From 2118ba08e8af58f4e096a1f51e25688cdf641f7b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 17:28:18 -0300 Subject: [PATCH 1/7] Mobile: read the placeholder a Compose Multiplatform field folds into its iOS label (upstream 75ae2ab) Co-Authored-By: Claude Sonnet 5.5 --- COMPATIBILITY.md | 1 + skills/e2e-dotnet/references/mobile.md | 1 + src/E2E.Mobile/SnapshotProjection.cs | 47 +++++++++++++++++-- .../Nodes/SnapshotProjectionTests.cs | 44 +++++++++++++++++ 4 files changed, 90 insertions(+), 3 deletions(-) diff --git a/COMPATIBILITY.md b/COMPATIBILITY.md index 31a08be8d..46442b603 100644 --- a/COMPATIBILITY.md +++ b/COMPATIBILITY.md @@ -308,6 +308,7 @@ The port installs the skill with `npx skills add hardkoded/e2e-dotnet` and regis - Upstream's `device.unstable_raw` (`feat(mobile): device.unstable_raw exposes the test's agent-device session`, `c3c4238`, tester-army/e2e#1380) is `Device.UnstableRaw`. Upstream's `DeviceSession` interface (`device.ts`) is `RawDeviceSession`, a .NET-only name, because the port's `DeviceSession` is the surface (upstream's `AgentDeviceSurface`) and a public type of that name exists. Its `client` is the session's `IDeviceClient` (the port's stand-in for `AgentDeviceClient`), `selection` is `DeviceSelection`, and `platform` is `MobilePlatform`. `BoundDevice` (`surface.ts`) is internal, set by the boot from the `id` and `device` of the `devices.boot` answer (`DeviceSession.BoundDevice`, upstream's `boundDevice`), and cleared when the session closes. Reading `UnstableRaw`, or `Client` of a handle taken earlier, after the session closed is `INVALID_STATE`. The client is `DeviceCommands.Translate`, which refuses a method it does not list, so `DeviceMethods.AppsList` (`apps.list`, the daemon's `apps` command) is added for upstream's reference example; the other methods of upstream's client stay unsupported until code calls them. Tests: the three `it` of `describe('unstable_raw')` in `Engine/DeviceFixtureUnstableRawTests`; the first is `Partial_`, since the port's fixture has no map of declared operations for `expect(declared).not.toHaveProperty('unstable_raw')`. Not ported: `unstable_raw reaches the test session through agent-device` (`apps/mobile-benchmark/tests/device-fixture.e2e.ts`), which needs the mobile benchmark app, a React Native app the port lacks. - Upstream's known-viewport fix (`fix(mobile): keep the known viewport when the app reads smaller than the screen`, `7674a2c`, tester-army/e2e#1291) is ported. `DeviceSession.Project` learns a snapshot's viewport only when it is plainly the screen's: the first one the session reads, the one it knows, or that one rotated (`SameScreen`, upstream's `sameScreen`; two sizes within a point are one screen, since rects from the tree carry fractions). `ViewportForAsync` (upstream's `viewportFor`) checks a tree of any other size against the device's screenshot, in either orientation, before it replaces the known one: a screenshot reports the portrait size even in landscape, so it confirms a size but cannot pick the orientation. The size it checks against is `DeviceSupport.LogicalScreenSize` (upstream's `logicalScreenSize`): the screenshot's logical dimensions, else its pixel dimensions over its density. A simulator's screenshot reports those; a real device's reports neither, so there the tree's size is taken as it comes. A screenshot that fails during that check fails the observation, as upstream's does. `TraceScreen` and `MaskedScreenshotAsync` size the trace screen and the secure-field masks by the known viewport, not the tree's. On iOS, a swipe from a sheet's grabber can pull down Notification Center, and the app's tree behind it reported a third of the screen (130x281 on a 390x844 device), which misplaced every rect and tap after it. