brainstorming

Brainstorming

Contract

Iron law: explore before committing — choosing wrong costs ten times what exploring costs. Invoked by /flow:brainstorm Phase 2 (GENERATE), Phase 3 (COMPARE), and Phase 4 (DECIDE), and whenever several valid approaches exist, requirements are ambiguous, or the team defaults to its first idea. Returns 2–4 genuinely distinct, implementable approaches (name, summary, pros, cons, effort, risk), a comparison table across the dimensions that matter for this decision, a recommendation with its key trade-off, and the user's choice via AskUserQuestion, logged to the decision journal. Permitted skips: none — the recommendation is never the decision; the user chooses.

Process

Track five tasks — clarify goal, check existing patterns, generate approaches, compare and recommend, user decision — and confirm with TaskList at the end.

1. Clarify the goal

Confirm what must happen (outcome, not implementation), why (motivation, constraints), and who is affected. Ambiguous → AskUserQuestion; do not guess.

2. Check existing patterns

grep -r "relevant_pattern" --include="*.{ts,js,py,rb}" -l
git log --oneline --all --grep="related keyword" | head -10

Existing patterns get priority; consistency has value.

3. Generate 2–4 approaches

Per approach: Name, Summary (one line), Pros, Cons, Effort (Small/Medium/Large), Risk (Low/Medium/High).

Rules:

• Minimum 2, maximum 4. More than 4 is analysis paralysis — cap and decide with available information.
• Genuinely distinct, not minor variations; "we can only do A or B" usually hides a C.
• Include the "do nothing" or "simplest possible" option when applicable.
• Each must be implementable — no hand-waving.
• Present all options before discussing any, so the first does not anchor.

4. Compare

Dimension	Approach A	Approach B	Approach C
Simplicity
Flexibility
Performance
Consistency
Effort
Risk