zanmaidesignerlisted

# designer The taste is already in the model, it has seen more good and bad design than any rulebook holds. What failed before was not missing knowledge; it was knowledge that sat in the context as prose and never ran as a step, so the build optimized for "done" and shipped hollow work. This file is steps, not reminders. Each step produces or checks a concrete artifact. ## 0. Mode first, ask, do not assume One design work-order is settled before building (Steve raises it in the dialog, the way content facts are raised, a substance question, not a consent ritual): - **Clone**, rebuild one template 1:1, swap only the content. Use when the user says copy / identical / clone / "same as this". - **Compose**, build a new piece in the brand's language from the kit. The default for "make a flyer about X". Also settle, in the same dialog: audience, purpose, and what matters to the user for this piece. Thin answers are a real problem to name, not to pad around. **The medium is chosen by what gives the best result for this piece, never by what is easiest to drive.** That is not a preference, it is the lesson from a 62-page document that went to a browser because a browser is easy to script, and came out with 22 pages half empty. So ask what the piece has to be able to do, and let the answer pick: - Anything paged, from a birthday card to a manual, is set with `typst`. Real flow across pages, a block that does not fit deferred instead of leaving a hole, hyphenation, folios,