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Engine/ObservationTests`, 'keeps the known viewport for an app drawn smaller than the screen, and follows a rotation' (new) and 'measures the screen in the viewport the snapshot publishes, and in its tree when it publishes none' (split into two harnesses, as upstream's); in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'drags at the centre of a screen rotated since the last observation' (new), since the keyboard drag measures its start through `ViewportForAsync` too. - Upstream's input budget (`fix(mobile): bound each input by its budget`, `14b77d3`, tester-army/e2e#1073) is ported. An input (a tap or other node action, a point action, typing, a key, `back`, a keyboard dismissal, and the node `LongPressAsync`, `HoverAsync`, `DragToAsync`, and `SwipeAsync` the port has in place of upstream's `longPress`, `hover`, `dragTo`, and `swipe` actions) runs its device commands under `InputBudget` (`DeviceSupport.cs`, upstream's `support.ts`): with no budget left the input is not sent and fails with `OPERATION_TIMEOUT`, an input that does not answer in time fails with `ACTION_MAY_HAVE_COMMITTED`, one sent with almost no budget left still gets 250 ms to answer, a hold (a long press or timed swipe) at least as long as the budget left is refused with `NOT_ACTIONABLE`, and each command of a multi-command input (a key on an unfocused field) is sent on its own. Reads, launches, artifacts, and the `Device` fixture calls (including `device.dismissKeyboard`, which passes no budget, as upstream's `dismissKeyboard(signal)` does) are unbounded. Differences: upstream reads the budget per call, from `operation.timeoutMs`, which is the call's `timeout` or `actionTimeout`. The port's session calls carry no timeout, so every input gets `DeviceSession.ActionTimeout`, which `MobileStartInfo.ActionTimeout` sets (`EngineStartOptions.ActionTimeout` fills it), as the web engine's operations get the action timeout. `DeviceSession.InputBudget` is upstream's `inputBudget` and `DeviceSession.SendInputAsync` its `sendInput`. `InputBudget`'s `track` is an `Action` that registers the task (upstream's returns the promise it is given); see [In-flight commands](#in-flight-commands). The `OperationContext.signal` doc change in `docs/reference/engine.mdx` and `contract.ts` has no counterpart: the port has no `OperationContext`. Tests: in `Engine/ObservationTests`, 'ends %s, hung on %s, at its budget as possibly committed' (a theory of the same seven entries; the screen swipe is a `perform` on the root), 'sends nothing once a settling target spent the budget, and no keystroke once a focus tap used it up', and 'refuses a long press that holds longer than its timeout leaves, before sending it' (also on the node `LongPressAsync`); in `Support/SupportHelpersTests`, 'sends an input with what is left of its budget, and reports one cut off as possibly committed'. These run on real timers of a few hundred milliseconds, where upstream's fake timers do not wait. 'lets a locate outlive its budget' was already ported. +- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the label before it is the name, and the child is left out. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change. - Upstream's WebDriver element-type fix (`fix(mobile): read XCUIElementType class names as their XCTest types`, `3f9a7d8`, tester-army/e2e#1133) is ported. `SnapshotProjection.NormalizeKind` is upstream's `normalizeKind` in `nodes.ts`: it drops `^XCUIElementType(?=[A-Z])` (`XcuiElementTypePrefix`, a .NET `[GeneratedRegex]` for upstream's inline regex) before it maps the type, so the `XCUIElementTypeButton` that agent-device's WebDriver runtime reports on hosted Appium clouds reads as `Button`. `DeviceSession.KeyboardShowing` is upstream's `keyboardShowing` in `surface.ts`: it routes node types through `NormalizeKind`, so an `XCUIElementTypeKeyboard` counts as a shown keyboard. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Nodes/SnapshotProjectionTests`, 'reads the XCUIElementType class names a WebDriver session reports as the native XCTest types'; in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'reads an XCUIElementTypeKeyboard from a WebDriver session as a shown keyboard'. - Upstream's held-device skip (`mobile: skip booted devices another agent-device session holds`, `9e53c40`, tester-army/e2e#1379) is ported. `BootedDevices` (`DevicePool.cs`, upstream's class in `pool.ts`) skips a booted device whose inventory entry has a `claimedBy.session` that is not one of this target's own slot sessions, logs the skipped ones, and fails `prepare` with a non-retryable `ENGINE_FAILURE` when every booted device is held. `BootedDevice` is upstream's interface of that name, as a record. `LifecycleTests.Skips_a_booted_device_another_agent_device_session_holds_keeps_one_its_own_slot_session_holds_and_fails_when_every_booted_one_is_held` ports the `lifecycle` `it` of `engine.test.ts`, whole. - Not ported: the tool-pixels test (needs the full runner and a scripted model), and the `device.openApp` check of an unknown permission name. diff --git a/skills/e2e-dotnet/references/mobile.md b/skills/e2e-dotnet/references/mobile.md index ea64f8f9c..0b9d25e70 100644 --- a/skills/e2e-dotnet/references/mobile.md +++ b/skills/e2e-dotnet/references/mobile.md @@ -163,6 +163,7 @@ Limits of the preview: the tests attach to the running app, so `Device.InstallAp ## Known limits +- Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The node's `Placeholder` is that text, its name is the label before it, and the child is left out of the tree, so `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. - On iOS, a plain React Native `View` has no children in the tree, so `Filter(has: ...)` on it and queries scoped to it match nothing, while `Filter(hasNot: ...)` keeps every such row, including one you meant to exclude. Scope to the screen's `ScrollView` (`Device.Locator("role=scroll-area")`), or give the children test ids. - `Locator.PressAsync` takes `Enter`, `Space`, or one character, sent through the soft keyboard. A field that is not focused is tapped first, for `Enter` as for a character, so the keyboard is up. Other named keys and every modifier fail with `UNSUPPORTED_CAPABILITY`. - Android needs `adb` from the Android SDK Platform-Tools on `PATH`, locally and for the Limrun tunnel. diff --git a/src/E2E.Mobile/SnapshotProjection.cs b/src/E2E.Mobile/SnapshotProjection.cs index 4488a43b2..a992b3369 100644 --- a/src/E2E.Mobile/SnapshotProjection.cs +++ b/src/E2E.Mobile/SnapshotProjection.cs @@ -311,7 +311,10 @@ SemanticNode Build(int position, ProjectedNode? parent) } var identifier = NonEmpty(source.Identifier); - var label = NonEmpty(source.Label); + IReadOnlyList kids = children.TryGetValue(position, out var below) ? below : []; + var blank = editable && !secure && string.IsNullOrEmpty(value) && NonEmpty(source.Placeholder) is null; + var folded = !android && blank ? FoldedPlaceholder(source, kids, raw, children) : null; + var label = folded is null ? NonEmpty(source.Label) : folded.Label; var name = NonEmpty(source.ContentDescription) ?? label; var entry = new ProjectedNode { @@ -324,7 +327,7 @@ SemanticNode Build(int position, ProjectedNode? parent) States = reported, }; index.Add(entry); - var childNodes = children.TryGetValue(position, out var below) ? below.Select(child => Build(child, entry)).ToList() : []; + var childNodes = kids.Where(child => child != folded?.Child).Select(child => Build(child, entry)).ToList(); // A secure field's value is never observed; the tree carries that it is secure, not what it // holds. Android echoes a text view's label as its value, which would render every line twice, @@ -355,7 +358,7 @@ SemanticNode Build(int position, ProjectedNode? parent) Checked = checkedState == true, Expanded = expanded == true, }, - Placeholder = NonEmpty(source.Placeholder), + Placeholder = folded?.Placeholder ?? NonEmpty(source.Placeholder), // The accessibility identifier (iOS) or resource id (Android) is the node's test id. TestId = identifier, @@ -601,6 +604,42 @@ private static bool ShowsHint(RawNode source, bool android, bool editable) return !string.IsNullOrEmpty(placeholder) && source.Value?.Trim() == placeholder; } + /// + /// The placeholder an empty iOS field folds into its label, with the child that draws it. Compose + /// Multiplatform reports no placeholder: it draws one as a text inside the field and appends that text to + /// the field's label, so an empty "Name" field with the example "Ada Lovelace" reads + /// "Name, Ada Lovelace" over one StaticText child. Null unless the field's only child is that + /// text and the field has a label of its own. + /// + private static Folded? FoldedPlaceholder(RawNode source, IReadOnlyList kids, IReadOnlyList raw, Dictionary> children) + { + if (kids.Count != 1) + { + return null; + } + + var child = kids[0]; + var text = raw[child]; + if (children.ContainsKey(child)) + { + return null; + } + + var placeholder = NonEmpty(text.Label); + var label = source.Label; + if (KindOf(text) != "static-text" || placeholder is null || label is null) + { + return null; + } + + if (!label.EndsWith($", {placeholder}", StringComparison.Ordinal) || label.Length == placeholder.Length + 2) + { + return null; + } + + return new Folded(label[..