Screen Monitor

**Verdict:** A high-utility screen-visibility and control skill that meaningfully upgrades an agent’s ability to debug, guide users, and automate browser-based workflows. The design balances a fast, low-friction sharing path with a more powerful full-control mode. **Strengths** - Dual modes cover both quick visual checks (WebRTC Fast Share) and deep interaction (browser relay with snapshot/click). - Model-agnostic vision analysis makes it flexible across different backends. - Good fit for debugging, live support, UI walkthroughs, and visually grounded reasoning on arbitrary web pages. **Key Risks / Limitations** - **Privacy and data exposure:** Sharing the screen or an attached browser tab can expose sensitive information (credentials, personal data, internal dashboards). Users must carefully limit what is visible and when sharing is active. - **Unintended actions:** With `browser.click` and full-control mode, the agent could perform destructive or irreversible actions (e.g., deleting data, confirming purchases, changing settings) if not properly constrained. - **Extension and network dependencies:** Reliability depends on the Chrome extension being correctly installed, the local backend running on port 18795, and WebRTC functioning in the user’s environment; these introduce potential setup and connectivity failure points. **Recommended Scenarios** - Supervised sessions where the user is actively watching and can immediately stop sharing if the agent behaves unexpectedly. - Technical support, QA, and debugging of browser-based tools, especially when describing the UI in text would be cumbersome. - Controlled UI automation tasks where the set of allowed actions and pages is pre-defined and monitored to mitigate safety and privacy risks.