Voices — editorial overlays
On this page
What voices are
Voices are editorial overlays, not content adapters or project profiles. The base scholarly standard of the platform is always active, and voices layer on top of this base to enforce specific editorial, grammatical, stylistic, or visual registers. They do not replace the base standard but rather add domain-specific rigor.
Where voices come from
This layer ships the mechanism — the folio-voice-skill/v1 shape, the
authoring and review skills, the BPMN process and the viewer — and no
voice. A voice is shipped by the instance that owns the register it encodes:
a house style by the instance that holds that organisation’s style manuals, an
exposition standard by the science layer that holds its exemplar paper. Each of
those instances documents its own voices beside their SKILL.md and
voice.json, and links down to this page for the mechanism.
This page does not list them, because every one of them sits above this
layer: a link from here to one would point up the dependency arrow, and a list
here would be a second catalogue free to drift from the first. The one
catalogue is the interactive viewer (bun run cat voices:viz, published at
cat-harness/docs/cat-harness/voices/), which re-reads every instance that
declares a voices graph on each run; skill_list answers the same question
from an agent.
Activation
To activate voices, list them in your folio.config.json under voices.active[].
{
"voices": {
"active": [
"who-editorial",
"who-guideline-development"
]
}
}
Platform docs themselves carry no voice, meaning their voices.active[] array is empty or absent.
The authoring cycle
Voices are designed to be part of the entire document lifecycle:
- Before writing: Follow the voice-authoring-guidance skill. It is much cheaper to honor terminology and stylistic rules as you type than to retrofit them later.
- After writing: Use the voice-overlay-review skill. This reviews your blocks against the rules of your active voices and places findings on the QA sidecar.
How voice review works
You can read about the formal review process in the Voice overlay review BPMN process page.
Every rule inside a voice carries a source citation that resolves to a real document or reference. When reviewing, you must open the cited page. The mechanical half of a rule (its patterns or terminology) acts as a question. The reviewer must check the context to decide if the rule truly applies or if it’s an exception (the judgement half). If a rule’s own citation does not support it, the defect is in the rule itself, not the prose.
You can explore all the active rules and their citations in the interactive viewer at cat-harness/docs/cat-harness/voices/.