design-ideate

What it does

Not variations but concepts: three models draw the same screen, each with its own MOVE — what the screen fundamentally is. The core rule comes from a run that failed: diverge along only ONE of three axes (taste, structure, move) and hold the other two fixed. Freedom on all three at once reads as “off-brief”, not “fresh” — the concepts drift out of the task and there is nothing to choose between. The working mode is divergence by move with taste and structure locked: options stay inside the task while making a move you hadn't considered. Before showing, the skill measures that concepts really differ (font-weight profiles, chromatics, radii): if two converge that's model convergence, not coincidence, and the weakest is regenerated.

How to use

1. 1Say “show me other options for this screen” or /ideate.
2. 2The first question is about grounding: do you have references you like and an agreed structure. You almost always do — without them the skill guesses taste from scratch and misses.
3. 3Taste is extracted from the references, the skeleton from the structure; it asks separately what must NOT be touched — “freshness” in a settled area reads as damage.
4. 4If the screen lives on content, real content goes in, not placeholders: a grey box instead of a photo kills the idea before anyone sees it.
5. 5The accent colour is computed from the key image rather than eyeballed — a colour taken from the shot reads as continuation, one assigned from outside as a sticker.
6. 6The set of moves is shown to you in one line BEFORE generation — cheaper than regenerating.
7. 7Concepts arrive on one canvas as separate artboards: compare side by side, name the winner and what's wrong with the rest — the choice feeds the tally.

When it triggers

Example

/ideate the hero of a personal page — three concepts side by side, diverging by move, taste and structure kept yours.