jkvc/brainstorm

Brainstorm

The user has a goal or direction but wants inspiration. Generate multiple concrete approaches, be opinionated about which are strongest, and let the user cherry-pick.

Steps

1. Understand the goal and the codebase — read what the user described, then thoroughly explore the relevant code before generating any ideas. Understand the data model, existing abstractions, current patterns, and constraints. Ideas that ignore the codebase are worthless.

2. Generate ideas — come up with 4-8 genuinely different approaches, grounded in what the code actually does today. Not minor variations — distinct takes on the problem. For each:

   • Name — short memorable label
   • What — 2-3 sentence description of the approach
   • Why it's interesting — what makes this worth considering
   • Trade-off — what you give up or what's hard about it

3. Rank and recommend — be opinionated. Mark your top 1-2 picks and say why. Flag any that are bad ideas and say why. Don't present everything as equally valid.

4. Invite reaction — ask which ideas resonate, which to kill, and whether to combine any. The user's picks feed into an interview or straight to implementation.

Rules

• Always explore the codebase first. Read the relevant files, understand the architecture, then ideate. Never brainstorm in a vacuum.
• Be creative, not safe. Include at least one ambitious/unexpected idea alongside pragmatic ones.
• Reference specific files, functions, patterns, and constraints from the codebase in every idea. Vague ideas that could apply to any project are not useful.
• Don't ask clarifying questions upfront. Make reasonable assumptions and generate. The user will correct what's off.
• Keep each idea concise. The full brainstorm should be scannable in under 3 minutes.
• After the user reacts, offer to refine the selected ideas or transition to interview/implementation.