GitHub

AI Skills

Give coding agents a safe, repeatable workflow for building complete Glass UI pages.

Portable Agent Skill

Turn a page brief into project-native Glass UI source, without hiding your UI behind a package.

React 18+Tailwind CSS v4MIT

Purpose

glass-ui-page-builder teaches OpenCode, Claude Code, Codex, and Gemini to inspect your project before generating a complete page. It preserves the copy-source model, reuses local components, applies the glass, glass-soft, and glass-strong visual grammar, and validates the result.

The skill includes practical patterns for landing pages, dashboards, settings screens, and content workspaces.

Requirements

  • An existing React 18+ project using Tailwind CSS v4
  • Glass UI initialized with a valid glass.config.json
  • A coding agent with Agent Skills support
  • Permission for the agent to inspect the project and run its validation scripts

Install

OpenCode, Codex, or Gemini

Install the skill in the portable .agents/skills location:

mkdir -p .agents/skills/glass-ui-page-builder
curl -fsSL https://ui-glass.vercel.app/skills/glass-ui-page-builder/SKILL.md \
-o .agents/skills/glass-ui-page-builder/SKILL.md

Claude Code

Claude Code can load the same file from .claude/skills:

mkdir -p .claude/skills/glass-ui-page-builder
curl -fsSL https://ui-glass.vercel.app/skills/glass-ui-page-builder/SKILL.md \
-o .claude/skills/glass-ui-page-builder/SKILL.md

Start a new agent session

Restart or open a new session so the agent discovers the installed skill. Keep the skill in the repository when you want the workflow shared with your team.

Invoke it

Name the skill and describe the route, audience, primary action, and required states. The agent should inspect the existing project before proposing files or CLI changes.

AgentInvocation
OpenCodeMention glass-ui-page-builder in the request or let the agent load it
Claude CodeRun /glass-ui-page-builder before the page brief
CodexInclude $glass-ui-page-builder in the prompt
GeminiConfirm it with /skills list, then mention glass-ui-page-builder
Use glass-ui-page-builder to build an operations dashboard at /dashboard.
Reuse local components, include an accessible activity table and mobile navigation,
support light and dark themes, and run the project's targeted checks.

More useful prompts:

  • Landing: “Create a conversion-focused landing page for a privacy-first finance app, with product proof, responsive navigation, and one dominant CTA.”
  • Settings: “Build /settings/profile with validation, save feedback, keyboard access, and a clearly separated danger zone.”
  • Workspace: “Create /articles as a responsive list-detail content workspace with search, draft status, empty states, and protected unsaved work.”
  • Audit: “Audit this page for Glass UI grammar, copy-source imports, accessibility, responsive behavior, and theme parity; make only the necessary fixes.”

Safety and limits

The skill directs agents to run read-only doctor, list, and info checks before using add, and to add only named components that are actually missing. It never permits package imports in generated application code.

It also prohibits --force, --overwrite, and --all unless you explicitly authorize the exact operation. Review generated source and diffs before merging; the skill improves agent behavior but does not replace product requirements, visual review, or accessibility testing with real users.

Inspect the portable source

One Markdown file, no manifest or runtime dependency.