Feature Request: Browser/Web Bridge Skill (CDP-based local browser control)

[lee-muriithi-kingori]

Feature Request: Browser/Web Bridge Skill (CDP-based local browser control)

Problem

AI agents can reason and plan but cannot act on the web. They can't click buttons, fill forms, extract data from authenticated pages, or navigate multi-step web workflows.

Proposed Solution

A browser bridge skill that lets agents control a real Chrome/Edge browser via Chrome DevTools Protocol (CDP):

• Connect to existing Chrome with remote-debugging-port
• Actions: navigate, click, type, scroll, screenshot, extract DOM
• Local-first: all data stays on-device (no cloud proxy)
• Supports authenticated sessions (agent works with real logged-in pages)

Reference Implementations

• browser-use (https://github.com/browser-use/browser-use) - 102k stars, MIT licensed, powers Manus AI
• Kimi WebBridge by Moonshot AI - Chrome extension + local service, supports Claude Code/Cursor/Codex
• MiniMax already has co-browser skill infrastructure to build on

Why This Matters for MiniMax Ecosystem

• Turns MiniMax agents into true hands-on agents that complete real tasks
• Competitive with Kimi WebBridge and Manus Browser Operator
• Open-source + local-first = privacy-respecting
• Enables: automated research, form filling, data extraction, cross-site workflows

What I Can Contribute

Happy to build and submit a PR for this skill. I have working CDP bridge implementation and Chrome extension scaffolding.