Publish ggui chat loop as a Claude Code skill (/ggui replaces /remote-control)

Summary

The ggui chat loop pattern (push UI → consume events → send responses) should be published as a Claude Code skill that any agent can invoke with /ggui.

This replaces the current /remote-control pattern with a first-class ggui integration.

How It Works

Today, the sample agent uses a system prompt (CLAUDE.md) that instructs Claude to use ggui_push, ggui_consume, and ggui_send in a loop. This pattern should be packaged as a reusable Claude Code skill:

/ggui "Start a customer support chat"

This would:

1. Configure MCP connection to ggui (if not already configured)
2. Call ggui_push to create the appropriate UI
3. Enter a ggui_consume → think → ggui_send loop
4. Continue until session completes

Core Principle

This aligns with the Zero Agent Code principle (docs/principles/zero-agent-code.md): agents need only MCP config + system prompt. A skill makes this even more frictionless — zero config, just /ggui.

Tasks

• [ ] Package the chat loop system prompt as a Claude Code skill
• [ ] Publish to Claude Code skill registry
• [ ] Update docs to reference /ggui instead of /remote-control
• [ ] Add skill variants: /ggui chat, /ggui form, /ggui dashboard

🤖 Generated with Claude Code