Environment:

• OpenClaw 2026.3.13 (61d171a)
• Windows 10 + WSL2
• Chrome CDP on 127.0.0.1:9222

Problem:
When browser-automation clicks a button that opens a native
Windows file picker dialog, the dialog is invisible to the
browser tool because it renders outside the browser context.
The agent gets stuck with no feedback.

Symptom:

• browser_click on "Agregar video" succeeds ✅
• Native Windows "Abrir" dialog opens ✅
• browser_screenshot shows nothing ❌
• Agent reports task complete or times out ❌

Root cause:
browser-automation uses Chrome CDP — it only sees the browser
DOM, not the Windows desktop layer above it.

Workaround (exec + PowerShell):
Use exec to run a PowerShell script that detects and interacts
with the native dialog.

Enable group:runtime first:
openclaw config set tools.allow '["browser","group:runtime","group:fs","group:web"]'

Then use this exec pattern from WSL2:
/mnt/c/Windows/System32/WindowsPowerShell/v1.0/powershell.exe -Command "python C:\path\to\upload_script.py video.mp4"

The script uses pyperclip + pyautogui to paste the file path
into the native dialog and confirm.

Tested with: Meta Business Suite Reels Composer
Result: Video uploaded and published successfully ✅

Related: PR #39407 (WSL2 + CDP troubleshooting guide)