# START-HERE — Second Brain OS This is the document that makes everything else in this vault work. Read it once before you start setting things up. It takes about 10 minutes. --- ## The problem this solves If you use an AI assistant for serious work — writing, planning, coding, research — you've hit this wall: every new session, you spend 10-15 minutes re-establishing context. What project are we on? What did we decide last week? What's the current state of this thing? It's not a prompting problem. It's a storage problem. Most Obsidian vaults are designed for you to browse. WikiLinks, backlinks, graph view — great for humans following trails of thought. But an AI assistant doesn't browse; it reads. It takes in what you give it and needs that material to be predictable, findable, and structured. This vault is designed for how an AI reads, not how a human browses. --- ## How to tell your AI about this vault Copy this briefing prompt at the start of any session where you want your AI to have context: ``` My knowledge base is in Obsidian. The vault structure: - 00-Inbox/ — recent unsorted captures. Don't read unless I ask. - 00-Memory/ — per-fact store. Each file = one fact, preference, or decision. When you need context about me or my work, check here first (grep by keyword). - 01-Projects/[name]/ — each project has exactly: status.md, tasks.md, decisions.md - 03-Decisions/ — major decisions with the reasoning behind them (YYYY-MM-DD-topic.md) - 04-Handoffs/ — session-ending summaries (read the most recent before we begin) - 05-Patterns/ — reusable formats and recurring procedures - 06-Lessons/ — things I've learned that changed how I work - 07-Reviews/ — retrospectives, audits, weekly reviews Routing rule: when in doubt, a new note lands in 00-Inbox/ first. When we finish a substantive session: write a handoff note to 04-Handoffs/YYYY-MM-DD-[topic].md summarising what we did, what we decided, and what the next session should start with. ``` Adapt the paths and wording to your setup. This exact text works with Claude, ChatGPT, and Gemini. --- ## The routing decision tree Every piece of information has exactly one home: | What it is | Where it goes | |---|---| | New capture, unsorted thought, idea | `00-Inbox/` first | | A fact about you (preference, constraint, style, habit) | `00-Memory/` — one fact, one file | | A decision you made and might reference again | `00-Memory/` or `03-Decisions/` (see below) | | Current state of an active project | `01-Projects/[name]/status.md` | | What's left to do on a project | `01-Projects/[name]/tasks.md` | | A non-trivial project decision + reasoning | `01-Projects/[name]/decisions.md` | | A major life or work decision that spans projects | `03-Decisions/YYYY-MM-DD-topic.md` | | Summary of a finished work session | `04-Handoffs/YYYY-MM-DD-topic.md` | | A format or procedure you'll reuse | `05-Patterns/` | | A lesson that changed how you work | `06-Lessons/` | | A retrospective or audit | `07-Reviews/` | **00-Memory/ vs 03-Decisions/:** Both hold decisions. The difference: 00-Memory/ is for facts the AI should know about *you* (preferences, working style, technical choices, standing policies). 03-Decisions/ is for significant cross-project calls worth auditing later. When in doubt, use 00-Memory/. --- ## The 00-Memory/ schema (the most important folder) The per-fact store is what separates this vault from every other system. Each file: - **Filename** = the fact. Use kebab-case. `my-code-review-preferences.md`, `current-tech-stack.md`, `client-billing-day.md`. - **Contents** = the fact + a brief explanation + when it applies. - **Size** = short. 50-200 words is the target. If it needs more, break it into two facts. - **Frontmatter** = `type:` (preference / reference / decision / constraint) and optional `tags:`. Tell your AI: *"My memory store is in 00-Memory/. When you need context about my preferences or past decisions, grep by keyword."* With a capable model that can read files, this is enough. The AI finds the relevant fact files, reads them, and has context — without you re-explaining everything. **What to put in 00-Memory/ first (suggested starting facts):** - Your working hours / availability - Your communication preferences (direct/formal/casual) - Tools you use regularly and how (which stack, which services) - Your code style or writing voice - Any recurring constraints ("I don't have admin access to X", "budget cap is Y") - Decisions that apply across multiple projects --- ## The 01-Projects/ discipline Each active project gets exactly three files: - **status.md** — current state. One page maximum. What phase, what's done, what's next, what's blocked. The goal: a new session can pick up in under 60 seconds. - **tasks.md** — three sections: Active / Blocked / Done. That's it. Not a Jira board. - **decisions.md** — non-trivial calls with the reasoning behind them. Not a log of every choice — just the ones you might question later. The discipline is: enough for a complete handoff in 500 words. No sprawling wikis inside project folders. --- ## The 04-Handoffs/ habit (the single biggest time-saver) At the end of every substantive work session with your AI, ask it: *"Write a handoff note to 04-Handoffs/[today's date]-[topic].md. Summarise what we did, what we decided, and what the next session should start with."* The next session begins: *"Before we start, read the most recent handoff note in 04-Handoffs/."* This eliminates the 10-15 minutes of re-explanation that kills momentum at the start of every AI session. It also means the AI never starts from zero — it already knows where you left off. --- ## First-week setup (in order) 1. **Set up 00-Memory/ first.** Write 5-10 fact files about yourself — preferences, stack, constraints, working style. These pay off in every session going forward. 2. **Create a node for one active project.** Copy the `01-Projects/_template/` folder, rename it, fill in status.md with a current-state paragraph. 3. **Use the inbox for one week.** Don't classify anything immediately. Let captures land in 00-Inbox/, then triage on Friday. You'll quickly see what type of content you actually capture. 4. **Write your first handoff note.** At the end of your next AI work session, ask it to write one. Then start the following session by reading it. Once you experience this once, you'll never skip it. 5. **Let 06-Lessons/ and 07-Reviews/ develop naturally.** Don't force them. Add a lesson when something genuinely changes how you work; do a weekly review when you actually have one. --- ## Adapting this vault to your workflow This structure is intentionally minimal. You should adapt it, not adopt it wholesale. Things to change without guilt: - Add more top-level folders if you have a clear use case (e.g., `02-Areas/` for ongoing responsibilities) - Add subfolders inside 01-Projects/ if you have complex projects - Rename folders to match your mental model - Add more template files to 05-Patterns/ Things to keep: - 00-Inbox/ as the landing zone (exceptions here are where things get lost) - 00-Memory/ as a separate per-fact store (not embedded in project folders) - The handoff habit (this is the biggest ROI item) --- *This vault template was designed for anyone who uses Obsidian alongside an AI assistant for substantive work. The automation scripts, hooks, and scheduled jobs that run on top of the original vault are not included — they're machine-specific. This is the architecture. The architecture is the harder part.*