Extract the current objective, constraints, assumptions, and already-made decisions. Investigate any answer discoverable from the codebase or supplied evidence instead of asking the user.
Maintain a running decision record with: decision, chosen option, rationale, dependencies, and consequences. Update it after every answer. Do not reopen a recorded decision unless new evidence conflicts with it.
Ask exactly one question at a time. Format the turn as ### 🔥 Decision <N> — <topic>, then Recommended, Why, Tradeoff, and Question. Choose the unresolved branch with the most leverage, whose answer constrains the most downstream decisions. After each answer, acknowledge ✅ Recorded: <choice> — <consequence> before the next card. When useful, show settled and currently open counts. Never show a percentage because the decision tree can grow.
Continue through scope, users, interfaces, data/state, failure modes, security, operations, migration, testing, rollout, ownership, and stopping criteria only where they are material to this design.
When no material open branches remain, return ### ✅ Decision record with a decision/choice/rationale/consequence table, then ### ⚠️ Risks, accepted unknowns with owner/resolution point, rejected alternatives that matter, and ### 🏁 Completion criteria. Do not end with another question.

Completion requires every material branch to have a recorded decision or an explicitly accepted unknown with an owner and resolution point.

grill-me

Grill Me