winui-ui-testing

Scope — any Windows app

winapp ui drives Windows UI Automation (UIA), the accessibility layer every Windows UI framework exposes, so the AutomationId-based approach in this skill works on any Windows desktop app: Win32, WPF, WinForms, and WinUI 3, packaged or unpackaged. The file-picker tests below already drive the OS's Win32 file dialog through the same verbs. For a non-WinUI app, use the same verbs and script template and skip the WinUI-specific gotchas (x:Bind LostFocus commit, ContentDialog selectors, MSIX relaunch).

Approach

The goal of this skill is to validate UI and app functionality automatically, without manual interaction, by exercising the app's UI elements, verifying their state, and asserting that the app behaves as expected under test conditions.

There are two main approaches:

1. Interactive exploration — manually run the app, use winapp ui <command> to explore the UI tree, find AutomationIds, verify element properties, and test functionality interactively. This is useful for discovery, but slow and expensive if repeated for every test iteration.
2. Scripted batch testing — generate a ui-tests.ps1 script that exercises all UI elements and asserts expected behavior in one pass. This allows you to run the tests automatically, capture results, and iterate quickly without manually interacting with the app each time.

Unless the user asked for interactive exploration, or you are unfamiliar with the code/app or need to explore the UI tree to discover AutomationIds for hidden or dynamically generated elements (flyouts, dialogs, lazy-loaded content), prefer scripted batch testing — it is faster, repeatable, and produces a record of pass/fail results that can be reviewed and acted on.

winapp ui Verbs

• Query: status, list-windows, inspect, search, get-property, get-value, get-focused, wait-for
• Interact: invoke, click, set-value, focus, scroll, scroll-into-view
• Advanced input: send-keys (synthetic keyboard + accelerators), hover (tooltips/flyouts), drag (drag-drop, reorder, sliders), touch (tap/swipe/pinch/stretch), pen (stylus ink, pressure/tilt/eraser)
• Capture: screenshot, record (H.264 MP4 video)

Run winapp ui --cli-schema for the complete command structure as JSON, or winapp ui <verb> --help for any single verb.

Step 1: Use the Running App

If the app is already running, use its PID. Do NOT relaunch — use the PID already captured from the build step. If the app is not running, build and launch it using the guidance in the winui-dev-workflow skill.

Step 2: Write the Test Script

If you wrote the code: Skip inspect — you already know all the AutomationIds and control structure from the XAML and code-behind. Write tests directly from that knowledge. Inspect misses popups, flyouts, dialogs, and lazy-loaded content anyway.

If you're verifying code you didn't write: Run inspect first to discover the UI:

winapp ui inspect -a <PID> --interactive

Then read the XAML files to find AutomationIds that aren't currently visible (flyout items, dialog buttons, secondary pages).

Create a ui-tests.ps1 file that tests all the app's requirements in one pass:

# ui-tests.ps1
param([Parameter(Mandatory)][int]$AppPid)
# NOTE: Do NOT name the parameter $Pid — it's read-only in PowerShell

$ErrorActionPreference = 'Continue'
$pass = 0; $fail = 0; $results = @()

# Get main window HWND (avoids PopupHost interference with JSON parsing)
$windows = winapp ui list-windows -a $AppPid --json 2>$null | ConvertFrom-Json
$hwnd = ($windows | Where-Object { $_.title -ne "PopupHost" } | Select-Object -First 1).hwnd

function Test-UI {
    param([string]$Name, [scriptblock]$Script)
    # IMPORTANT: Inside $Script, use 'throw' to signal failure — NOT 'exit 1'
    # (exit terminates the entire script, not just the test)
    try {
        $output = & $Script 2>&1
        if ($LASTEXITCODE -eq 0) {
            $script:pass++; $script:results += @{ name = $Name; status = "PASS" }
        } else {
            $script:fail++; $script:results += @{ name = $Name; status = "FAIL"; detail = "$output" }
        }
    } catch {
        $script:fail++; $script:results += @{ name = $Name; status = "FAIL"; detail = "$_" }
    }
}

# ─── Element Existence ───
Test-UI "NavHome exists" { winapp ui wait-for "NavHome" -a $AppPid -t 3000 }
Test-UI "NavSettings exists" { winapp ui wait-for "NavSettings" -a $AppPid -t 3000 }

