brainstorm

Files (1)

• SKILL.md

SKILL.md (first lines)

# brainstorm — diverge, cluster, converge

Generates real optionality before committing. The failure mode this guards against is grabbing the
first plausible idea; the fix is separating generation from judgement.

## Step 1 — frame
State the question sharply and the constraints that any answer must satisfy (must-haves) vs
preferences (nice-to-haves). A good frame is specific enough to judge ideas against.

## Step 2 — diverge (no judging)
Generate many ideas — aim wide, include the obvious, the ambitious, and at least two deliberately
weird ones. **No evaluation yet.** Pull from different angles: user need, constraint inversion,
analogous domains, "what would make this trivial", "what if we had no budget / infinite budget". Use
web search for prior art where it helps. Quantity now, quality later.

## Step 3 — cluster
Group the ideas into themes; name each theme. Collapse duplicates. Surface the underlying dimensions
the ideas vary along (e.g. depth vs breadth, guided vs exploratory) — often the dimensions matter more
than any single idea.

## Step 4 — converge
Score the strongest candidates against the Step 1 criteria (impact × feasibility × fit). Show the
scoring. Recommend one primary direction and one hedge, and name what you'd need to learn to be sure.
Hand the winner to `/design-architecture` (to design it) or `/write-story` (to slice it). If the
outcome needs the operator's call, append a **DECISION** line to the INBOX.

## Do not
- Collapse to one idea in Step 2 — diverge first, always.
- Judge during divergence — it kills the weird ideas that are sometimes the good ones.
- Stop at a list — the value is the cluster + the recommendation, not raw quantity.

From PlatformRelay/Kaddy, MIT.

More from PlatformRelay/Kaddy