Agent skill

Design Intelligence Skill (RYUX): Ask Before You Design

The RYUX design intelligence skill makes an AI agent ask about the product before designing, leave unknown facts out, design for failure states and explain each decision.

Install
npx @ryuxdsgn/design-intelligence
View on GitHub

Stars, license and last push from GitHub. Install count as reported by the open skills directory.

What is the design intelligence skill?

RYUX, in the design-intelligence repo, is a design skill that makes an AI agent ask about the product first, keep unknown prices and policies off the screen, design loading, error and offline states, and explain its choices. A scan command also checks code and rendered pages without an LLM.

RYUX is a design skill for AI agents published in the design-intelligence repository. Its pitch is to understand before designing: the agent asks about the product, keeps what it does not know out of the screens, designs for when things go wrong and explains each decision.

It ships as one skill folder, ryux, with a router SKILL.md, five capability files and knowledge modules. One workflow sits behind every step: understand, frame, explore alternatives, evaluate, build and critique, then verify.

What the README reports from its benchmarks

  • In an order-tracking brief, agents without the skill showed 28 invented business rules as fact across three runs, against 1 with RYUX, counted blind.
  • Offline, loading and error states appeared in every RYUX run and in none of the no-skill runs.
  • A senior designer rated one checkout blind at 4.33 with RYUX against 2.83 without.
  • One illustration benchmark failed at first, then passed a rerun after a fix.
  • A RYUX run costs about 1.5 to 7 times a run without a skill.

Who the skill is for

Best for

  • Prototypes where invented fees, promos and policies would mislead stakeholders.
  • Teams that want critique with severity and evidence, like critique, plus a QA pass at several widths.
  • Developers who want a no-LLM scan in CI for alt text, contrast and sideways scroll.

Not for

  • Quick, cheap drafts. It asks first and costs more per run.
  • Full reference-data features. The README says reference screens from real products are private, optional and in early access.
  • Guaranteed rating gains; the benchmarks are two to three runs per side and describe those runs, not rates.

Use it with UiChemy

Run the Analyze or Critique step on the WordPress page you ship, and use the scan on a rendered URL. Build the page itself with UiChemy Composer or AI to WordPress. Click Get Started on uichemy.com.