ai-creators-second-brain
Obsidian vault template + compiled-brain builder: let AI agents load a creator's entire knowledge base in one file read. Born from production practice with Claude Code.
AI Creator's Second Brain 🧠
An Obsidian vault template + "compiled brain" builder for AI creators working with Claude Code (or any AI agent).
Your prompts, your past works, your taste — scattered across thousands of files — become one queryable knowledge system: browsable by you in Obsidian, and loadable by your AI agent in a single file read.
🇯🇵 日本語の説明は下にあります。
The idea
Creative knowledge has different shelf lives. This vault splits it into three layers:
| Layer | Contains | Shelf life |
|---|---|---|
10_principles | How you think — aesthetics, framing rules, creative doctrine | Forever |
20_works | One note per work: the prompt, the result, what worked (○), what failed (×), what surprised you (☆) | Forever (dated specimens) |
30_recipes | Tool-specific know-how (model quirks, camera grammar, prompt formats) | Expires — tracked with status: current/superseded |
Then compile_brain.py compiles the whole vault into one small digest file (~15 KB). Your AI agent reads that single file first — the whole brain in one context load — and only deep-dives into individual notes when needed. The digest ends with a "what this brain does NOT know yet" section, so the agent knows when to go read the source.
Quick start
- Copy
vault-template/into your Obsidian vault (or use it as a new vault) - Fill a few notes: one principle, one work, one recipe (templates included)
- Build the digest:
python3 compile_brain.py --vault /path/to/your/vault - Tell your AI agent (e.g. in
CLAUDE.md): "Before any creative task, read00_MOC/compiled-brain.mdin one go, then deep-dive individual notes as needed. Rebuild with compile_brain.py after updating the vault."
The verification memo (○×△☆)
Binary "worked / failed" doesn't match creative reality. Each work note records per-element outcomes:
○did what I intended×didn't work△partially worked☆unplanned but good → adopted (the most valuable symbol: knowledge nobody designed)
Disaster recovery
A second brain you can lose to a repair shop isn't infrastructure — it's a pet. docs/disaster-recovery.md covers backing up the brain itself (agent memory, conversation logs, skills, vault) with a restore README written for the AI, not for you — field-tested the night before a MacBook repair.
Born from practice
This system is extracted from the real workflow of notfukudaSTUDIO — a former fashion-magazine editor-in-chief turned AI creator (work screened at Expo 2025 Osaka) — who manages 80+ AI works, 50+ distilled recipes and 30 creative principles with it, in production, daily.
🇯🇵 AIクリエイターのための第二の脳
Obsidianの3層Vault(原理/事例/レシピ)テンプレートと、Vault全体を1ファイルに固める「圧縮脳」コンパイラのセットです。散らばったプロンプト・過去作・美意識を、人間はObsidianで、AIエージェントは1回のファイル読み込みで参照できる知識システムにします。
- 知識を賞味期限で3層に分ける(不変の原理 / 日付つきの事例 / 期限つきのレシピ)
- 検証メモは○×△☆の4記号(「狙ってないのに良かった→採用」の☆が最重要)
compile_brain.pyがVault全体を約15KBのダイジェストに圧縮 — エージェントが冒頭1Readで全脳ロード- ダイジェスト末尾に「この脳がまだ知らないこと」を自動生成(エージェントの知ったかぶり防止)
- 災害復旧ガイド(docs/disaster-recovery.md)— 脳ごとバックアップする方法。復元手順書を「AI向けプロンプト」として書くのがキモ。MacBook修理前夜に実戦検証済み
License
MIT
How to Install
- Download the ZIP or clone the repository
- Open the folder as a vault in Obsidian (File → Open Vault)
- Obsidian will prompt you to install required plugins
Stats
Stars
0
Forks
0
License
MIT
Last updated 3mo ago
Categories
Tags