Update to Lumen UI

{
  "description": "Review an existing Lumen interface or proposed UI change for component contracts, setup, semantic tokens, accessibility, and responsive behavior. Use for requested Lumen UI audits and reviews.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 229
    }
  ],
  "name": "lumen-review",
  "skill_md_contents": "---\nname: lumen-review\ndescription: Review an existing Lumen interface or proposed UI change for component contracts, setup, semantic tokens, accessibility, and responsive behavior. Use for requested Lumen UI audits and reviews.\n---\n\n# Review a Lumen interface\n\nReview the requested surface and return actionable findings with file locations, observable\nconsequences, and suggested fixes. Preserve the user's explicit review scope. A review request\ndoes not authorize edits or dependency upgrades.\n\nIdentify the framework and resolved installed Lumen versions. Compare them with `lumen_get_meta`\nand `lumen_check_compatibility` when MCP is connected. Read installed public types and documentation\nwhen versions differ; a healthy server is not proof that its catalog matches the application.\nWithout MCP, use installed types, package READMEs, and `lumen show <component>`.\n\nInspect public component imports and props, theme/style ownership, and the framework's setup:\nAstro runtime once, React hooks/controllers, Elements registration, or native providers/themes.\nRead only the matching setup reference from\n[framework contracts](../lumen-ui/references/frameworks.md) when checking adapter-specific details.\nFor v5, check component CSS coverage, await Astro enhancement, and account for lazy first\ninteractions and navigation cleanup. Full stylesheet usage is supported, not automatically a defect.\nKeep source-only setup findings separate from measured route requests.\nLook for duplicate primitive implementations, incorrect compound children, raw palettes,\napplication CSS that breaks public slots, and unsupported adapter APIs.\n\nReview actual interactions and states: keyboard/focus and dismissal, accessible names, validation,\nloading/error/empty feedback, touch targets, reduced motion, and phone/desktop layout. Use existing\nproject checks and rendered evidence when permitted; distinguish source findings from verified\nbehavior. Native reviews use platform semantics and supported target availability. Use\n[the interaction checklist](../lumen-ui/references/verification.md) for the reviewed behavior;\nreport unavailable checks without treating source inspection as rendered verification.\n\nFor visualization, check that the encoding answers the product question, the selected chart exists\nin the installed adapter, and its data shape preserves unique identities, missing values, and real\nzeros. Verify localized context, readable exact data, and supported keyboard inspection. Applications\nown aggregation, conversion rates, binning, and statistical policy; charts must not invent them.\n\nUse `lumen doctor --json` or `lumen doctor-native --json` for existing diagnostics when installed.\nExplain applicability instead of treating every advisory as a defect. Do not upload private\nsource, credentials, or user data to the public catalog service.\n\nReport only evidence-backed issues in priority order. Say when no actionable findings remain,\nlist the checks performed, and identify any unverified behavior. Do not claim WCAG certification\nfrom an automated accessibility scan.\n"
}