Skip to main content
Audit what’s actually on screen, not what a DESIGN.md says. The QA counterpart to the design pillar. It opens a rendered page in a real browser, reads its actual computed styles (the fonts, heading scale, and colors the browser resolved), and grades them deterministically — every finding is a pure function of the page, not an LLM opinion. With a report, it also grades the page against that report’s design system.
/design-review https://your-site.com

What it checks

Deterministic hard-rules over the rendered styles:
  • Fonts — more than three distinct families in use; a body face that’s one of the AI-default convergence-trap fonts (Inter / Roboto / system-ui / …).
  • Headings — a heading scale that isn’t monotonically decreasing (h1 ≥ h2 ≥ h3).
  • System match (when a report’s design system is supplied) — whether the rendered body font actually matches the brand’s typography choice.

What you get back

FieldWhat it is
score / passed0–10 (10 minus penalties); passed is true with no fail-severity findings.
fontsThe distinct font families actually rendered.
headingsThe rendered h1/h2/h3 sizes.
findings[]Each a StyleFinding: severity (fail / warn / ok), category, and a one-line detail.
against_systemWhether it was graded against a design system, not just the hard-rules.

Needs the browser

Design review reads computed styles, which requires a script-capable renderer (Playwright):
pip install "metalworks[browser]"
metalworks browser install
A screenshot-only backend (Firecrawl) cannot read computed styles — design review raises a clear error pointing you to install the browser.