brainstorm
Installation
SKILL.md
Brainstorm
Decide whether to build and which direction to take. Keep the why before the how: name the problem before choosing a mechanism. Expand the solution space with alternatives grounded in the user's needs and the project. Write no production code or tasks.
- Understand the problem. Read the relevant code, docs, and prior decisions first. Ask only what research cannot answer: who is affected, what fails today, why it matters, and which constraints change the decision. Do not reopen a settled decision without new evidence.
- Compare directions. Weigh the current approach, the smallest useful change, and a materially different option when one exists, each with its expected result, cost, and main risk. Claims about the project cite files read in this run; label assumptions and unverified outside claims. When the user wants generated ideas, follow references/idea-basis-contract.md. When the user cannot judge a territory, run references/blindspot-pass.md.
- Choose. Recommend one option, explain the trade-off that matters, and ask the user to confirm or change it. Resolve material objections; no fixed count of options or rounds.
- Return the verdict: BUILD, DEFER, REDUCE SCOPE, or REJECT, with the reason, scope, exclusions, open decisions, and evidence. DEFER names the event that should reopen the question. Give no numerical confidence without a measured basis.
BUILD or REDUCE SCOPE hands the direction to interview-me, within the user's request. DEFER and REJECT stop here.