^(placeholder.Length + 2)], placeholder, child); + } + /// The string itself, or null for an absent or empty one. private static string? NonEmpty(string? value) => string.IsNullOrEmpty(value) ? null : value; @@ -747,4 +786,6 @@ private sealed record SpelledStates(bool? Checked = null, bool? Expanded = null) /// What React Native spelled into an iOS accessibility value. private sealed record SpelledValue(string Role, SpelledStates States, string? Value); + + private sealed record Folded(string Label, string Placeholder, int Child); } diff --git a/tests/E2E.Mobile.Tests/Nodes/SnapshotProjectionTests.cs b/tests/E2E.Mobile.Tests/Nodes/SnapshotProjectionTests.cs index 5ba5800f0..d99917eab 100644 --- a/tests/E2E.Mobile.Tests/Nodes/SnapshotProjectionTests.cs +++ b/tests/E2E.Mobile.Tests/Nodes/SnapshotProjectionTests.cs @@ -446,6 +446,50 @@ public void Reads_the_role_and_state_React_Native_spells_into_an_iOS_accessibili Assert.Empty(android.Index[0].States); } + [Fact] + public void Reads_the_placeholder_an_empty_Compose_Multiplatform_field_folds_into_its_label() + { + // examples/with-kotlin-multiplatform on iOS 27 (agent-device 0.21.33's XCTest tree, Compose Multiplatform 1.12.1): + // the field has no placeholder attribute; it appends the text it draws to its label. + static ProjectedSnapshot Field(string label, IReadOnlyList children, string? type = null, bool editable = true, string? value = null) => Project( + [ + N("e1", index: 0, depth: 0, type: "Application", label: "with-kotlin-multiplatform"), + N("e2", index: 1, parentIndex: 0, depth: 1, type: type ?? "TextView", label: label, identifier: "name", editable: editable, value: value), + .. children.Select((child, offset) => child with { Ref = $"e{3 + offset}", Index = 2 + offset, ParentIndex = 1, Depth = 2 }), + ]); + + var hint = N(type: "StaticText", label: "Ada Lovelace"); + var empty = Field("Name, Ada Lovelace", [hint]); + Assert.Equal(["with-kotlin-multiplatform", "Name"], Names(empty)); + Assert.Equal("textbox", empty.Index[1].Node.Role); + Assert.Equal("Ada Lovelace", empty.Index[1].Node.Placeholder); + Assert.Empty(empty.Index[1].Node.Children); + + // Typed text drops the placeholder from the label and is the value. + var typed = Field("Name", [], value: "Grace"); + Assert.Equal("Name", typed.Index[1].Node.Name); + Assert.Equal("Grace", typed.Index[1].Node.Value); + + // Anything but one text that ends the label stays as reported. + foreach (var (label, children, type, editable, value) in new (string, RawNode[], string?, bool, string?)[] + { + ("Name, Ada Lovelace", [hint], null, true, "Ada"), + ("Name, Ada Lovelace", [hint], "Other", false, null), + ("Name, Ada Lovelace", [hint, N(type: "Button", label: "Clear")], null, true, null), + ("Name, Ada Lovelace", [N(type: "Image", label: "Ada Lovelace")], null, true, null), + ("Name, Grace Hopper", [hint], null, true, null), + + // A label equal to the text it holds may be the field's own name (a Material label drawn inside). + ("Ada Lovelace", [hint], null, true, null), + }) + { + var kept = Field(label, children, type, editable, value).Index[1].Node; + Assert.Equal(label, kept.Name); + Assert.Null(kept.Placeholder); + Assert.Equal(children.Length, kept.Children.Count); + } + } + [Fact] public void Keeps_the_value_of_an_iOS_field_whose_text_is_also_its_label_and_drops_only_the_Android_text_view_echo() { From de3bfa5f778a255cf01ebcb6497105a882385a0d Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 17:36:07 -0300 Subject: [PATCH 2/7] Link the follow-up issue for the fill and clear refusal Co-Authored-By: Claude Sonnet 5.5 --- COMPATIBILITY.