screen-monitor/SKILL.md

name, description, metadata

name	description	metadata
screen-monitor	Dual-mode screen sharing and analysis. Model-agnostic (Gemini/Claude/Qwen3-VL).	bot emoji requires 🖥️ model_features vision

Screen Monitor

This skill provides two ways for the agent to see and interact with your screen.

🟢 Path A: Fast Share (WebRTC)

Best for: Quick visual checks, restricted browsers, or non-technical environments.

Tools

• screen_share_link: Generates a local WebRTC portal URL.
• screen_analyze: Captures the current frame from the portal and analyzes it with vision.

Usage:

# Get the link
bash command:"{baseDir}/references/get-share-url.sh"

# Analyze
bash command:"{baseDir}/references/screen-analyze.sh"

---

🔵 Path B: Full Control (Browser Relay)

Best for: Deep debugging, UI automation, and clicking/typing in tabs.

Setup

1. Run bot browser extension install.
2. Load the unpacked extension from bot browser extension path.
3. Click the Bot icon in your Chrome toolbar to Attach.

Tools

• browser action:snapshot: Take a precise screenshot of the attached tab.
• browser action:click: Interact with elements (requires profile="chrome").

---

Technical Details

• Port: 18795 (WebRTC Backend)
• Files:
  • web/screen-share.html: The sharing portal.
  • references/backend-endpoint.js: Frame storage server.