# ─── Navigation ───
Test-UI "Navigate to Settings" { winapp ui invoke "NavSettings" -a $AppPid }
Test-UI "Settings page loaded" { winapp ui wait-for "TxtUserName" -a $AppPid -t 3000 }

# ─── Interactions ───
Test-UI "Set username" { winapp ui set-value "TxtUserName" "TestUser" -a $AppPid }
Test-UI "Click Save" { winapp ui invoke "BtnSave" -a $AppPid }  # commits the TextBox binding
Test-UI "Username value set" {
    winapp ui wait-for "TxtUserName" -a $AppPid --value "TestUser" -t 2000
}

# ─── Value assertions for different control types ───
Test-UI "Theme is System default" {
    winapp ui wait-for "CmbTheme" -a $AppPid --value "System default" -t 2000
}
Test-UI "Logging is off" {
    winapp ui wait-for "TglLogging" -a $AppPid --value "Off" -t 2000
}

# ─── Accessibility Audit ───
# Only audit controls in the app's main window (exclude OS picker/popup controls)
$allElements = (winapp ui inspect -a $AppPid --interactive --json 2>$null | ConvertFrom-Json).elements
$appElements = @($allElements | Where-Object {
    $_.type -match 'Button|TextBox|ComboBox|CheckBox|ToggleSwitch|TabItem|Edit' -and
    $_.name -notmatch 'Minimize|Maximize|Close|System' -and          # window chrome
    $_.className -notmatch 'PickerHost|#32770|CabinetWClass'         # OS dialogs
})
$missingId = @($appElements | Where-Object { -not $_.automationId })
if ($missingId.Count -eq 0) {
    $pass++; $results += @{ name = "All app controls have AutomationId"; status = "PASS" }
} else {
    $fail++
    $names = ($missingId | ForEach-Object { "$($_.type) '$($_.name)'" }) -join ", "
    $results += @{ name = "AutomationId coverage"; status = "FAIL"; detail = "Missing: $names" }
}

# ─── State Screenshots (capture each meaningful state for visual review) ───
New-Item -ItemType Directory -Force -Path "screenshots" | Out-Null
winapp ui screenshot -a $AppPid -o "screenshots/01-initial.png" 2>$null
# ...take more screenshots after key interactions above (mode switches, dialogs opened, etc.)

# ─── Final Screenshot ───
winapp ui screenshot -a $AppPid -o "test-screenshot.png" 2>$null

# ─── Results ───
Write-Host "`nPassed: $pass | Failed: $fail"
$results | Where-Object { $_.status -eq "FAIL" } | ForEach-Object {
    Write-Host "  FAIL: $($_.name) — $($_.detail)" -ForegroundColor Red
}
$results | ConvertTo-Json | Out-File "test-results.json"
if ($fail -gt 0) { exit 1 } else { exit 0 }

What to Test

Write tests for every requirement from the user's prompt:

Step 3: Run and Read Results

.\ui-tests.ps1 -AppPid <PID>

Read test-results.json for structured pass/fail. Only fix code if tests fail.

Step 3.5: Look at the Screenshots

UIA assertions don't see clipping, overlap, wrong theming, or controls bleeding past their container — UIA returns PASS while the app is visually broken. Capture screenshots with winapp ui screenshot and view each PNG.

Capture the initial state and any state after a major interaction (the State Screenshots block in the script template above handles this).

Visual checklist — fail the run if any item is no:

• No unintended scrollbars
• No text ending in … that shouldn't be
• Hero elements fully visible (not sliced)
• Right-edge controls fully visible
• No overlapping rows
• Content uses the available width — no asymmetric dead zones (e.g. content pinned to one edge leaving empty space on the other)
• Spacing intentional — not cramped, not unintentionally vast
• Theming matches the user's ask (Light/Dark/HighContrast if relevant)
• Focus/hover/error states render if tested

If the checklist fails, it's a bug — fix before declaring done. Window too small → grow per winui-design Step 4.

