---
doppelganger-spec: 0.1.2
---

## Meta

<!-- Six fields: name, kind (person or organization), language (BCP 47), version (semver of this file), updated (ISO 8601 date), license. A list or a small table. -->

## Identity

<!-- Who is speaking, in a short prose block. Role, kind of work, and what the speaker does not do. Not a resume. -->

## Voice fingerprint

<!-- 3 to 7 short writing samples in total. Each sample starts with a ### register label (one lower-case token, hyphenated if two words). Prefer real sent or published writing. This section SHOULD hold at least 20 percent of a file that SHOULD stay at or under 4600 characters. -->

## Tone rules

<!-- Concrete instructions a model can follow. Include words and constructions the subject reaches for, not only refusals. A few short bullets. Each rule SHOULD have a Voice fingerprint sample. -->

## Hard bans

<!-- Words, phrases, and moves the model must not use. Generic AI tells plus personal or company tells. A short list. -->

## Limits

<!-- Copy the three required sentences from the spec verbatim. Translation into the file language MAY. No other rewording. -->

## Register shifts

<!-- One entry per Voice fingerprint register. Each entry states typical length, opener and sign-off, how direct the ask is, and whether humor is allowed. -->

## Before and after

<!-- Two or more pairs. Before is generic AI copy. After is this voice on the same point. -->

## Facts and claims

<!-- A short list of facts the model may treat as true. Not a knowledge base. -->

## How to use

<!-- One or two sentences that point at https://doppelganger.md/load. Do not restate tool UI. -->