This repository documents a writer's voice in enough detail that a machine can write in it without erasing the person who built it.
The artefact at the centre is a voice profile. It records how I write, and underneath that, why I write at all. It was compiled on 25 May 2026 from a long excavation interview: ninety-five questions that began with craft and kept going until they reached the motivations the craft was serving. Around that profile sits a skill, a compressed and operational version of the same rules, structured so it can be loaded into an AI writing session and executed against. The profile is the reasoning. The skill is the instrument.
The premise is narrow, and worth stating plainly. AI-assisted writing earns its credibility only when the voice it executes already exists, in full, as the writer's own work. There is a real line between assisted and generated, and it does not run where most people draw it. It is not about how much of the typing the machine did. It is about whether the thinking, the stakes, and the register were the writer's before the machine ever touched the page.
That is what makes this document necessary rather than indulgent. The labour of building a voice is what earns the right to delegate its execution. Strip the labour out and AI-assisted writing is just AI-generated writing with better manners. So this repository is the labour, made visible and kept. The next time a tool generates something in my voice, it does so because the voice was built first, by hand, across ninety-five questions, and written down.
Start with the voice profile, in voice-profile/. It is the primary document and the source of record: a manifesto of intent and an operational ruleset in one file, dated deliberately so that a later version of me can measure distance from it. Read it the way you would read an essay, not a manual. The rules only make sense once the reasoning underneath them is visible.
The skill/ folder is for operating, not for understanding. SKILL.md is the entry point. reference-card.md is the two-minute recalibration before a writing session. risk-matrix.md decides how much to trust a given draft based on where the thinking came from. voice-rules-full.md is the complete working manual for when the entry point is not detail enough. The profile and the skill describe the same voice; the difference between them is depth, not direction.
/
├── README.md
├── LICENSE
├── CHANGELOG.md
├── voice-profile/
│ └── voice-profile-azryl-ali-v1.md
└── skill/
├── SKILL.md
├── voice-rules-full.md
├── risk-matrix.md
└── reference-card.md
The voice is mine. The document was compiled by Claude, in collaboration, across the interview that produced it. That division is honest and worth keeping straight: the thinking and the stakes are mine, the structuring and the pressure-testing were shared, and the result is a record of my voice rather than a model's idea of it. The same distinction the document defends is the distinction by which it was made.
The licence is CC BY-NC-SA 4.0. You may share and adapt the work with attribution, for non-commercial purposes, and any derivative carries the same terms. It is the right licence for a personal-voice artefact that I want visible and citable, but not commercially appropriated. The full text is in LICENSE.
The construction of this profile has its own record, an annotated transcript and the load-bearing moments inside it, and that record is held in a separate private archive. It is not lost, and it is not here. Whether any of it is ever made public is a decision on its own, and it has not been made.
This is part of a body of work, not a standalone tool. It will be cited in later writing, revised as the voice moves, and kept whole as it changes. The original v1 stays fixed; the corpus around it continues.