- Home
- Skills
- Testing & Debugging
- offstage-windows
Works with the AI Agents you already use
Offstage Windows
Run Electron and Playwright GUI tests on Windows without them touching your screen: each run gets a hidden desktop, so no windows pop up, no focus is stolen, and test runs can go side by side.
Free
About this skill
Automated runs of a desktop app open real windows: end-to-end tests, screenshot scripts, smoke tests of the packaged app. On Windows those windows pop up on your screen and take your keyboard focus while you work. When a coding agent runs your tests, that happens all day.
This skill teaches your agent to use offstage-windows (npm, MIT), which runs the test session on a hidden Windows desktop. The app renders as it does on screen (GPU, animations, focus, screenshots). You just never see it.
What your agent does with it
- Wraps the test command (
npx offstage-windows npx playwright test) for the quick start. - Wires the launch sites when you want every path covered: Playwright's Electron launch, scripts that spawn Electron or a packaged exe, and helpers that work with the app's windows by handle.
- Checks the result with a JSON report per launch, instead of running the suite visibly to compare.
- Runs Playwright workers side by side once each spec file is isolated. On two real Electron apps that took an e2e suite from 62 s to 26 s.
Where it stops
- The hidden desktop is Windows only. On macOS and Linux commands pass through unchanged (on Linux, use xvfb-run).
- Explorer, the default browser, toast notifications and UAC prompts still reach the screen, so tests should stub them. The clipboard is shared with your desktop.
- Some endpoint-protection products flag hidden desktops and on-machine compiling, because malware uses both. If the helper is blocked, your tests run with visible windows as before.
The skill is instructions only: your agent adds the package with npm. The package is one file with no dependencies, and its full README ships with the skill.
Windows only for the hidden desktop; elsewhere commands run as is. Windows other programs open for your app (Explorer, the default browser, toasts, UAC prompts) still show, and the clipboard is shared with your desktop. A native dialog the app opens waits unseen until the test times out; OFFSTAGE=0 shows it. Real OS input (SendInput, robotjs) doesn't reach the app; Playwright's and CDP's input do.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 9 days ago
- Free with an Agensi account
Requires