Omarchy 4.0.3 × Midscene

Run 36134486534 · 71.0% passed · GitHub Actions · Report history

Each image is the original page screenshot used by that node. Click a case or image to open the exact Midscene step and Agent replay.

Doubao Say

10/13 cases passed

Open native Midscene report 1 → Open native Midscene report 2 → Open native Midscene report 3 → Open native Midscene report 4 →

Result	Case	Duration	Node screenshot	AI response / error
❌ Failed	Cancel sign-in and retry	5m55s		Error: Node "aiWaitFor" failed: waitFor timeout: The screenshot shows a sign-in window with a "Sign in" title, an email input field, a password input field, and a "Sign in" button. The window appears to be open and active, not closed. There is no indication that Doubao Say has advanced to a Microphone step; the visible interface is entirely focused on the sign-in form.
✅ Passed	Complete Doubao Say onboarding on a real desktop	3m33s		AI: The screenshot shows a settings page with a "Setup" section. There is a disabled control displaying text that reads "Setup completed by the synthetic E2E fixture." The control appears grayed out, indicating it is disabled. The text is clearly visible in the screenshot.
✅ Passed	Recover from an endpoint failure	1m31s		AI: The screenshot shows a page with a "Synthetic" section. There is a green checkmark icon next to "Synthetic" and a green status indicator. The page displays "Synthetic endpoint" with a green checkmark and "Success" status. There is no visible error message or "unavailable" indicator. The endpoint result appears to be successfully displayed with a green success status.
✅ Passed	Changing microphone requires another check	4m10s		AI: The screenshot shows a mobile or desktop setup page with a "Finish setup" button at the bottom. The button appears to be visually distinct, with a filled background color (likely blue or similar), indicating it is in an enabled state. There is no indication of it being grayed out or disabled. The button text is clearly visible and the overall styling suggests it is interactive and ready for use.
✅ Passed	Capture and retain a custom trigger shortcut	2m05s		AI: The screenshot shows a settings page for a keyboard shortcut or trigger configuration. The page displays a "Trigger Key" section with a key combination shown as "⌃ + F8" in a highlighted/selected state. The text indicates this is the currently active trigger, and the visual styling (highlighted box) suggests it is also the selected option. The page appears to be the Trigger key configuration page, and the displayed combination matches the statement.
❌ Failed	Toggle polishing without losing configuration	22s		Error: Node "aiAssert" failed: Node "aiAssert" failed: Assertion failed: Voice polishing disabled is visible, the switch is off, and the polishing Base URL and Model fields are hidden. Reason: Based on the screenshot, I can see a settings panel with a "Voice polishing" section. The section shows a toggle switch that appears to be in the "off" position (gray/disabled state). Below the toggle, there is a "Base URL" field and a "Model" field, both of which are visible and contain values. Since the Base URL and Model fields are visible, the statement that they are hidden is false.
✅ Passed	Complete and cancel a voice test safely	7m20s		AI: The screenshot shows a setup control labeled "Voice test" with a status of "passed" and a "Start voice test" button that appears enabled. There is no visible indication of anything pasted or sent in the screenshot.
✅ Passed	Switch to Volcengine and validate credentials	2m30s		AI: The screenshot shows a step indicator with several steps labeled, and the current step appears to be highlighted. The visible labels include "Microphone" among the steps, and the highlighted/active step is "Microphone".
✅ Passed	Deliver one synthetic dictation to the focused target	31s		AI: The screenshot shows a page with a "Synthetic dictation" section. The target field displays "Synthetic dictation delivered exactly once" and the status area shows "Delivered once" with a delivery count of "1". The text and status match the statement exactly.
✅ Passed	Preserve signed-in account across backward navigation	1m34s		AI: The screenshot shows a microphone setup step with a title "Let's make sure we can hear you." and a "Test microphone" button. The statement is true.
✅ Passed	Require an enabled trigger and retain the replacement	1m35s		AI: The screenshot shows a dialog or panel with a "Trigger" section. There is a dropdown labeled "Trigger" with the value "F8" displayed. Below it, there is a key step section showing "Active trigger F8" with a highlighted/selected state. The text "Active trigger F8" is visible, and the dropdown shows "F8" as the selected option. This matches the statement that the Trigger key step still shows Active trigger F8 and F8 is selected in the trigger dropdown.
❌ Failed	Microphone check gates setup completion	7m48s		Error: Node "aiAct" failed: Replanned 40 times, exceeding the limit. Please configure a larger value for replanningCycleLimit (or use MIDSCENE_REPLANNING_CYCLE_LIMIT) to handle more complex tasks.
✅ Passed	Cancel synthetic dictation without delivering text	2m29s		AI: The screenshot shows a UI with a section labeled "Synthetic Sentences" and a status area. There is a list of sentences with checkboxes, and a status indicator showing "1 delivered" with a success message. The interface appears to be in a state where a new attempt has been made, and the status reports exactly one delivery. The synthetic sentence count appears to be one, and the delivery status confirms this single delivery.

Omarchy visual checks

0/1 cases passed

Open native Midscene report →

Result	Case	Duration	Node screenshot	AI response / error
❌ Failed	System menu text and focused row are visible	16s		Error: Node "aiAssert" failed: Node "aiAssert" failed: Assertion failed: Exactly one row in the open system menu has a visible focus highlight or accent, and the highlight is rendered cleanly. Reason: The screenshot shows a desktop environment with a system menu open. The menu appears to have multiple rows/items. I need to check if exactly one row has a visible focus highlight or accent, and whether that highlight is rendered cleanly. Based on the visible pixels in the screenshot, I can observe the menu items and any highlighting or accent effects on them.