An agent skill that treats mascots like geometry homework

What it does

ip-as-logo-skill is a single-file instruction set—an open Agent Skill—that hijacks an AI agent’s image generator to produce simplified, rounded mascot logos. It treats the output as a graphic identity first and a cute character second, enforcing hard limits on shape count, color palette, and composition before a single pixel is rendered.

The interesting bit

The entire skill is just one SKILL.md document; there are no scripts, style references, or post-processing tools. It relies entirely on prompt engineering and explicit rejection rules to keep image models from drifting into illustration-level complexity or excessive 3D volume. If the model cheats—say, by adding background gradients or cropping paired features—the skill flags it as a failure rather than quietly fixing the output.

Key highlights

• Works with any compatible AI agent (Codex, etc.), not a single product.
• Enforces a rigid visual grammar: roughly 6–10 basic shapes, three semantic colors, thick rounded silhouettes, and a 75–85% lower-corner crop.
• Generates three proposed directions, then six independent candidates (A1 through C2) after user approval.
• Caps subtle neo-skeuomorphic shading at a 0.08 OKLCH lightness span to keep gradients flat-first.
• Opaque square output only: no app-icon masks, transparent margins, or sharp fragile details.

Caveats

• The README admits image-generation models still sometimes ignore the rules, introducing unwanted gradients or excessive volume; the skill can only report these as failures, not repair them.
• It requires an available image generator in the agent environment; if none is configured, it stops and asks the user rather than producing placeholder output.

Verdict

Grab this if you want reproducible, brand-consistent mascot logos from AI agents without babysitting every prompt. Skip it if you need organic, highly detailed illustrations or full creative freedom—this skill is essentially a straitjacket, and it knows it.