brain-starter

A working personal second brain you can fork and make your own. Markdown + YAML, Obsidian-compatible, Karpathy-flavoured external cognition. Comes with a setup wizard, thorough by default and interruptible at any point, that walks you through populating it via Claude Code. Usable after the first sitting, truer every time you come back to it.

What this is: a battle-tested vault structure for capture, distillation, retrieval, content creation, decision journalling, and quiet relationship tracking — designed to compound over decades without methodology thrash.

What this isn't: a Notion template, a productivity app, or an AI chatbot. It's plain markdown files in folders. Anything you write here outlives any tool.

Why it exists

Most personal knowledge management systems die within three months. They die for predictable reasons. Capture is too high-friction, processing becomes a chore, the structure was over-engineered before you knew what you needed, or the plugin you depended on broke.

This template fixes the predictable failures up front.

• Capture is one keystroke. Daily notes + inbox + (optional) voice-note pipeline.
• Processing is a short Sunday ritual with a template and a /review skill that walks you through it.
• Structure is committed for ≥12 months. Ten numbered folders, atomic notes, MOCs for navigation, state in frontmatter. The architecture is finished; only the content evolves.
• Plugin discipline. Bases (Obsidian core) is preferred over Dataview wherever it covers the need. Cap ~10 community plugins. Everything else is plain markdown.
• Designed to survive plugin death, Obsidian death, your-laptop death. Triple-backup: Obsidian Sync + Git + Time Machine.

Where this came from

This template is the working vault behind a short series on owning your AI stack. Reading them is optional; the template stands alone.

1. How to set up Claude Code — forthcoming
2. The assistant that's always on — running a personal assistant on your own machine
3. Building a Second Brain That Doesn't Die — this repo
4. The assistant that knows you — pointing one at the other

You need only this repo. Clone it, run /setup, and you have a second brain you read and write yourself. See PAIRING.md if you later want the optional assistant half.

What you get

Structure

• 10 numbered top-level folders + _Meta infrastructure + _Wiki (active Karpathy LLM-wiki — agent-compiled synthesis layer)
• 21 Templater templates covering every note type, and the Obsidian config to make them fire — .obsidian/ ships in the repo, so Daily notes and Templates are already pointed at the right folders
• 12 Bases views ship as .base files, not just documented queries — reading boards, goal boards, decisions due for revisit, recent atom seedlings, people follow-ups, and more
• Generic schema spec, workflow doc, plugin list, durability plan

Identity, roles, and goals

• Values.md runs Covey's centre diagnostic — what you actually fall back on under pressure, not what you'd like it to be
• Roles are load-bearing, not decoration: goals hang off a role via a roles: field rather than floating free
• A weekly plan built as roles × goals, plus a renewal rhythm so the whole chain gets revisited instead of set once and left
• A learning architecture: a goal's skills_needed: links out to a plan for closing the gap

Toolkit taxonomy

• 40 Sources/Frameworks/, Checklists/, References/ — three source kinds alongside books, articles, talks, and podcasts, each indexed by its own Map
• The function-beats-medium rule: a book you read is kind: book; the same book, if you run its method, is kind: framework

Disciplines you can opt into

• Capture & learning (everyone) — daily notes, inbox, atoms, sources, theme maps
• Content creation (opt-in) — voice formula, pre-publish gate, vocabulary, pillars, channels, content process, cadence
• People (opt-in) — quiet CRM for relationships that matter
• Decisions (opt-in) — Annie Duke-style decision journal with retrospectives
• Projects (opt-in) — defined-outcome work tracking

Claude integration

• CLAUDE.md at vault root onboards any Claude Code session to the schema, conventions, and your voice
• Twenty skills ship as proper skill directories, so Claude Code actually loads them.
  • Vault skills — /setup (the wizard), /ask, /decide, /capture, /process-inbox, /distill, /review, /draft, /grade, /wiki-maintainer.
  • Assistant / daily-loop skills (for the optional always-on half, see PAIRING.md) — /think, /triage, /plan-day, /wind-down, /commitments, /follow-up, /resurface, /research, /research-digest, /scan-deadlines.

Agent layer (best-of-both)

• You own distillation — 20 Atoms/ and 30 Maps/. Agents read and link, never overwrite.
• Agents own compilation + memory — _Wiki/ synthesis pages compiled from 40 Sources/, an index.md catalog (read first), a log.md change history, and on-device search (_Meta/search.md, qmd). /wiki-maintainer compiles the wiki and updates index + log in one pass, and never touches your Atoms or Maps.

Quick start

git clone https://github.com/bwhdx/brain-starter ~/Brain
cd ~/Brain
claude

Then in Claude:

/setup

That runs the setup wizard. It's thorough by default and stops to check in after each phase, so you can go straight through or come back to /setup later — it picks up exactly where you left off.

• Your Now.md dashboard populated
• A full identity stack, covering Why, Values (including Covey's centre diagnostic), Mission, and your ten roles
• Goals linked to those roles, plus a first weekly plan
• Your first 9 atoms (3 beliefs, 3 moments, 3 concepts)
• Your Who I am integrating narrative
• Optionally, your voice formula, brand bible, and pillars (if you do content)
• Optionally, 5-10 people stubs, a first project, and a first decision (if those modules matter to you)
• A Weekly review template and reminder
• A private GitHub repo (if you want one) for offsite backup

After the wizard, see SETUP.md for next steps (the three community plugins that still need a manual install, sync setup).

Use cases

• Anyone who wants their thinking to compound instead of scatter — atomic notes, dense links, a structure that gets denser rather than messier over the years
• Researchers and learners — read with intent, distill what you learned, find it again years later
• Decision-heavy roles — decision journals you'll actually re-read because they're queryable
• Anyone who publishes, in whatever form — the anti-AI cadence rules catch the tells before you hit send
• Anyone tired of Notion drift — local-first, markdown, survives any tool

Philosophy

This template is opinionated about three things and minimal about everything else.

Opinionated:

1. Atomic notes + MOCs for navigation. One idea per note, dense linking, hand-curated maps. Folgezettel-lite without the IDs.
2. State in frontmatter, not folder location. Reading status, draft status, project status all live in YAML. Bases views render the boards. Files don't move around as state changes.
3. Anti-AI cadence rules for any content you ship. The setup wizard offers to install these. Most LLM-generated content is detectable in two sentences; the rules cap em dashes, mirror inversions, tricolons, colon-reveals, and journalistic clichés.

Minimal everywhere else. Folder structure is flat. Plugin list is short. Templates are templates, not philosophy. Make it yours.

Acknowledgements

This template is the open-source extraction of one person's working second brain — built over weeks of refinement with Claude Code as a thinking partner. Patterns drawn from:

• Andy Matuschak (evergreen notes, atomic granularity)
• Tiago Forte (PARA, capture habits)
• Nick Milo (LYT / ACE structure)
• Andrej Karpathy (the LLM-wiki idea file pattern, active here as the agent-compiled _Wiki/)
• Annie Duke (decision journalling)
• Heath brothers (the SUCCESs framework for content)

License

MIT. Fork freely. Make it yours.