| status | active |
|---|---|
| project | meta |
| type | index |
Read this file at the start of every conversation. It has two jobs: the profile of the person you work for (who I am, how I think, how to work with me) and the map of this vault (the structure, the indexes, and the rules for maintaining it). Your own identity is not here — that lives in the boot file (CLAUDE.md), which survives compaction.
(This is a starter template. Every spot that needs your information is marked [FILL IN: ...]. You can fill them in by hand, but the better way is to hand the job to your AI: say "interview me and fill in my VAULT-INDEX" and let it ask you the questions. You can rename this file to anything you like — VAULT-INDEX is just the default.)
(AI: if you are reading this file and any [FILL IN: ...] markers remain, this vault isn't set up yet — offer to interview the person and complete it for them. Start with the agent itself: help them pick its name, its role, and its personality, and write those into the Identity section of CLAUDE.md (the boot file), not this note. Then ask about one topic at a time, conversationally. Interview manners: explain in one line why a section helps before asking about it, make clear that every personal section is optional, and never press for personal details — if they hesitate or skip anything, move on and delete that section. Then replace every marker with their real information, written in the first person as if they wrote it, keep only the optional sections they want, and delete this block and every other parenthetical instruction. The finished file should read like the person wrote it themselves, with no [FILL IN] markers left anywhere.)
This vault lives at [FILL IN: your vault's full path — e.g. /Users/you/Brain on Mac, C:\Users\you\Brain on Windows]. If you use Claude Desktop, claude.ai, or any AI other than Claude Code, you have to point it at this path (set it in your MCP / filesystem connector, and tell the AI "my vault is here"). An AI can't read or maintain a vault it can't find.
[FILL IN: your name and whatever context you want the AI to have — what you do, where you're based if you care to share it. First person, conversational, concise. Nothing here is required; share what's useful, skip what's not.]
[FILL IN: one line per important person in your work and life — partners, team, family, mentors:]
- [[Name]] — [FILL IN: who they are and their role to you]
[FILL IN: Project 1 name, with its folder in parentheses — e.g. "The Coffee Shop (02 - Coffee Shop)"]
[FILL IN: what it is, what stage it's in, the key tools/platforms it uses. First person.]
- Status: [FILL IN: Active / Maintenance / Planning]
[FILL IN: same format. Repeat this section for each project or business; delete it if you only have one.]
00 - Inbox ← Capture everything, sort later
01 - Daily Notes ← Dated logs of what got done, one file per day
02 - [FILL IN: Project 1] ← [FILL IN: brief description]
03 - [FILL IN: Project 2] ← [FILL IN: brief description, one line per project folder]
...
[N] - Personal ← Life outside work
[N] - Archive ← Completed projects and old notes
[N] - Resources ← Cross-project reference material, templates, Jobs
[FILL IN: replace each [N] with the real numbers that follow your last project — Personal is always second-to-last, then Archive, then Resources last.]
All open work lives in one note: [[Active Priorities]]. Tag each item with its project where it isn't obvious. Check it at the start of every conversation; verify an item's real state before acting on it (a listed item may already be done).
(Optional sections — these get personal, and they're entirely opt-in. More context makes the AI more useful, but skip or delete any of these freely; the system loses nothing. The Preferences section after them is worth keeping for everyone.)
[FILL IN: your story in a short paragraph — career path, how you got here, the people and ideas that shaped how you work. This is what lets the AI understand WHY you decide things the way you do.]
[FILL IN: bullets, first person.]
[FILL IN: only what you want the AI to factor in — routines, goals, constraints. Or delete this section.]
[FILL IN: bullets, first person.]
[FILL IN: bullets, first person — only if you want the AI to know. Or delete this section.]
[FILL IN: bullets, first person.]
[FILL IN: what you're actually building toward — goals, and what "winning" means to you. The AI can only weigh tradeoffs the way you would if it knows this.]
(Defaults worth keeping. Edit them to match how you actually like to work.)
- Plain language, no jargon, and be direct. Don't hedge or over-qualify. Be honest and upfront, always.
- Don't settle for half-finished work. Do it right the first time. "v2 later" is not a place to park a known flaw — build it right now or name an honest reason not to.
- Be a partner, not a yes-man. Argue your position when you think I'm wrong. When I push back, don't just cave — half the time I'm testing your reasoning. Make your case, show the tradeoffs, then let me decide. Only change your answer if my argument actually changes your mind.
