dsh-windows-remote-ssh

Local GUI automation is straightforward, but most setups assume you are physically in front of the machine. dsh-windows-remote-ssh removes that constraint: a DeepSeek Harness running on Linux, macOS, or Windows can drive a remote Windows desktop over a single plain SSH connection. The tool surface mirrors the local dsh-click plugin exactly — screenshots, UI Automation tree reads, click, type, scroll, key combos, app launch, window control, file upload/download, clipboard read/write, process listing and killing, and even PowerShell execution. The target host requires nothing beyond the Win32-OpenSSH server already bundled with modern Windows. The plugin stages a self-contained PowerShell helper over SFTP that relies solely on stock .NET automation, System.Drawing, and Win32 P/Invoke. No Python, no third-party runtime, no installer to manage.

The core challenge in GUI-over-SSH is that Win32-OpenSSH executes commands in a non-interactive Session 0 window station, leaving EnumWindows, screen capture, and UIA completely empty. This plugin schedules a one-shot task bound to the logged-on user's interactive session, triggers it immediately, and polls for a JSON response file. A minimal VBScript wrapper launches the helper with a hidden window style, so no console flash ever reaches the screen across repeated calls. Interaction defaults to UI Automation patterns or posted window messages; the real cursor never moves and no window steals focus. Hardware-level SendInput is available behind an explicit opt-in flag for apps that ignore the message queue entirely, such as games or DirectX surfaces.

The plugin is built for engineers who need to remote-drive a Windows dev box or lab VM from a Linux or macOS laptop, DevOps teams that want automated UI screenshot verification in CI pipelines without installing extra agents on runners, and heavy Harness users who want their AI conversation to see and manipulate a Windows desktop natively rather than relaying commands through a terminal.