Update skill description to focus on second opinions and code review

Auto-determine approval mode from quick reference instead of asking user

Hardcode model choices and use generic prompt function

Remove sandbox references and make agent-harness-agnostic

Improve step clarity, yolo/approval-mode docs, and add large I/O guidance

Limit model options to gemini-3.1-pro-preview and gemini-3-flash-preview

Include model name in peer AI identification

Make agent identity generic, not Claude-specific

Initial release of skill-gemini v1.0.0