new-deck

# Start a new deck Decks in this repo are graded against a profile, and the profile only knows what to enforce if someone writes down who is listening and for how long. Before this skill existed those premises lived in the presenter's head, leaked into the deck as inconsistencies, and were re-derived from scratch every time anyone touched it. Ask once, write it down, and every downstream tool reads it from the same place. Two halves, deliberately: **you run the interview, `scripts/new_deck.py` writes the files.** Ask the questions however the conversation needs them asked. The scaffold stays byte-identical for the same answers. ## 1. Ask Use `AskUserQuestion` — this is a set of judgment calls, not a form. Batch related ones; do not ask about anything the user already told you. Lead with a recommendation when there is an obvious default. **Genre** — `papers`, `talks`, or `lectures`. Run `python3 scripts/deckpath.py --genres` for the live list. This picks the directory and, unless overridden, the profile. **Audience prior knowledge** — `none`, `practitioner`, or `expert`. The single most consequential answer: it sets which profile's bullet budget and checks apply (and it is what the review agents read the deck against); guessing `practitioner` for a room of first-years produces a deck that scores well and lands badly. The acronym-expansion check is off for every shipped profile (D17: the matcher cannot tell a model name from a term), so spelling terms out is judgment, not...