Skip to content

Windows support #2

@localkin

Description

@localkin

Phase 6 Windows port shipped 2026-05-12 — code complete, awaiting community runtime testing.

This is the Windows-side companion to #1 (Linux). Same status: the agent wrote everything from API docs without a Windows box to run against. Builds are green for both windows/amd64 and windows/arm64; everything past that needs someone with hardware.

What landed

  • 4 native claws (pkg/skill/*_windows.go)
    • screen — PowerShell + System.Drawing.Graphics.CopyFromScreen
    • inputuser32.dll P/Invoke (SetCursorPos, mouse_event) + SendKeys
    • ui — UI Automation 2.0 (UIAutomationClient + UIAutomationTypes)
    • recordffmpeg -f gdigrab -i desktop
  • 4 cerebellum categories (skills/cerebellum/categories/windows-*.sh)
    • windows-files, windows-apps, windows-settings, windows-clipboard
    • All shell out to powershell.exe; runs under WSL, Git Bash, MSYS2
  • souls/pilot_windows.soul.md — 23 skills

Details: see CHANGELOG entry 2026-05-12 (overnight) and roadmap Phase 6.

What we need from testers

Grab the Windows binary from the latest release (or from the latest CI run — artifacts kept 7 days) and try a few common flows. See TESTING.md for the suggested smoke commands + report template.

Tested-most-likely-to-need-refinement areas — please prioritise these:

  1. PowerShell escape semantics inside Go raw strings — one backtick clash already caught + fixed during build; more may surface at runtime
  2. UIA InvokePattern on UWP apps — Win32 buttons fine; UWP sometimes only expose SelectionItemPattern / TogglePattern
  3. ffmpeg gdigrab on Windows 11 — DRM-protected windows may show black
  4. Windows.Devices.Radios Wi-Fi/Bluetooth toggle — may need admin on some configs (the soul's toggle_wifi OFF is intentionally refused per paper #11 §5.1)
  5. CloseMainWindow → Stop-Process fallback — apps with "Save changes?" dialogs will currently force-kill after 2s
  6. key_down / key_up — currently degraded (SendKeys auto-releases). True modifier-hold needs raw SendInput; not implemented yet

Both ✅ and ❌ reports are useful. If you only have 10 minutes, the highest-value tests are:

  • kinclaw.exe -soul souls\pilot_windows.soul.md -exec "screenshot the desktop" (screen claw)
  • kinclaw.exe -soul souls\pilot_windows.soul.md -exec "what app is focused" (ui claw)
  • kinclaw.exe -soul souls\pilot_windows.soul.md -exec "copy hello to clipboard" (cerebellum windows-clipboard)

Environment

PowerShell 5.1 ships with every supported Windows version (7 SP1+) so no third-party install is needed beyond ffmpeg if you want record. Tested on amd64 + arm64 (Surface, Snapdragon X).

cc @tinkerbaj — same testing-request pattern as #1.

Metadata

Metadata

Assignees

No one assigned

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions