The problem
The wrapper is a desktop shell I built to replace the Windows taskbar, designed around running many Claude Code sessions at once.
How it works
The stock taskbar shows windows, not work. With a dozen agent sessions open across two monitors, I need to see which session is doing what, which one is waiting for me, and how much of my plan I have used. The wrapper draws its own bars on every monitor at that screen's native scaling, and rebuilds them when a screen is plugged in or removed. The real taskbar stays hidden.
Under the hood, a small local server feeds web views docked as app bars to the screen edges. A supervisor watches every piece and restarts any that crash. Every fifteen seconds a gate checks for changes and runs the fast test suite on them: a pass auto-commits, a failure reverts and restarts.
- Bars on every monitor at native DPI
- Rebuilt when a screen is plugged in
- A supervisor restarts any crashed piece
Outcome
This is the desktop I work in every day. I edit the shell constantly, and the safety net keeps it working anyway.
Stack
- claude-code
- windows
- webview2
- desktop-shell
- developer-tools
Transcript
I replaced the Windows taskbar with a shell built for AI sessions. Every monitor gets its own bars, drawn at that screen's scaling, and they rebuild when a screen is plugged or unplugged. The real taskbar stays hidden, and a supervisor restarts any piece that crashes. Behind it is a small local server and web views docked to the screen edges. Every fifteen seconds a gate runs the fast tests on any change: pass commits, fail reverts.
How this was made
Voiceover is text-to-speech (Microsoft Edge neural voice). Background art is AI-generated (Gemini). Titles and captions are set in code, not generated. The build itself was made with Claude Code.