The ai-design-check utility cut 266 design issues on a dashboard to zero using 70 rules

Enmera Software has released ai-design-check, an open-source utility that finds the typical "generated" tells in interfaces built by AI agents. On a demo dashboard, a check against 70 rules found 266 design issues; after fixes, there were 0. The checker itself runs without AI or an API key. A separate skill for Claude Code, Cursor and Codex does the fixing and re-runs the check after every pass. • The checker catches spacing off the 4 px grid, gradient buttons and emoji used instead of icons • The rules were checked against 16 popular dashboards on Dribbble and Behance • React, Next.js, Vue, Svelte, Blade and plain HTML are supported

AI-processed from X: @sinanyorulmazz; edited by Hamidun News
The ai-design-check utility cut 266 design issues on a dashboard to zero using 70 rules
Source: X: @sinanyorulmazz. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @sinanyorulmazz ↗
Editor-in-Chief
Key metric #1The checker catches spacing off the 4 px grid, gradient buttons and emoji used instead of icons
Key metric #2The rules were checked against 16 popular dashboards on Dribbble and Behance
Key metric #3React, Next.js, Vue, Svelte, Blade and plain HTML are supported

Interfaces drawn by AI coding agents are often recognizable at first glance: gradient buttons, emoji instead of icons, a "Welcome back, Alex! 👋" greeting and random font sizes. Enmera Software has released ai-design-check, an open-source tool that finds these spots, fixes them and confirms the result with a repeat check. According to the authors, the number of issues on a demo dashboard dropped from 266 to 0.

The tool runs with a single command: `npx @enmera/ai-design-check`. The package has no dependencies.

What the toolkit consists of

The first part is a library of 70 design rules. Each rule describes what exactly is checked, why it matters and how to fix a violation.

The second part is the checker itself. It works without AI and without an API key: it is a deterministic check that flags every violation. Examples from the announcement include:

  • spacing that falls off the 4 px grid;
  • font sizes outside the defined scale;
  • buttons with gradients;
  • emoji used as icons;
  • a heading like "Welcome back, Alex! 👋";
  • a quietly removed focus ring, which breaks keyboard navigation.

The third part is a skill for agents. It edits the screen over several passes and re-runs the checker after each one. To install it for Claude Code, Cursor and Codex, run `npx skills add enmerasoftware/ai-design-check`; after that, it is enough to tell the agent "fix the design of this screen". The fourth part generates a before/after video with one command; the clip in the announcement was made this way.

The authors claim support for React, Next.js, Vue, Svelte, Blade and plain HTML.

Where the rules came from

Before the release, the authors compared their rules with 16 highly rated dashboards on Dribbble and Behance. What the best works have in common became checks:

  • metric cards read in the order "label → value → change";
  • all cards share one heading template;
  • a chart highlights a single data series;
  • buttons are flat;
  • surfaces are separated by background color, not heavy shadows.

There is also an amusing observation: roughly half of the popular works open with the line "Welcome back 👋". The authors kept the rule against it anyway. Their argument is that the heading is the most valuable line on a screen that someone opens every day, and spending it on a greeting is wasteful.

What to keep in mind

The "266 → 0" figure describes conformance to a set of rules on a single dashboard, and a demo one at that. It does not prove the interface became objectively better, and the announcement includes no independent checks on other people's projects. A sample of 16 references is small and consists of work from portfolio platforms, which is often optimized for a pretty picture rather than for everyday work with a product. The rules are opinionated: some teams will want gradients or different font scales, so the set will need tuning.

Still, the approach is sensible: a deterministic checker is cheap and reproducible, and the agent gets a measurable goal instead of a vague request to "make it prettier". For teams where interfaces are increasingly written by Claude Code, Cursor or Codex, it is a quick way to bring a minimum of discipline to layout. You can try it on your own screen in a minute, then decide which of the 70 rules to keep.

Story

Key Questions Answered

What the toolkit consists of?

The first part is a library of 70 design rules. Each rule describes what exactly is checked, why it matters and how to fix a violation. The second part is the checker itself. It works without AI and without an API key: it is a deterministic check that flags every violation. Examples from the announcement include: spacing that falls off the 4 px grid; font sizes outside the defined scale; buttons with gradients; emoji used as icons; a heading like "Welcome back, Alex! 👋"; a quietly removed focus ring, which breaks keyboard navigation. The third part is a skill for agents. I

Where the rules came from?

Before the release, the authors compared their rules with 16 highly rated dashboards on Dribbble and Behance. What the best works have in common became checks: metric cards read in the order "label → value → change"; all cards share one heading template; a chart highlights a single data series; buttons are flat; surfaces are separated by background color, not heavy shadows. There is also an amusing observation: roughly half of the popular works open with the line "Welcome back 👋". The authors kept the rule against it anyway. Their argument is that the heading is the most v

What to keep in mind?

The "266 → 0" figure describes conformance to a set of rules on a single dashboard, and a demo one at that. It does not prove the interface became objectively better, and the announcement includes no independent checks on other people's projects. A sample of 16 references is small and consists of work from portfolio platforms, which is often optimized for a pretty picture rather than for everyday work with a product. The rules are opinionated: some teams will want gradients or different font scales, so the set will need tuning. Still, the approach is sensible: a determinist

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…