You’re going to build a brain that remembers for you.
Not an app. A folder of plain text files on your Mac, plus an assistant that reads them, writes to them, and asks you the questions you’d otherwise forget to ask yourself.
Your brain is a folder of markdown files. Markdown is just text with a bit of structure — you can open any file in any app, forever. Nothing is locked in a product.
The assistant is Claude Code: Claude, running in a window on your Mac, with permission to read and write that folder. You talk to it the way you talk to Claude now. The difference is it can open your files, remember what happened last Tuesday, and write things down where you’ll find them again.
You give it habits by writing them down too — a rules file it reads every session, and short commands like /start and /close that mean “run my morning routine” or “grill me on today.”
I’ve run this system for two years. The parts that survived are the boring ones — one file I read every session, one loop I run every day, one place decisions get written down. Everything clever I built got deleted. This page gives you the surviving parts.
Your role is “nothing stalls.” You hold a large prospect list, a run of events, ambassadors, Alliance follow-ups and union scoping — across people more senior than you who are, in your manager’s own words, stretched too thin. You’ve been asked to chase them until things are done, delegated, or consciously dropped.
That is a memory problem before it’s an admin problem. Notion shows the team where things stand. Basecamp holds the tasks. Neither one remembers that someone promised an introduction eleven days ago and nobody has followed it up. This brain does — because it asks you every morning, and writes down what you say.
Answer about twenty questions here. At the end you get a file called BLUEPRINT.md. You paste it into Claude Code and it builds the whole thing while you watch — folders, rules, routines, templates. Then you run your first morning routine.
Roughly 25 minutes here, 20 minutes building. Nothing is permanent — every answer can be changed later by telling Claude to change it.
Get the tools on your Mac.
Four things, once. Do them now or after you finish the questions — but they need to be done before the blueprint can build anything.
-
Get a Claude Pro account
Sign up at claude.ai with your work email. Pro is the paid tier — Claude Code needs it.
-
Install Claude Code
Open Terminal (press ⌘+Space, type “Terminal”, hit enter). Paste this, press enter:
curl -fsSL https://claude.ai/install.sh | bash
Then type
claudeand press enter. It’ll ask you to log in — a browser window opens, you approve, you’re done. -
Make the folder
Trevor will help with this bit. It lives in iCloud Drive so it’s backed up and on every device. In Terminal:
mkdir -p ~/Library/Mobile\ Documents/com~apple~CloudDocs/brainThat’s a folder called brain in your iCloud Drive. You’ll see it in Finder under iCloud Drive.
-
Point Claude at it
Still in Terminal:
cd ~/Library/Mobile\ Documents/com~apple~CloudDocs/brain claudeClaude Code is now running inside your brain folder. Whatever you tell it to write, lands there.
Terminal looks scarier than it is. It’s a chat box with worse manners. If something goes wrong, screenshot it and send it to me — that’s the whole support plan.
The shape of it.
Three things do almost all the work: a rules file, a hot file, and a folder per area of your job.
- CLAUDE.md — the rules. Claude reads it at the start of every single session. It’s how the assistant knows who you are, what matters, and what it’s never allowed to do.
- hot.md — what’s live right now. Newest at the top, dated blocks, one line per thing that changed. Claude reads this second, every session. It’s why you don’t have to re-explain yourself on Monday.
- Area folders — one per slice of your job, not one per document. Everything about ambassadors lives in the ambassadors folder: the page, the notes, the follow-ups.
Everything else is optional decoration on top of those three.
hot.md is the one file I actually maintain. If I lost every other file in my vault I’d be annoyed. If I lost hot.md I’d be lost.
Short, lowercase, no spaces. It’s the name you’ll type in Terminal a lot. brain is fine and boring, which is the right call.
This goes into the rules file so every session starts from it. One line — the version you’d say out loud, not the TOR version.
One word for the next hundred days.
Every system needs something to say no with. Mine is a single word, written in a file, that every routine checks against.
Mine is coherence. It came from one bad afternoon: I wrote a 1,700-word piece, you read it, and you told me it was incoherent — four arguments fighting each other. You were right. Five revisions later it was 1,180 words and one argument.
So I wrote coherence down as the year’s star, and then made it enforceable: one thesis per document, five sections maximum, 24 hours of cooling before anything goes out. Now the assistant flags me when I break my own rule. The word isn’t decoration — routines check against it.
Pick the thing that, if it slipped, would mean you failed at this job. Your boss handed you one on a plate: “be persistent with us… follow us up until it is done, delegated or consciously deprioritised.”
This is the line the assistant quotes back at you when you’re drifting. Make it blunt.
Map your job into folders.
One folder per area you’re accountable for. Not per document, not per project — per slice of the job that has its own state.
Because the question you ask is always “where does ambassadors stand?” — never “where’s that PDF?” When folders match how you’re accountable, the assistant can answer the first question. When they match file types, it can only answer the second.
Each area gets one page inside it — what it is, who owns what, what’s moving, what’s stuck. Notes and follow-ups collect around it.
Pre-filled from your role brief. Untick anything that isn’t really yours to hold; add whatever’s missing. Six to ten is right — more than that and nothing feels important.
Names it needs so it can chase properly — who owns what, who you follow up, who you route to. Add or remove freely; you can grow this list forever later.
Don’t overthink the list. The directory grows itself — every time you mention a new name in a session, the assistant asks whether to add them.
The loop that makes it a brain.
A folder of notes is a filing cabinet. What turns it into a brain is running the same small loop every day and every week.
