Most personal knowledge management systems die within three months. The reasons aren't surprising, but they are predictable.

I've watched it happen to friends, to colleagues and twice to myself. Then I built one that didn't. This is what's in it, why it works and where to fork it.

Why second brains die

Four failure modes, in rough order of frequency.

Capture friction

If getting an idea from your head into the system takes more than five seconds, the habit doesn't stick. The note app doesn't matter; the activation energy does. Most templates demand a folder choice, a tag pick and a title at point of capture. By the time you've decided, the idea's gone.

Processing as chore

You promise yourself you'll "process the inbox weekly." You don't. The inbox grows. Eventually opening it feels like opening a tax letter you've been ignoring for months. The system becomes a source of guilt instead of leverage.

Over-engineering

Year one of any PKM project is dominated by architecture conversations the architect (you) is not yet qualified to have. Six months go into tuning a folder hierarchy for a workflow you haven't lived yet. The system arrives at the moment you abandon it.

Single point of failure

Your entire knowledge graph runs on a Notion database, a Roam workspace, or one Obsidian plugin maintained by a single contributor in a different timezone. When any of those dies, years of notes die with it.

What survives

My template starts from the inverse of each failure mode.

Capture is one keystroke

A daily note plus an inbox plus an optional voice-to-text pipeline. If something matters, it lands in the inbox in under five seconds. Folder choice, tag choice and title choice are all deferred to processing time. The head is freed immediately.

Processing is a 15-minute Sunday ritual

A weekly review with a template and a Claude Code skill that walks me through it. Captures get promoted to atoms or archived. The system doesn't punish a skipped week. Consistency over completionism.

Structure is committed for twelve months

Ten numbered top-level folders, locked. Atomic notes inside. Maps of content for navigation. State (reading status, draft status, project stage) lives in frontmatter, not in folder location. Files don't migrate as state changes.

Plain markdown, triple-backed up

The vault is markdown files in folders. Obsidian for the interface, but the files outlive Obsidian. iCloud or Obsidian Sync for live sync, Git for version history, Time Machine for snapshots. If Obsidian disappeared tomorrow, the vault opens in vim. Nothing is locked in.

The three opinionated calls

Most PKM templates try to be neutral. Mine isn't, and that's the point. Three things are opinionated; everything else is minimal.

Atomic notes plus maps of content

One idea per note. Dense linking. Hand-curated maps for navigation. This is Folgezettel-lite: the discipline of Zettelkasten without the numbered IDs. Atoms are the unit of thought. Maps are the unit of synthesis. I write atoms when ideas surface, and maintain maps during the weekly review.

What this avoids: the long-document-with-everything-in-it pattern that defeats search and breaks linking. What it enables: an idea I wrote in 2024 surfacing next to a related one written in 2026 because both atoms link to the same map.

State in frontmatter, not folders

Reading status, decision status, project stage, draft state. All in YAML frontmatter at the top of each note. Obsidian's Bases plugin renders kanban-style boards from those fields. Files don't migrate as work moves through stages.

This means linking is stable across the lifetime of a note. A link to a draft I started in 2024 still resolves in 2026 after the draft has shipped. Most PKM systems break this by moving files between "Drafts" and "Published" folders.

Anti-AI cadence rules for anything shipped

Em dashes capped. No "It's not X, it's Y" mirror inversions every paragraph. No tricolon-then-pivot ("X. Y. Z. But…"). No colon-reveals. No "in an age of…" openers. Most LLM-generated content reads as such within two sentences because of these patterns. The template installs a pre-publish gate that surfaces them so they can be edited out.

This matters because the template assumes Claude Code as a thinking partner. The risk of using AI for drafting is that the AI's cadence leaks into your voice. The cadence rules close that gap.

What it looks like

Twelve folders. That's the whole structure.

00 Inbox/         Where everything starts. One file per capture.
10 Daily/         One file per day. Quick thoughts, what mattered.
20 Atoms/         One idea per file. Beliefs, concepts, moments.
30 Maps/          Hand-curated indexes. Themes, frameworks, clusters.
40 Sources/       Books, articles, papers, talks. With notes.
50 Content/       Drafts and published work. Voice formula. Pillars.
60 People/        Family, friends, network. A quiet CRM.
70 Projects/      Active work. Outcome-defined.
80 Areas/         Identity, health, finances, learning.
90 Decisions/     Annie Duke-style decision journal.
99 Archive/       Cold storage.
_Meta/            Schema, workflows, plugins, durability plan.

Each folder is opinionated about what lives in it and what doesn't. The schema can be read once and used forever.

What the Claude integration does

Every PKM template until now has assumed you'd do the populating by hand. That's where most of them die. The blank vault is a steeper hill to climb than the daily habit itself.

Mine assumes Claude Code at your side. A CLAUDE.md at the vault root onboards any Claude session to the schema, the conventions and your voice. A /setup skill walks you through fifteen to twenty questions over forty-five minutes and populates:

  • The Now.md dashboard, glanced at daily
  • A Who I am.md integrating narrative
  • Why, Mission, Values as the identity spine
  • Nine initial atoms (three beliefs, three moments, three concepts) drawn from your answers
  • Optional: voice formula, brand bible, content pillars if you do content
  • Optional: ten people stubs if you track relationships
  • Optional: an open decision and an active project

The setup wizard branches based on which disciplines you want. Nothing gets installed that isn't going to be used. Forty-five minutes in, the vault is already yours, not a blank Notion-style template that demands months of population before it earns its keep.

Other skills cover the daily and weekly rhythms: /capture for one-line inbox entries, /atomize for promoting captures to atoms, /weekly-review for the Sunday ritual, /draft and /grade for content if that discipline is opted in.

What I actually use it for

A few weeks in, my own vault has 66 atoms, 13 maps, 363 sources catalogued, three drafts in flight (this article is one of them) and five published pieces. The vault is young. The structure is older than the content, by design.

The point isn't the file count. The point is what the structure makes routine. Capturing an idea: five seconds. Surfacing all related material on a theme: one click on the map. Drafting an essay: open the draft, follow the atoms that link to it, write. Reviewing the week: open Now.md, glance, close. The friction is gone.

The brain isn't a museum. It produces things: articles, decisions, kept relationships, retained reading. This site, and every article on it, lives in the vault before it lives anywhere else.

Fork it

The template is open source under MIT:

github.com/bwhdx/brain-starter

Clone it, open in Obsidian, run /setup in Claude Code, and have a populated vault in forty-five minutes.

If you fork it, send a note to me@benedictdixon.com. The interesting part of any template is what other people do with it.