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/COMPATIBILITY.md b/COMPATIBILITY.md index 46442b603..cb58d1506 100644 --- a/COMPATIBILITY.md +++ b/COMPATIBILITY.md @@ -308,7 +308,7 @@ The port installs the skill with `npx skills add hardkoded/e2e-dotnet` and regis - Upstream's `device.unstable_raw` (`feat(mobile): device.unstable_raw exposes the test's agent-device session`, `c3c4238`, tester-army/e2e#1380) is `Device.UnstableRaw`. Upstream's `DeviceSession` interface (`device.ts`) is `RawDeviceSession`, a .NET-only name, because the port's `DeviceSession` is the surface (upstream's `AgentDeviceSurface`) and a public type of that name exists. Its `client` is the session's `IDeviceClient` (the port's stand-in for `AgentDeviceClient`), `selection` is `DeviceSelection`, and `platform` is `MobilePlatform`. `BoundDevice` (`surface.ts`) is internal, set by the boot from the `id` and `device` of the `devices.boot` answer (`DeviceSession.BoundDevice`, upstream's `boundDevice`), and cleared when the session closes. Reading `UnstableRaw`, or `Client` of a handle taken earlier, after the session closed is `INVALID_STATE`. The client is `DeviceCommands.Translate`, which refuses a method it does not list, so `DeviceMethods.AppsList` (`apps.list`, the daemon's `apps` command) is added for upstream's reference example; the other methods of upstream's client stay unsupported until code calls them. Tests: the three `it` of `describe('unstable_raw')` in `Engine/DeviceFixtureUnstableRawTests`; the first is `Partial_`, since the port's fixture has no map of declared operations for `expect(declared).not.toHaveProperty('unstable_raw')`. Not ported: `unstable_raw reaches the test session through agent-device` (`apps/mobile-benchmark/tests/device-fixture.e2e.ts`), which needs the mobile benchmark app, a React Native app the port lacks. - Upstream's known-viewport fix (`fix(mobile): keep the known viewport when the app reads smaller than the screen`, `7674a2c`, tester-army/e2e#1291) is ported. `DeviceSession.Project` learns a snapshot's viewport only when it is plainly the screen's: the first one the session reads, the one it knows, or that one rotated (`SameScreen`, upstream's `sameScreen`; two sizes within a point are one screen, since rects from the tree carry fractions). `ViewportForAsync` (upstream's `viewportFor`) checks a tree of any other size against the device's screenshot, in either orientation, before it replaces the known one: a screenshot reports the portrait size even in landscape, so it confirms a size but cannot pick the orientation. The size it checks against is `DeviceSupport.LogicalScreenSize` (upstream's `logicalScreenSize`): the screenshot's logical dimensions, else its pixel dimensions over its density. A simulator's screenshot reports those; a real device's reports neither, so there the tree's size is taken as it comes. A screenshot that fails during that check fails the observation, as upstream's does. `TraceScreen` and `MaskedScreenshotAsync` size the trace screen and the secure-field masks by the known viewport, not the tree's. On iOS, a swipe from a sheet's grabber can pull down Notification Center, and the app's tree behind it reported a third of the screen (130x281 on a 390x844 device), which misplaced every rect and tap after it. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Engine/ObservationTests`, 'keeps the known viewport for an app drawn smaller than the screen, and follows a rotation' (new) and 'measures the screen in the viewport the snapshot publishes, and in its tree when it publishes none' (split into two harnesses, as upstream's); in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'drags at the centre of a screen rotated since the last observation' (new), since the keyboard drag measures its start through `ViewportForAsync` too. - Upstream's input budget (`fix(mobile): bound each input by its budget`, `14b77d3`, tester-army/e2e#1073) is ported. An input (a tap or other node action, a point action, typing, a key, `back`, a keyboard dismissal, and the node `LongPressAsync`, `HoverAsync`, `DragToAsync`, and `SwipeAsync` the port has in place of upstream's `longPress`, `hover`, `dragTo`, and `swipe` actions) runs its device commands under `InputBudget` (`DeviceSupport.cs`, upstream's `support.ts`): with no budget left the