The setup, shown Read
Read
my files.
The structure is open. Seven markdown files, a folder system, a few rules about who writes what. You could copy it this afternoon. So here it is, the setup that runs WiZRD. Same structure a client gets, filled with me instead of them.
The files are easy.
Filling them isn't.
Anyone can copy a folder of empty markdown. What took years is the judgment that fills it, one approved decision at a time, and the discipline that keeps it true after month one.
I've written the argument elsewhere: why setups rot in a month, and why the part that needs you is the part you can't automate. This page is the other half. Not the argument. The artifact.
Three of the seven.
Each file owns one job. Here's what's actually in mine, lightly redacted, with what each one changes about how the AI works.
ME.md
Cold · changes yearly
I'm a builder, not a consultant. I prototype, design, write copy, structure pitches, and architect AI systems. I pressure-test before I commit, but lightly. A poke or two, then once the reasoning holds I decide fast and move. I don't revisit settled decisions without new information.
// what it changes
Six lines and the AI stops handing me three options and a recommendation. "Builder, not consultant" changes what it drafts. "Decide fast and move" changes how it argues with me. It barely shifts in a year, so it's a cold file: the AI proposes edits, I approve them, nothing here gets rewritten behind my back.
STYLE.md
Cold · the voice
No emdashes. Ever. Restructure the sentence instead.
Wrong: The deck is the artifact — what matters is the
conversation it starts.
Right: The deck is the artifact. What matters is the
conversation it starts. // what it changes
Voice survives as rules with examples. The file is mostly before-and-after pairs like this one, so the AI writes in my cadence, not a generic founder's. This is the file that takes longest to get right and matters most.
DECISIONS.md
Log · append only
2026-05 · Write permissions follow file temperature Cold files: Claude drafts, I approve. Hot files: Claude writes directly. The log: append only, never rewrite. One rule for every file fails in opposite directions.
// what it changes
The log only grows. New decisions go in dated, underneath. I never rewrite an old call, I let it stand as history and add the new one below. That's how the system remembers why something was chosen, not just what.
This is the part
that needs a person.
Cold files don't update themselves. The AI drafts a change, I keep it, edit it, or drop it. Here's one that got dropped.
Claude proposed ↓
ME.md: "I help teams leverage AI to accelerate concept development."
I dropped it ↓
"Leverage" and "accelerate" are the tell. Not how I talk. If that lands in a cold file, everything it writes afterward sounds a little off and I won't catch it for weeks.
Kept instead →
ME.md: "I help teams turn vague ideas into things they can build."
// Write discipline ·
The system keeps what passed. Patterns that repeat get promoted into the rules. Nothing lands without me. That's what keeps the context mine, and what lets me trust it to act on the work.
This is the thing
running the practice.
The one you just read about. Want one? Thirty minutes, no deck.
This page was written through the system it describes.