Output

I want one command to turn a piece of writing into a finished deck, document, or web page, and I want it to look like mine every time, not like whatever the model happened to produce that day. The system keeps three questions separate, because conflating them is what makes output systems collapse: they add, they never multiply.

Mode: how is the output produced? Two engines, never mixed. The deterministic engine renders a locked template: same input, identical output, no AI anywhere in the render path, for when I want a known, repeatable look. The generative engine is the opposite philosophy: it produces three divergent visual concepts, iterates each against written gates, then combines the best features of all three into a merged version for me, with the three originals presented as backups. Each engine is its own skill, because a single instruction cannot honestly promise to never deviate and to explore at the same time.

Type: what shape is the output? Documents (explainers, briefs, reports), slide decks, posters, one-page sites. Each shape is a sub-page inside an engine, not a separate system.

Design system: what does it look like? Every visual identity I use lives in one small token file: a colour palette, a type scale, spacing rules, named once. Both engines read the same files, so declaring a colour once reaches a report, a deck, and a web page at the same time; a new look is a new token file and zero code changes. For slide decks the deterministic engine composes the look from layers: shared chrome (navigation, overview, a notes panel, print-to-PDF), a layout frame, and a skin, which is a design system applied to the deck's shared content blocks by a bridge that reads the same token file. The blocks are written once, so the same deck re-skins without touching its structure.