The paste.
Every new chat starts the same way. Who you are. What the business does. How you like things written. You have it saved in a note somewhere, and you paste it in again.
Stop re-explaining who you are, how you write and what you're building. Give your assistant a set of folders it reads before you say a word.
Free download. No email, no signup. 10 files, about 12 KB.
You're in Instagram's browser, and it can't download files.
Two ways around it. Tap the … menu and choose Open in browser, or leave your email below and I'll send you the link.
Email me the link insteadThree moments that happen to everyone who uses AI seriously for more than a week.
Every new chat starts the same way. Who you are. What the business does. How you like things written. You have it saved in a note somewhere, and you paste it in again.
It writes something technically correct and completely unlike you. So you spend twenty minutes editing the voice back in, which is the part you wanted help with.
One chat knows your client work. Another knows your uni deadlines. A third has that good prompt in it. None of them know about each other, and you can't find any of them.
The model isn't the problem. It has never been given your context in a way it can find.
A CLAUDE.md file loads automatically when you open the folder it sits in. That one behaviour is the whole system. The folder you open decides what your assistant knows.
A small file at the top that points at everything else. It does no work itself.
Who you are, how you write, what you're building. It loads every single session, everywhere.
One folder per area of life. Business, study, finances, personal. Only the relevant one loads.
AI Ops/ ├── CLAUDE.md the root loader │ ├── Claude/ the global layer │ ├── CLAUDE.md loads every session │ ├── context/ what is true │ │ ├── about_me.md │ │ ├── writing_style.md │ │ └── memory.md │ └── skills/ how work gets done │ └── weekly_summary.md │ ├── Business/ domain │ ├── CLAUDE.md overrides the global layer │ ├── context/ clients, offers, pricing │ └── skills/ proposals, outreach, reports │ ├── University/ domain │ ├── CLAUDE.md │ ├── context/ units, due dates, marking rubrics │ └── skills/ essay drafting, revision plans │ └── Personal/ domain ├── CLAUDE.md ├── context/ health, money, admin └── skills/ weekly review, budgeting
The rule that holds it together: the domain wins.If your global file says write casually and your study domain says write formally, the study domain is right for study work. Narrower context beats broader context, every time.
I'm Will. I'm at uni in Adelaide, and I'm building a two-person marketing agency with my best mate Calem, from nothing.
I built this because I was the person with 40 tabs open. Now the whole agency runs on it: client work, uni, money, the lot.
I'm posting the build in public, the ugly parts included.
Follow along at @watchwillbuild
The real structure, empty and ready for your own material. Unzip it, open it in Claude Code, start filling it in.
The structure takes ten minutes. Knowing what to write in it is the hard part, and the README teaches you exactly that.
Direct download. No email required.
I keep writing skills and domain packs. Leave an email and I'll send them as they land, plus a note when The Builder is ready. Nothing else.
This one belongs on your computer, not your phone. Leave an email and I'll send you the link so you can grab it there. You'll also get new skills as I write them, and one click gets you out.
No spam, and one click to leave. The starter above is yours either way.No spam, one click to leave. You can still download it here if you want it now.
A plugin that builds the whole system with you, end to end. Roughly an hour from install to a system that is populated and working, not empty.
One question at a time, in plain language. Then it writes your context files from your answers, in your words. No blank page.
Business, study, finances, personal. Each one arrives with its own structure and starting context already shaped for that area.
Twelve working procedures covering client work, study, money and the weekly reset. Written out and ready to run, not described.
How the system remembers you and keeps remembering: what gets written down, what never does, and why your work rules stay out of your personal ones.
The rules that stop the system drifting: what never goes in a file, when a domain overrides the global layer, how memory gets updated.
A plain list of the tools and MCPs I run day to day. No setup guides, no config to copy. Just what's worth knowing exists.
Everything above, as a thing you install and run yourself. I'm still finishing it.
Includes 12 months of updates. What you own keeps working either way.
Tell me when it's readyNinety minutes on a call, screen shared. I interview you and write your context files as you talk, set up your domains, drop in the skills that match how you actually work, then we run one real task live before you hang up.
Goes straight to my inbox. I'll reply with times and what to do beforehand.
A Claude subscription with Claude Code. That's it. It works on Mac and Windows. Everything in the system is plain markdown files in ordinary folders, so nothing is locked to my setup or anyone else's.
No. It's the real structure, just empty. Same three layers, same routing, same rules. The difference is that you fill it in yourself with the README next to you, instead of being interviewed through it.
If you can make a folder and edit a text file, you can do this. The README walks you through it step by step. If you'd still rather someone set it up with you on a call, that's coming later.
Anyone running more than one thing at once and using AI for all of it. Freelancers, small operators, students, founders. If you only ever ask it one kind of question, you don't need a system for it.
If we finish the ninety minutes and you haven't got a system that works, I refund it. No forms, no argument. You'd still have everything we built, because it's your files on your machine.
Ten files. About twelve kilobytes. The whole shape of the thing.