Taste protocol layer

Taste needs its own
protocol layer.

Vertical packs cover industry workflows. Eng skills cover how code gets written. Taste protocols cover what makes design good — for your brand, on every AI agent you use.

Free to try · No signup needed for conversion · GitHub OAuth to publish

The AI agent stack, today

Four layers. Only one is still open.

Layer 01

Vertical Pack

Example
anthropics/financial-services
16.7k ⭐ · +3,077 today
Solves

Industry-specific workflows — investment banking, legal, healthcare. Bundle of named agents + MCP connectors per vertical.

Who's building

Anthropic + enterprise partners. Big-co territory.

Status

Industrializing fast

Layer 02

Horizontal Eng

Example
addyosmani/agent-skills
36.9k ⭐ · +2,801 today
Solves

Generic engineering skills — TDD, debugging, ship workflows, code review. Tools an agent reaches for, not domain knowledge.

Who's building

Indie engineers + big-co. MIT-license abundance.

Status

Industrializing fast

Layer 03

Pattern Library

Example
mobbin.com/mcp
621K+ shipped screens · launched May 2026
Solves

Real-world design pattern reference — "how do shipped apps handle KYC / paywall / onboarding?" Industry-wide pattern average.

Who's building

Mobbin + similar reference catalogs. Proprietary MCP behind paywalls.

Status

Just launched

We're here
Layer 04

Taste / Design

Example
designskills.xyz
The protocol layer for taste
Solves

Color, type, spacing, motion, brand voice — encoded as portable skills any AI agent can read. Your taste, on every tool.

Who's building

Open category. No incumbent yet.

Status

Still being named.

The first three layers are filling up fast — workflows, code, and now patterns. The fourth one — the layer that decides whether the output is on-brand, not just average — is mostly empty. That's the one we're building.

Why now

In a world where everyone can ship a product, "what's good" becomes the scarce resource.

01

Taste is what's scarce now.

Vertical packs know how KYC runs. Eng skills know how to write a passing test. Neither of them knows whether a button color is right, or whether a spacing rhythm makes the brand feel cheap. When generation gets free, judgment is the bottleneck — and judgment lives in taste.

02

Taste can't be bundled into a vertical pack.

An investment-bank financial report doesn't look like a DTC brand site doesn't look like a family memory app. Taste belongs to the designer or founder, not the industry. Bundle it into a vertical and you get a thousand products that look the same.

03

Taste can't ride along with eng skills either.

Engineering skills are verifiable — the test passes or it doesn't. Taste is evaluable — it looks right or it doesn't. Two different evaluation mechanisms, two different review surfaces. Mixing them into the same skill registry pollutes both. Taste deserves its own protocol, its own evaluators, its own community.

Already on the protocol

Designers shipping their
taste as portable skills.

View all skills →

Name the layer.
Or someone else will.

Six to twelve months until the big tools roll their own design layer. Until then, the taste protocol is a public good — and the people defining it now will have shaped the category they're using.

Built by designers, used by anyone who cares about taste.

— A protocol layer for taste · Open spec · Free to publish