The vault: numbered categories
You cloned vault-template/. You have twelve numbered directories staring at you, and most of them don't quite fit your domains. This chapter is how to adapt them without breaking the model.
The default twelve
| # | Name | What goes here |
|---|---|---|
| 01 | operations | SOPs, runbooks, dashboards, weekly reviews |
| 02 | finance | Banking, tax, P&L, cashflow, insurance |
| 03 | people | Team, family, household, key relationships |
| 04 | resources | Vendors, suppliers, subscriptions, service providers |
| 05 | legal | Contracts, compliance, IP, entity docs, I-9 records |
| 06 | marketing | Brand, content, channels, campaigns |
| 07 | revenue | Sales pipeline, customers, pricing, retention |
| 08 | relationships | Network, mentors, peers, high-touch clients |
| 09 | automations | Cron, n8n, webhooks, scheduled scripts |
| 10 | tools | Code, dashboards, internal tools, PRDs |
| 11 | strategy | Annual plans, quarterly reviews, strategic bets |
| 12 | daily-log | One file per working day |
You don't have to use all twelve. You don't have to keep these names. But you should keep the numbered structure even if categories are empty for you.
Renaming categories
If you don't have employees, 03-people might become 03-household or just stay 03-people with only family inside. If you don't sell to customers, 07-revenue might become 07-output (for your investing thesis output) or simply stay empty.
Rename the directory and update its README to match what you put inside. Update CLAUDE.md so your AI tools know the new naming.
What you should NOT do: collapse two categories into one, or add a thirteenth. Twelve is the right number because it covers every domain most operators run while still fitting on one screen. More than twelve gets ambiguous. Fewer leaves some things with nowhere to live.
When a document belongs in two places
It happens. A signed vendor contract is both 04-resources/vendors/[vendor]/contracts/ AND 05-legal/contracts/. A new-hire offer letter is both 03-people/onboarding/ AND 05-legal/contracts/.
The rule: pick one location, put the file there, and add a one-line pointer in the other:
Files that exist in only one place are easier to maintain. Pointers go stale; the file does not.
Empty categories
Some categories will be empty for the first several months. That's normal. Don't delete them. Empty categories are placeholders that signal future fit.
The exception: if a category has been empty for 6 months and you can't think of anything that would ever go in it, rename it to something more relevant or merge its purpose into another.
The _loop/ subfolder
Every numbered category has (or should have) a _loop/ subfolder:
This is the continuous-improvement engine for each domain. Most days you'll add nothing. Most weeks you'll add one or two files. Over a year, _loop/ accumulates into the most valuable knowledge in the vault.
We'll come back to loops in chapter 07.
Naming files inside categories
- Lowercase, kebab-case.
- Date-prefixed when temporal:
2026-05-20_lease-amendment-208-e-main.pdf - Topic-prefixed when not:
vendor-evaluation-criteria.md - Spaces are forbidden. Underscores are okay. Hyphens are preferred.
Why this matters: AI tools navigate the vault by file name. Consistent naming means consistent navigation. Random naming means the AI guesses, and guessing produces broken links.
The two special directories
_AI/ and _System/ start with underscores. They sort to the top, marking them as "system files, not domain content."
_AI/holds cross-domain intelligence: analyses spanning multiple categories, your AI model inventory, schedules._System/holds drive management: cron scripts, hooks, slash command directories.
You'll touch _AI/ rarely. You'll touch _System/ when adding new automations.
The workspace/ directory
This is the brain (Layer 2). It contains BRAIN.md, MEMORY.md, CHANGELOG.md, PLAYBOOK.md, VOICE.md.
It sits at the root of the vault, not under any numbered category. The reason: the brain isn't a domain. It compresses across all of them.
The default template names it workspace/ (no number prefix) so it doesn't pretend to be a domain. If you want it to sort first, rename it 00-workspace/. Either works.
Migrating from your current setup
You probably have documents in Notion, Obsidian, Google Drive, Dropbox, iCloud, or somewhere worse. The migration plan:
Set up the vault. Don't migrate anything yet. Start writing today's documents directly into the vault.
Migrate the documents you reference most often. Bank statements, active contracts, current customers, current vendors. Don't migrate archives.
Migrate older documents only when you actually need them. Most archives sit unused. Migrating them is wasted effort.
The trap: people try to migrate everything in a weekend. The migration takes a month and they exhaust themselves before the system is established. Migrate as you go.
Customization vs. discipline
You can customize the vault. You should customize the vault. But every customization has a cost: future-you forgets the customization, and your AI tools don't know about it.
The defaults are opinionated. Run them for 30 days before customizing. Most customizations people propose on day 1 (rename five categories, restructure the subfolders, add a thirteenth) get reversed by day 21 because the defaults turn out to be reasonable.
If you're going to customize, document it in workspace/MEMORY.md under "Vault customizations" so future-you remembers what was changed and why.
The first thing to put in the vault
The most-used document in your operation. The thing you reference weekly.
For me, it was a vendor invoice file. For Paula, it was the household chore schedule. For Bayes-Invest-prototype, it was the advisor persona document.
Drop one document into the right category. Notice how it feels. The harness becomes real when something inside it is actually useful, not when the structure is "complete."
Next chapter: the brain — the five documents that compress everything else.