What's this about?
s1dashu built and shared a Logo Skill, and the observation behind it is the interesting part. He noticed that Grok Bot, Coze, WorkBuddy, Doubao and Kiro all use their product's IP character directly as the logo — which produces very high recognizability while reinforcing the character in users' memory. So all his products now have a matching cute IP character used as the logo, and after enough of them the pattern became a reusable skill. Its three core rules: Logo first, IP second — make sure it's a clean, clear, identifiable logo before it's a cute character, avoiding complex illustration; minimal composition — keep only the most recognizable outline and features, using as few shapes as possible so it stays legible when shrunk; and rounded, solid lines — avoid sharp corners, thin strokes and jagged structure, which makes it friendlier and better suited to an app icon.
What exactly is the difference between a logo and an illustration?
A logo has to survive being small, monochrome, and glanced at. It appears as a 16-pixel favicon, an embossed shape, a shadow on a dark background. That forces extreme reduction: a few shapes, one strong silhouette. An illustration can be detailed because it's viewed large and looked at. The failure mode this skill targets is real and common with generated art — the model produces something charming at full size that becomes an unreadable smudge at icon size. Encoding "must work small" as a hard rule is what turns a picture into a logo.
Why is this worth noticing?
- It's a skill derived from doing the work repeatedly, not designed in the abstract, which matches the finding that skills encode procedure rather than knowledge.
- The trend observation stands on its own. IP-as-logo is a genuine pattern across recent AI products, and naming it is useful whether or not you use the skill.
- The constraints are all negative and physical — no sharp corners, no thin lines, fewest shapes — which is the same shape as the Hallmark skill's approach of enumerating what not to do.