Mikkel Manniche

Claude Code skill

page-critic

Reviews the pages of a web app in three layers: does it work, does it hold up, is it delightful. It measures in a real browser, lets a blind customer try the task, and gives a report with scores and concrete fixes.

  • Claude Code skill
  • MIT licence
  • Needs Node.js and Playwright

macOS and Linux

git clone https://github.com/manniche-labs/ui.git ~/manniche-labs-ui
ln -s ~/manniche-labs-ui/skills/page-critic ~/.claude/skills/page-critic

Windows (PowerShell)

git clone https://github.com/manniche-labs/ui.git "$env:USERPROFILE\manniche-labs-ui"
New-Item -ItemType Junction -Path "$env:USERPROFILE\.claude\skills\page-critic" -Target "$env:USERPROFILE\manniche-labs-ui\skills\page-critic"

Ask Claude Code to “review the pages”, or type /page-critic.

The three layers

  1. Does it work

    Console errors, dead links, forms, keyboard and focus, contrast and the size of what you tap.

  2. Does it hold up

    Long text, narrow screens, slow networks, zoom and dark mode. It finds what breaks.

  3. Is it delightful

    Rhythm, type, motion and finish, and whether the page looks like an AI made it.

  4. Before and after

    It shows three variants of a visible fix, writes a prompt for whoever fixes it, and measures again afterwards.

What you get

  • A report with scores per page and layer.
  • Faults with the exact element and reason.
  • Proposals for UI, animation and finish.
  • Numbers before and after, so the improvement shows.