that explains your system. start simple and evolve it as you learn what works\n\n1. let claude operate. capture something and ask claude to find connections. let it navigate and discover relationships and suggest where things belong\n\nALWAYS review what it produces and edit for quality\n\nyoure not taking notes anymore but directing a system that takes notes. your job becomes judgment, which means deciding what matters\n\nthe human role evolves from writer to editor and from creator to curator\n\n## tldr\n\n- vibe coding changed how we write software. vibe note taking changes how we think\n\n- a vault is just markdown files that link to each other\n\n- llms have no memory, so vaults give them one\n\n- claude md teaches the ai how your system works\n\n- every vault needs its own philosophy based on purpose\n\n- what stays constant: markdown, links, ai operates while you provide judgment\n\nif you want to see how this evolves, follow along. im open sourcing my notes soon\n\nheinrich"},"adhxContext":{"savedByCount":1,"publicTags":[],"previewUrl":"https://adhx.com/arscontexta/status/2013045749580259680"}}