input is not sent and fails with `OPERATION_TIMEOUT`, an input that does not answer in time fails with `ACTION_MAY_HAVE_COMMITTED`, one sent with almost no budget left still gets 250 ms to answer, a hold (a long press or timed swipe) at least as long as the budget left is refused with `NOT_ACTIONABLE`, and each command of a multi-command input (a key on an unfocused field) is sent on its own. Reads, launches, artifacts, and the `Device` fixture calls (including `device.dismissKeyboard`, which passes no budget, as upstream's `dismissKeyboard(signal)` does) are unbounded. Differences: upstream reads the budget per call, from `operation.timeoutMs`, which is the call's `timeout` or `actionTimeout`. The port's session calls carry no timeout, so every input gets `DeviceSession.ActionTimeout`, which `MobileStartInfo.ActionTimeout` sets (`EngineStartOptions.ActionTimeout` fills it), as the web engine's operations get the action timeout. `DeviceSession.InputBudget` is upstream's `inputBudget` and `DeviceSession.SendInputAsync` its `sendInput`. `InputBudget`'s `track` is an `Action` that registers the task (upstream's returns the promise it is given); see [In-flight commands](#in-flight-commands). The `OperationContext.signal` doc change in `docs/reference/engine.mdx` and `contract.ts` has no counterpart: the port has no `OperationContext`. Tests: in `Engine/ObservationTests`, 'ends %s, hung on %s, at its budget as possibly committed' (a theory of the same seven entries; the screen swipe is a `perform` on the root), 'sends nothing once a settling target spent the budget, and no keystroke once a focus tap used it up', and 'refuses a long press that holds longer than its timeout leaves, before sending it' (also on the node `LongPressAsync`); in `Support/SupportHelpersTests`, 'sends an input with what is left of its budget, and reports one cut off as possibly committed'. These run on real timers of a few hundred milliseconds, where upstream's fake timers do not wait. 'lets a locate outlive its budget' was already ported. -- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the label before it is the name, and the child is left out. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change. +- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the label before it is the name, and the child is left out. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change; it is tracked in #850. - Upstream's WebDriver element-type fix (`fix(mobile): read XCUIElementType class names as their XCTest types`, `3f9a7d8`, tester-army/e2e#1133) is ported. `SnapshotProjection.NormalizeKind` is upstream's `normalizeKind` in `nodes.ts`: it drops `^XCUIElementType(?=[A-Z])` (`XcuiElementTypePrefix`, a .NET `[GeneratedRegex]` for upstream's inline regex) before it maps the type, so the `XCUIElementTypeButton` that agent-device's WebDriver runtime reports on hosted Appium clouds reads as `Button`. `DeviceSession.KeyboardShowing` is upstream's `keyboardShowing` in `surface.ts`: it routes node types through `NormalizeKind`, so an `XCUIElementTypeKeyboard` counts as a shown keyboard. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Nodes/SnapshotProjectionTests`, 'reads the XCUIElementType class names a WebDriver session reports as the native XCTest types'; in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'reads an XCUIElementTypeKeyboard from a WebDriver session as a shown keyboard'. - Upstream's held-device skip (`mobile: skip booted devices another agent-device session holds`, `9e53c40`, tester-army/e2e#1379) is ported. `BootedDevices` (`DevicePool.cs`, upstream's class in `pool.ts`) skips a booted device whose inventory entry has a `claimedBy.session` that is not one of this target's own slot sessions, logs the skipped ones, and fails `prepare` with a non-retryable `ENGINE_FAILURE` when every booted device is held. `BootedDevice` is upstream's interface of that name, as a record. `LifecycleTests.Skips_a_booted_device_another_agent_device_session_holds_keeps_one_its_own_slot_session_holds_and_fails_when_every_booted_one_is_held` ports the `lifecycle` `it` of `engine.test.ts`, whole. - Not ported: the tool-pixels test (needs the full runner and a scripted model), and the `device.openApp` check of an unknown permission name. From 7a9a1fc1b2ec2c38331e9afcd15d624f18efb7b3 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 17:49:12 -0300 Subject: [PATCH 3/7] Document that ContentDescription still names a folded field Co-Authored-By: Claude Sonnet 5.5 --- COMPATIBILITY.