brainstormlisted

# Brainstorm Two failure modes, opposite and equally common: converging on the first plausible idea, or producing twelve options and leaving the choice to someone else. This separates the two phases so neither happens. The `innovator` agent is the 10x-vision counterpart; this is the structured pass you run inline. Whatever comes out goes to the `devil` before it becomes code (`rules/risk.md`). ## 1. Ground it — 5 minutes, not skippable ```sh .claude/tools/digest.sh ``` Dreaming in a vacuum produces ideas that cannot be built here. Before generating anything, know: the stack, what already exists that could be reused, the actual constraint (time? correctness? a quota? someone's patience?), and what has already been tried and rejected — and why. Restate the problem in one sentence. Half of all brainstorms are answering a mis-stated question, and it is cheapest to catch that now. ## 2. Diverge — no judging, quantity first Generate at least **six** genuinely different options before evaluating any of them. Not six variations of one idea. Judging while generating kills the unusual options first, and the unusual one is the reason you ran this. Force range by attacking the problem from different angles: - **Do nothing.** What happens if this is never built? Sometimes correct, always a useful baseline (`rules/minimalism-ladder.md` rung 0). - **Delete instead.** Can the need disappear by removing something upstream? - **Buy / borrow.** Does the platform, the stdlib, or an e