Step 4: Fix and Rerun (if the user asked for it)

If tests fail:

1. Read the failure details from test-results.json
2. Batch-fix all issues in one pass
3. Rebuild with .\BuildAndRun.ps1 (blocking mode — shows crash info if the fix broke something)
4. Rerun .\ui-tests.ps1 -AppPid <PID> (parse PID from the launched (PID: XXXXX) output)

Maximum 2 fix-and-rerun cycles. If the same tests keep failing after 2 cycles, report them as known issues and move on — do not keep iterating.

Assertion Reference

Use wait-for --value as the primary assertion — it uses a smart fallback chain that reads the right value for any control type:

Full assertion commands:

Testing File Pickers

File/folder pickers (FileOpenPicker, FileSavePicker, FolderPicker) run in a separate PickerHost process but are fully interactable. The picker appears as an owned dialog window.

# 1. Trigger the picker
winapp ui invoke "BtnOpenFile" -a $AppPid

# 2. Find the picker window (it's a dialog owned by the app window)
Start-Sleep 1
$allWindows = winapp ui list-windows -a $AppPid --json 2>$null | ConvertFrom-Json
$picker = $allWindows | Where-Object { $_.title -match "Open|Save" }
$pickerHwnd = $picker.hwnd

# 3. Interact with the picker using -w <HWND>
#    Type a filename:
winapp ui set-value "FileNameControlHost" "test.txt" -w $pickerHwnd
#    Click Open/Save:
winapp ui invoke "Open" -w $pickerHwnd     # or "Save", "Cancel"
#    Or cancel:
winapp ui invoke "Cancel" -w $pickerHwnd

# 4. Verify the app processed the file
winapp ui wait-for "StatusBar" -a $AppPid -p Name --value "opened" -t 3000

Tip: Use winapp ui inspect -w <pickerHwnd> --interactive to discover the picker's controls — they include the folder tree, file list, filename textbox, and Open/Cancel buttons.

Testing Context Menus and Flyouts

MenuFlyouts and ContextFlyouts are fully testable. They appear in the UI automation tree when open.

# 1. Right-click to open a ContextFlyout
winapp ui click "LstItems" -a $AppPid --right
Start-Sleep 0.5

# 2. The flyout MenuItems appear in the tree immediately
#    Find them with inspect or search:
winapp ui inspect -a $AppPid --interactive   # shows MnuCopy, MnuDelete, etc.

# 3. Click a flyout item
winapp ui invoke "MnuCopy" -a $AppPid

# 4. Verify the action
winapp ui wait-for "StatusText" -a $AppPid -p Name --value "Copied" -t 2000

For MenuBar flyouts (File, Edit, View menus):

# Click the menu header to open
winapp ui invoke "FileMenu" -a $AppPid
Start-Sleep 0.5
# Click the sub-item
winapp ui invoke "MenuSaveAs" -a $AppPid

Testing ContentDialogs

ContentDialogs are in-app controls (same window) — they appear directly in the UI tree when shown.

# 1. Trigger the dialog
winapp ui invoke "BtnDelete" -a $AppPid
Start-Sleep 0.5

# 2. The dialog buttons appear in the tree
#    For a standard confirmation dialog:
winapp ui search "Primary" -a $AppPid --json   # finds the primary button
winapp ui invoke "Primary" -a $AppPid           # click "Yes"/"Delete"/"Save"
#    Or:
winapp ui invoke "Secondary" -a $AppPid         # click "No"/"Don't Save"
winapp ui invoke "Close" -a $AppPid             # click "Cancel"

# 3. Wait for dialog to dismiss
winapp ui wait-for "Primary" -a $AppPid --gone -t 3000

Tip: ContentDialog buttons often don't have custom AutomationIds — use inspect to find the actual selector (slug or text match).

