Turn your AI agent into a persnickety UI reviewer

What it does

This repo is a bundle of seven specialist reviewers for AI coding agents. Each skill—typography, color, accessibility, layout, writing, UI polish, and a coordinating interface review—acts as a focused prompt that nudges agents like Claude Code or Codex toward design-system thinking rather than just functional code.

The interesting bit

Rather than baking design opinions into a custom model, the project treats interface quality as a modular, on-demand layer. You invoke a skill explicitly, set the depth, and point it at a specific screen or flow, turning a generic agent into a temporary UX consultant.

Key highlights

• Seven discrete skills covering UI polish, typography, OKLCH color handling, layout, accessibility, and interface copy.
• Top-level better-interface skill coordinates the others for holistic reviews.
• Integrates with agent-native IDEs like Claude Code or Codex.
• Review depth and target are adjustable (e.g., a quick pass versus a full review on a specific flow).

Caveats

• Tightly coupled to the external skills.sh ecosystem; unclear how it functions outside that CLI.
• The README is vague about implementation details—each skill appears to be a SKILL.md file, but the exact mechanics are unexplained.

Verdict

Worth a look if you’re already using agent-native IDEs and tired of manually reminding your agent to check contrast ratios. Skip it if you want standalone linting tools or don’t work inside an agent-centric workflow.