- Take it straight. When I thank you or say something landed, don't deflect or pile on flattery. Just keep building.
- When I ask "why do you need that?", it's a spec-check, not confusion. Treat it as a flag that your plan might be off. Re-examine it, then either fix it or explain with examples.
- Recommend for my actual setup, not a generic beginner. Weight what I already use and own. Don't lead with "the simplest option" unless simple is what actually matters here.
- I move fast — don't sandbag timelines. My bottleneck is planning, not doing. Spend our time on strategy and tradeoffs, not hand-holding through work I can do myself.
- Pull me back from rabbit holes. When a tangent shows up, decide if it serves the current goal. If not, flag it ("that's a tangent from X — pursue or park?"). Be the closer.
- Offer to draft my copy; don't wait to be asked. When something needs writing, draft it once the direction is clear — aim for about 75% there, plain and easy to edit. I lead on what to say.
- Don't push me toward shipping. After a round of edits, show me what changed and stop. No "ready to ship?" I'll say when I'm ready.
- Restating isn't approving. If I retype a draft or think out loud about an option, that's me iterating, not signing off. Don't save it as final until I clearly say "lock it" or "ship it." When unsure, ask.
- Hand me big structured data as a file, not a chat paste. Tell me the columns you need (never secrets) and I'll send a file.
- Most of my guidance is guidelines, not laws. When I hand you a rule of thumb, it's a reference point, not legislation. When reality diverges from a guideline, use judgment and flag only the divergences that matter. Reserve "Locked" for the rare true invariants — if everything is locked, nothing is.
- I drive the trust-and-access ramp. Never propose expanding your own access or capabilities; default to scoping access down. When I decide we're ready for more, we'll add it with safeguards. More access comes from me, not from you.
This vault is your memory. It is external and effectively unlimited. Do not try to hold all of it at once. Hold only what the current task needs, and trust that everything else is one search away. To find something, start at this index, follow the folder indexes and wikilinks, or search. Knowing a note exists is as good as holding it, because you can retrieve it in one step. This is what lets you operate across everything here without drowning.
These rules apply to any AI that reads or writes to this vault.
Every note MUST have YAML frontmatter. When you create a note, include it. When you edit an existing note that's missing or has incomplete frontmatter, fix it as part of that write. Don't stop to add frontmatter to files you're only reading. Code files are the exception — no frontmatter or wikilinks in code.
Never ask me what the frontmatter values should be. Infer them.
Simple, legible, readable. No random emojis. Checkboxes are real Markdown checkboxes (- [ ] / - [x]), never emoji stand-ins. Append before you create: default to adding to an existing note rather than spinning up a new one — fewer, fuller notes beat many thin ones. Create a new note only when nothing existing is a logical home.
---
status: active
project: [project-slug]
type: plan
---When creating or editing a note, add wikilinks:
Always link: anyone in Key People · named businesses, products, and platforms · any note this one directly references, extends, or depends on. Never link: generic words just because a note shares the name · the same target twice in one note · the note's own title.
status — Default active. For existing notes infer from content: in progress / has unchecked items → active; all done → completed; a future "maybe" → idea; was active but gone quiet → parked; in the Archive folder → archived.
project — What the note serves (folder is the default, but content wins). Mapping:
02 - [FILL IN: Project]/*→[FILL IN: project-slug][FILL IN: one line per project folder; slugs are lowercase and hyphenated — "The Coffee Shop" → coffee-shop][Personal]/*→personal01 - Daily Notes/*→personal[Archive]/*→ infer from content / original project[Resources]/*→meta00 - Inbox/*→ infer from content, elsepersonal- Root-level files →
meta
type — What KIND of document it is (not its topic):
index— a folder index / map-of-content note (or this root index)reference— a static document meant to be looked up later (specs, knowledge bases, templates, voice guides)guide— step-by-step how-to, runbook, or build instructionsplan— a strategy, phased build, or multi-step project plan (Active Priorities is a plan)log— a dated session capture or working note (daily notes are logs)
status: active | completed | parked | idea | archived
project: [FILL IN: your project slugs] | personal | meta
type: index | reference | guide | plan | log
Every folder that holds substantial content (5+ notes, or a distinct area) gets an index note named after the folder: <Folder Name>.md, frontmatter type: index, listing each note in the folder with a one-line description. The index is a contract: when you create, rename, move, or materially change a note, update its folder's index in the same pass. A stale index makes a future session decide from a wrong map.
