AgentUI

AgentUI is a skill-driven benchmark project to improve UI implementation quality for coding agents (Codex + Claude).

Goals

• Improve trigger accuracy for the ui-architect skill frontmatter
• Benchmark output quality before and with skill instructions
• Produce auditable side-by-side artifacts and scoring

Structure

• .agents/skills/ui-architect/ — skill package used by agents
• benchmarks/prompts/ — test prompts
• benchmarks/baseline/ — outputs without skill guidance
• benchmarks/with-skill/ — outputs with skill guidance
• benchmarks/analysis/ — side-by-side evaluation docs
• docs/ — reports and drive-share artifacts

Benchmark Domains

1. Fintech dashboard
2. Healthcare settings/preferences
3. Ecommerce search results + filters