v2.0.6 — Background Window Capture

What's New

Background Window Capture

• Screenshots now capture windows even when they're behind other windows or in another Space
• Removed optionOnScreenOnly restriction from CGWindowListCopyWindowInfo
• No need to focus an app before taking a screenshot (unless it's minimized)

Better Error Messages

• Specific error messages instead of generic "Screenshot capture failed"
• "Application has no open windows" when app has no windows
• "Screen Recording permission not granted" only when actually missing
• Window ID and app name included in error details

FocusManager Improvements

• Polling-based focus verification (up to 1s) instead of fixed sleeps
• More reliable focus confirmation

VisionPerception Retry Logic

• Vision capture now retries with app activation (matching screenshot behavior)
• Defense-in-depth for VLM grounding pipeline

Updated Agent Instructions (GHOST-MCP.md)

• Rule 8: Web App Interaction — use dom_id for Chrome/Electron apps
• Rule 9: Gmail/Email Pattern — URL-based compose for reliable email sending
• Rule 10: Coordinate Mapping — screenshot coords vs screen coords
• Rule 11: Vision Grounding Best Practices — always use crop_box
• Rule 12: Wait Between Actions — ghost_wait after clicks

Screenshot Response

• Now includes window_frame in response for coordinate mapping

Full Changelog: v2.0.5...v2.0.6