Turning a company's own writing into an operable brand voice

Warm but not cutesy, premium but not stiff. Everyone nods; nobody can hand it off. The guide describes a vibe and hopes the next writer absorbs it, and that fails the only test that matters: could a new hire, freelancer, or AI tool write on-brand on day one using only the guide?

The operator is a set of markdown files you drop into a Claude project. Claude reads them and becomes the operator: an identity file, a rules file with the staged logic, and a reference folder holding the method. It runs on Interpretable Context Methodology, folder structure doing the work code would otherwise do. No scripts, no install. The files are the program, which is why it runs in a plain browser project.

Two rules hold it up. Evidence over invention:every attribute is backed by two real quoted lines from your writing, or it isn’t claimed. The sharpness bar: anything that would fit any brand in the category gets rewritten or dropped.

Anyone can fork the repo; it’s a dozen markdown files. What doesn’t come with the fork is what makes it work: the calibrated examples that define “sharp,” the never-list method that makes a rule generalize, and the discipline to refuse output that reads like a machine wrote it.

Each run produces one artifact: a north star sentence, the tension the voice lives between, three to five evidence-backed attributes, a numbered never-list, channel playbooks, and an AI prompt pack. Plain markdown, so it feeds any model or a new hire’s first week. Three sample voices it produced sit in the voice gallery.

The proof point is portability: the same folder runs in Claude Code for rigor or a browser project for reach, and ingests a voice doc from another tool as happily as it reads raw writing. One set of files, many runtimes, one contract.

It’s only as sharp as its examples, and right now those examples are mine. Every real voice I run through it yields one validated, anonymized input-to-output pair. A curated set of those is what a competitor can’t fork, and it compounds: more voices, sharper inference, better output.

The scaffold was always going to be easy. The calibration is the work.