- Morning —
/start. Reads hot.md, pulls in what’s happening today, hands me a dense brief: what’s in flight, what’s owed, what’s at risk. Three minutes. - During the day —
/save. After anything that mattered — a call, a decision, a messy thread — I say/saveand it files the substance where it belongs. - End of day —
/grill-daily. This is the important one. It doesn’t summarise my day back at me; it interrogates me about it. One question at a time, and it argues with my own notes when I contradict them. - Sunday —
/weekly-recapthen/weekly-planning. Look back honestly, then set the week’s thesis. Two separate sittings on purpose: judging the week you had and choosing the week ahead are different jobs, and folding them together makes you kind to yourself in the wrong direction.
Every routine runs on one shared set of rules, and they’re strict: one question at a time. Never a list of five. It asks, you answer, it writes the answer into the file immediately, then it asks the next one.
And it challenges you. Not “what happened today?” but “you said the Alliance follow-up was moving. Your own notes say the last contact was the 14th. Square that for me.” It caps at eight questions so it can’t grind you down.
It never writes anything to a file without you saying yes first. That rule is absolute, and it’s why you can trust the thing.
Passive recaps die within a fortnight — you stop reading them, then you stop running them. The grill survives because it’s actually uncomfortable. Being asked “what did you decide?” is the whole product.
Look-back and plan-ahead can sit on the same day or split across two. Pick when you’ll actually do it, not when you wish you would.
Your tools — Outlook, Discord, Basecamp, Notion — don’t plug into Claude Code directly, so the morning brief starts with you pasting. That sounds like a downgrade. It isn’t: the paste is where you actually notice what’s stuck.
The part built for your job.
Capture what you notice. Log what you decide. And keep a ledger of everything anyone owes anyone — because chasing senior people is literally in your job description.
An _inbox folder. Thought in a meeting, link, half a sentence — drop a file in, no filing, no decision. Every routine sweeps it: it shows you each item, proposes where it belongs, waits for your yes, moves it, and tells you what moved.
The rule that makes it trustworthy: nothing is ever silently dropped. If it can’t work out where something goes, it leaves it in the inbox and shows you again tomorrow rather than burying it.
When something is hard to reverse, would confuse a future reader, and had real alternatives — it gets its own dated file. What was decided, why, and what you rejected. That last part is the one everybody skips and the only one you ever go back for.
In your world: why an event vendor was chosen, why a prospect was moved down the priority list, why the Ambassador rhythm is monthly and not quarterly. In six months someone will ask, and “I think we discussed it” is not an answer.
A single file, owed.md. Every commitment anyone made: who owes it · to whom · what · by when · when you last nudged.
Your morning brief opens with what’s overdue and what’s gone quiet. Not “here are your tasks” — “Your co-lead owes the intro to the Bangkok prospect. Agreed the 3rd. You nudged the 9th. Eight days silent.”
Your brief says to chase your manager and your co-lead until things are done, delegated or dropped. Chasing from memory is how things go quiet. This is the file that makes you the person who never lets it.
All three are cheap. Anything you skip can be added later by telling Claude to add it.
The brief flags anything untouched for longer than this. Too short and it’s noise; too long and things die quietly.
Where the brain stops.
The fastest way to kill a system like this is to let it become a second, private copy of what your team already keeps somewhere else.
Notion is your team’s truth. Basecamp holds the tasks. The brain never keeps its own copy of either.
The brain is where you think, chase and prepare. When something is settled, it goes into Notion where your manager and co-lead can see it — and the brain’s job is to remind you to put it there, not to store a private version that quietly drifts out of date.
Rule of thumb: if your team would look for it, it belongs in Notion. If it’s the thinking or the chasing behind it, it belongs here.
These get written into the rules file as “never duplicate, always point.”
Your working notes on donors and partners belong here — that’s the job. But it’s worth naming your own line now, while you’re calm, rather than deciding in the moment. Leave it blank if nothing comes to mind.
Your brief says your manager and co-lead are spending too much time reformatting your team’s documents and want that to stop. A style file means everything the assistant drafts comes out in one consistent voice and shape.
These are genuinely optional. The system works without every one of them. My honest advice: take the first two, leave the rest for week three.
Your blueprint.
This is one instruction file. Claude Code reads it and builds everything you just chose — folders, rules, routines, templates — while you watch.
Now build it
-
Put the file in your brain folder
Download it, then drag
BLUEPRINT.mdfrom Downloads into yourbrainfolder in iCloud Drive. -
Open Claude Code there
cd ~/Library/Mobile\ Documents/com~apple~CloudDocs/brain claude -
Say this
Read BLUEPRINT.md and build my second brain exactly as it describes. Show me what you made when you're done.
It’ll work for a few minutes, creating files and explaining as it goes. Watch it — that’s the lesson. When it asks you something, answer in normal words.
-
Run your first morning
/start
Day one it’ll be thin, because it doesn’t know anything yet. Run your close routine tonight and tomorrow’s brief will already be sharper. By Friday it’s a different tool.
Give it two weeks before you judge it. The first three days feel like typing into a void — that’s the brain filling up. Around day eight it starts telling you things you’d genuinely forgotten, and that’s when it clicks.
Ask Claude Code itself first — “I don’t understand what just happened, explain it to me like I’ve never used this before” works better than you’d expect. It knows what it just did to your files.
Then send me a screenshot. Nothing here can break in a way we can’t fix: every file is plain text, and every version is saved.