023 / 024 · Brand · v0.17.1 · MIT
logo-banner
logo-banner does not start by generating an image. It starts by settling the direction.
- When it fires
The prompts CI scores this skill against, on every release.
Fires on
- Design a logo and GitHub banner for my new personal knowledge app.
- The light Hero exists. Please make a matching dark-mode brand banner and put it in the README.
- I need an app icon and a favicon generated from the mark we settled on.
Stays out of
- Make the cards on this landing page line up better on mobile.
- Pick the accent colour for our design-system tokens.
The direction is agreed before anything is drawn
It reads the project first — the name, what it is for, who it is for, the visual language already in use, the constraints it has to respect — and then offers exactly three concrete directions, each stating a mood, a palette, a motif and a type character. You pick or amend one, and only then does generation begin.
Three is deliberate. One is no choice; five makes you compare near-identical options, and the usual outcome of that is starting over.
One direction produces the whole set
The logo, the app icon, the favicon, the header image and the light and dark social banners all derive from that one agreed direction. That is why they look like a set rather than five separate drawings — assets generated independently drift apart in weight, spacing and motif even when the style description is the same.
Without a generator, it stops
When no image generator is available it says so and stops rather than substituting another tool. An improvised logo gets used as though it were real, and the gap only surfaces later, in the one place that needs a vector source.
What it does not do
It does not vectorise, does not write brand guidelines, does not do print layout, and does not generate anything before the direction is agreed.