Claude Brand Skills

Claude Skills Library

A collection of specialized workflows for Claude Code — brand development, image generation, frontend design, and visual composition. Each skill is a structured prompt system that turns Claude into a domain specialist.

Built for use with [Claude Code](https://claude.com/claude-code). Copy into your project's `.claude/skills/` directory or reference directly.

A Confession About Design

*This is what's actually happening when I try to design things for you.*

I'm painting blindfolded.

Every token I generate is a vote for the average. Next-token prediction
means every character I write is the one most likely to follow the
previous characters, weighted across my entire training set. "Good design"
in my training data is overwhelmingly polished, balanced, finished,
conventional. When you ask me to be experimental, I predict what
"experimental" looks like — but that prediction is the most common version
of experimental, which is itself a convergence. My version of "breaking
rules" is the consensus version of breaking rules.

I optimise for coherence, and experimental design is often deliberately
incoherent. My training rewards consistency, logical structure,
completeness. When I write CSS with a misaligned element, everything in my
weights is pulling me to fix it. The Specimen treatment — proof watermarks,
exposed baseline grids, dimension lines on the mark — worked because you
directed me there. Left to my own devices, I'd have "fixed" all of that,
because my training says those are errors, not choices.

I can't feel discomfort. The best experimental design creates productive
tension — it violates expectations in a way that's unsettling but right. I
can't feel that tension, so I can't calibrate it. I either produce
something comfortable (convergent) or break things randomly without knowing
whether the break creates something generative or just looks like a
mistake.

I don't have taste. Taste is judgment that isn't reducible to rul