# Tone Book

> A brand tone of voice guide (TONE.md) that writers and bots can actually follow: the fixed voice, then the tone per
> channel. Answer the discovery questions and paste the brand's own past copy; the browser checks the intake against
> the skill's minimum batches, flags a category that needs research first, measures what the samples already show,
> lists every channel's hard limits and checks any pasted copy against the guide. Metered runs define the voice -
> three to five "X, but never Y" attributes with do's, don'ts and anti-examples taken word for word from the brand's
> own writing, the archetype, the NN/g four dimensions, a lexicon led by a banned list, mechanics, inclusive language,
> global do's and don'ts and before/after examples - and port it to up to four channels at a time with a tone
> modulation matrix. Built on the SkillSafe skill @samber/copywriting-tone-of-voice-creator, from samber/cc-skills
> (MIT), by Samuel Berthe. It never invents a fact about the brand and never cites a source.

https://tone-book.skillsafe.ai/

## What is free
- The minimum discovery answers (brand, category, markets, goal, audience, channels, reading age, risk tolerance, archetype guess, references and anti-references, regulatory regime, localisation), each marked given or missing.
- A research-first flag for the categories the skill researches before defining a voice: political, religious, defense, gaming, healthcare professional comms, sports, fintech and crypto, regulated substances, insurance and legal, other.
- What the brand's own samples show: average and longest sentence, negative and positive contractions, we versus you, exclamation marks, emoji, dashes, ellipses, the Oxford comma, passive voice, Title Case headings, banned terms and machine-writing tells.
- The NN/g lean check: at least three of the four dimensions off neutral.
- The hard constraints of every channel in scope (approximate - platforms change them).
- A copy checker: banned terms (yours, the skill's starter list and the guide's), machine-writing tells, sentence length for the reading age, exclamation marks, emoji and the channel's limits.
- Exports: TONE.md with everything written so far, the channel limits (.csv), everything the page worked out (.json).

## The metered lanes
Two tasks on gpt-terra, metered per run, routed by `task`. Input string fields: `task` (voice or channel), `brief`, `samples` (voice only), `guide` (channel only) and `facts`.
- `voice` replies with `summary`, `attributes` (3-5, each name, definition, sounds_like, not_like, three do, three dont, anti_example), `archetype`, `nng` (four dimensions in order), `lexicon` (preferred, banned, power, jargon, naming), `mechanics` (eleven decisions in order), `inclusive`, `dos`, `donts`, `examples`, `research` and `caveats`.
- `channel` replies with `summary`, `channels` (per channel: constraints copied from the page, shift, three do, three dont, opening, and a matrix column of ten situations with a dominant tone and two or three prohibited tones) and `caveats`.
The page holds every reply to the skill's validation: three to five attributes, anti-examples found verbatim in the samples, at least three NN/g dimensions off neutral, a banned list that carries every term you gave, one section per channel and every situation in the matrix.

## Links
- API tutorial: https://tone-book.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@samber/copywriting-tone-of-voice-creator