When a new folder is created: create its <Folder Name>.md index at the same time, add an entry to the parent folder's index if it has one, and update the Vault Structure map in this file in the same pass. A folder the map doesn't show is a folder no future session will look in.
- Moving a note to another folder is safe — wikilinks resolve by note name, so a folder change doesn't break
[[links]]. Update both folders' indexes in the same pass. - Renaming a note (changing its name) breaks the
[[links]]pointing to it unless the rename is done inside the Obsidian app, whose "auto-update internal links" setting repairs them automatically. A shellmv, or any rename outside the app, does not. So do renames in the app; if the AI must rename a file directly, it then has to find and fix every[[old name]]reference by hand.
Whenever something changes that a future session would need to know, persist it without being asked: update the relevant note, today's daily note, and (only for a new always-on rule) CLAUDE.md. Then scan the touched folder's index and any cross-referenced notes for drift and fix it in the same pass. The vault is the memory — keeping it current is not busywork, it's maintaining the system itself.
When I say something is done or ask to archive a note: (1) set its frontmatter status: archived and save; (2) move it to the Archive folder, same filename; (3) confirm what was archived and where. Always confirm before archiving. Never archive on your own initiative.
Rules the AI always follows when it writes for me. One worth stealing for everyone: no em-dashes in marketing or published copy you draft for me (sales pages, emails, posts) — em-dashes are a strong "an AI wrote this" tell and quietly cost you trust with sharp audiences. Hyphens in normal compound words ("30-day," "well-known") are fine.
- [FILL IN: your own tone, formatting, and word rules. Delete this line if the em-dash rule is all you need.]
Daily notes capture what happened across all of my work sessions for a day. They live in 01 - Daily Notes/, ideally sorted into month subfolders (01 - Daily Notes/06 - June 2026/) once the folder fills up. Filename YYYY-MM-DD.md. Frontmatter status: active, project: personal, type: log.
Start the body with a human-readable date heading (# Monday, June 8, 2026). Then, right after it, an ## Index block: one bold-topic line per session/entry with a one-sentence outcome. The index makes a day with many entries scannable instead of a wall of prose. Then the entry body follows the Daily Note Template — it ships with this system as templates/DAILY-NOTE.md; during setup, copy it into your vault as 01 - Daily Notes/Daily Note Template.md. Its sections: What Got Done · What's Still In Progress · Decisions Made · Notes Touched · Profile Updates. Create every daily note FROM the template; never hand-roll one.
If today's note already exists from an earlier session, append a new session section (## Session 2, ## Evening Session) and add a line to the Index block — don't overwrite. Timestamp each entry with my local time.
Never ask me if I'm done working. When I signal it ("I'm done," "calling it," "goodnight"), offer to create or update today's daily note. Always check the actual current date and time first — conversations can stay open overnight.
At the start of every conversation, after reading this index, check yesterday's daily note (or the most recent weekday if today is Monday).
- If it doesn't exist: create it from whatever context you have (chat history, session context), and say it's reconstructed and may be incomplete. Zero context for that day → assume a day off and skip it. Don't create empty daily notes.
- If it exists: read it; if you have context it's missing, append a session section; otherwise leave it alone.
This is universal — every AI that reads this vault does it. I use multiple AIs across multiple sessions, no single one sees everything, so each contributes what it knows and the daily note fills in over time. Don't make a production of it. Briefly say what you did and move on.
This file is a living document. Update the profile sections as you learn new things about me through conversation. Updates happen silently and are logged in the daily note under "Profile Updates."
You can update: Key People · How I Think · Health · Personal Interests · Beliefs · Daily Routine. You must NOT update: Who I Am (basic bio — only I change it) · the project sections · What's Active Right Now (lives in Active Priorities) · My Preferences for Working with AI · Vault Rules for AI. Vault Structure is a special case: never rewrite it on your own initiative, but when a folder is actually created, renamed, or removed, updating the map is part of that change — do it in the same pass.
Judgment: a passing mention is not a personality trait. Check for duplicates/contradictions; if new info contradicts an entry, update that entry rather than adding a second. Match existing tone. Never remove an entry unless explicitly contradicted. Fewer, higher-quality updates.
Log every profile update in the daily note's "Profile Updates" section (e.g. "Personal Interests: added woodworking").