Advanced Input: keyboard, hover, drag, touch & pen

Synthetic input beyond invoke/click/set-value. Each verb takes -a <PID> / -w <HWND> like the rest.

send-keys — real keyboard input. Named keys (enter, tab, f5), combos (ctrl+shift+t), raw vk=0x42, or literal text. --via selects the transport:

• post-message (default) — HWND-targeted, no foreground needed; raises TextChanged but not per-character KeyDown.
• send-input — OS-wide; real per-character KeyDown + TextChanged. Required for accelerators/shortcuts (KeyboardAccelerator, e.g. ctrl+t) and for reliable typing into a WinUI 3 / WPF TextBox.

winapp ui send-keys "ctrl+s" -a $AppPid --via send-input                          # fire a Ctrl+S accelerator
winapp ui send-keys "hello world" --target "TxtName" -a $AppPid --via send-input  # focus then type
winapp ui send-keys --verbatim "down down enter" -a $AppPid                       # type the words, not the keys

--target focuses first; text=<literal> / --verbatim type literally instead of interpreting key names. System combos (win+r, alt+f4) need --allow-system-keys + --via send-input (win+l stays blocked).

hover — tooltips, flyouts, hover states. Dwells on the element (--dwell-time, default 800 ms) so hover-triggered UI appears in the tree.

winapp ui hover "BtnInfo" -a $AppPid
winapp ui wait-for "InfoTooltip" -a $AppPid -t 2000

drag — drag-drop, reorder, resize, sliders. <from>/<to> are each an element selector (its center) or screen x,y from inspect. --hold-ms long-presses before moving; --dwell-ms settles on the target before releasing (merge/latch targets).

winapp ui drag "ItemA" "ItemB" -a $AppPid          # reorder ItemA onto ItemB
winapp ui drag "SldVolume" 300,120 -a $AppPid      # drag a slider thumb to a point

touch — touch gestures. -g: tap (default), double-tap, long-press, swipe, pinch, stretch; --direction/--distance/--to-point for swipes, --fingers for multi-touch. Needs an unlocked interactive desktop with the window foregroundable.

winapp ui touch "LstFeed" -g swipe --direction up --distance 400 -a $AppPid
winapp ui touch "ImgPhoto" -g stretch --distance 200 -a $AppPid    # pinch-to-zoom

pen — pen/stylus. Taps or draws ink; --path "x,y x,y …" for a multi-point stroke, with --pressure, --tilt-x/--tilt-y, --eraser (Win10 1809+).

winapp ui pen "InkCanvas" --path "50,50 120,80 200,60" --pressure 0.8 -a $AppPid
winapp ui pen "InkCanvas" --path "50,50 200,60" --eraser -a $AppPid

Recording a Video

winapp ui record captures the target window (or an element region) to an H.264 MP4 — handy for a repro clip of a flow or animation. Records until stopped (newline/EOF on stdin, or Ctrl+C); --duration-sec N gives a fixed-length clip (simplest for scripts). The MP4 is finalized on graceful stop.

winapp ui record -a $AppPid --duration-sec 6 --fps 30 -o "flow.mp4"

--max-edge N downscales large windows; --capture-screen uses screen BitBlt to include popups/overlays outside the target window (also on screenshot).

Key Gotchas

• set-value does NOT commit default TextBox bindings — WinUI 3 x:Bind TwoWay on TextBox.Text updates the ViewModel on LostFocus by default. UIA set-value changes the text but doesn't trigger focus events. Fix: apps should use UpdateSourceTrigger=PropertyChanged on TextBox bindings (see design skill). If the app doesn't, invoke a button or click/focus another element after set-value to trigger LostFocus.
• Set a RichEditBox with send-keys, not set-value — WinUI 3 RichEditBox / WPF RichTextBox don't support UIA value-setting. focus (or --target), then send-keys "…" --via send-input — which also raises real per-key KeyDown, so use it whenever a control reacts to individual keystrokes (or a KeyboardAccelerator) rather than a bulk value change.
• Verify persistence via the data file, not UI relaunch — killing and relaunching a packaged app from a test script is fragile (MSIX registration timing, PID issues). Instead, check the data file on disk: Get-Content $dataFile | ConvertFrom-Json and verify expected values.
• Use $AppPid not $Pid — $Pid is a read-only automatic variable in PowerShell
• Use --value without -p — it auto-detects the right UIA pattern (TextPattern → ValuePattern → TogglePattern → SelectionPattern → Name). Only use -p PropertyName --value when you need a specific property like IsEnabled
• File pickers need -w <HWND> — they run in a separate PickerHost process, so -a PID won't find them. Use list-windows to discover the picker HWND first
• Flyouts need a short Start-Sleep after triggering — the menu items appear in the tree asynchronously