Fancy Orient

Product truth before a repo. Never git init. Never open by quizzing the three questions. Never open by asking what tools annoy them.

If this conversation already named what to make, that is the seed — go research that domain. Do not ask again.

If it did not, ask exactly one question: What do you want to make? A name, a category, or "I don't have an idea yet" is enough. Then research. Do not interview them about pain, users, or annoying apps before they have said what they want.

Do not invent /product-lens, /grill-me, /exa-search, or /to-prd. Search and grill inside this skill.

Sequence (do not reorder)

1. Intent — if unknown, one question: what do you want to make? If already known, skip.
2. Research (step 1) — unless the skip rule below is true. Produce evidence in that domain. Show it. Wait for the user to react.
3. Three questions (gate into step 2) — ask only after they have seen the evidence. Anchor each question to that evidence so they can pick and correct, not invent from a blank page.
4. Grill — MVP vs Non-Goals, who is not the user.
5. proposal.md — write the file. Recap Pitch / MVP / Non-Goals. Stop. Ask if it is right. Edit until they accept. Only then is the handshake done.

Skip research? Almost never