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/COMPATIBILITY.md b/COMPATIBILITY.md index cb58d1506..8c8843a5e 100644 --- a/COMPATIBILITY.md +++ b/COMPATIBILITY.md @@ -308,7 +308,7 @@ The port installs the skill with `npx skills add hardkoded/e2e-dotnet` and regis - Upstream's `device.unstable_raw` (`feat(mobile): device.unstable_raw exposes the test's agent-device session`, `c3c4238`, tester-army/e2e#1380) is `Device.UnstableRaw`. Upstream's `DeviceSession` interface (`device.ts`) is `RawDeviceSession`, a .NET-only name, because the port's `DeviceSession` is the surface (upstream's `AgentDeviceSurface`) and a public type of that name exists. Its `client` is the session's `IDeviceClient` (the port's stand-in for `AgentDeviceClient`), `selection` is `DeviceSelection`, and `platform` is `MobilePlatform`. `BoundDevice` (`surface.ts`) is internal, set by the boot from the `id` and `device` of the `devices.boot` answer (`DeviceSession.BoundDevice`, upstream's `boundDevice`), and cleared when the session closes. Reading `UnstableRaw`, or `Client` of a handle taken earlier, after the session closed is `INVALID_STATE`. The client is `DeviceCommands.Translate`, which refuses a method it does not list, so `DeviceMethods.AppsList` (`apps.list`, the daemon's `apps` command) is added for upstream's reference example; the other methods of upstream's client stay unsupported until code calls them. Tests: the three `it` of `describe('unstable_raw')` in `Engine/DeviceFixtureUnstableRawTests`; the first is `Partial_`, since the port's fixture has no map of declared operations for `expect(declared).not.toHaveProperty('unstable_raw')`. Not ported: `unstable_raw reaches the test session through agent-device` (`apps/mobile-benchmark/tests/device-fixture.e2e.ts`), which needs the mobile benchmark app, a React Native app the port lacks. - Upstream's known-viewport fix (`fix(mobile): keep the known viewport when the app reads smaller than the screen`, `7674a2c`, tester-army/e2e#1291) is ported. `DeviceSession.Project` learns a snapshot's viewport only when it is plainly the screen's: the first one the session reads, the one it knows, or that one rotated (`SameScreen`, upstream's `sameScreen`; two sizes within a point are one screen, since rects from the tree carry fractions). `ViewportForAsync` (upstream's `viewportFor`) checks a tree of any other size against the device's screenshot, in either orientation, before it replaces the known one: a screenshot reports the portrait size even in landscape, so it confirms a size but cannot pick the orientation. The size it checks against is `DeviceSupport.LogicalScreenSize` (upstream's `logicalScreenSize`): the screenshot's logical dimensions, else its pixel dimensions over its density. A simulator's screenshot reports those; a real device's reports neither, so there the tree's size is taken as it comes. A screenshot that fails during that check fails the observation, as upstream's does. `TraceScreen` and `MaskedScreenshotAsync` size the trace screen and the secure-field masks by the known viewport, not the tree's. On iOS, a swipe from a sheet's grabber can pull down Notification Center, and the app's tree behind it reported a third of the screen (130x281 on a 390x844 device), which misplaced every rect and tap after it. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Engine/ObservationTests`, 'keeps the known viewport for an app drawn smaller than the screen, and follows a rotation' (new) and 'measures the screen in the viewport the snapshot publishes, and in its tree when it publishes none' (split into two harnesses, as upstream's); in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'drags at the centre of a screen rotated since the last observation' (new), since the keyboard drag measures its start through `ViewportForAsync` too. - Upstream's input budget (`fix(mobile): bound each input by its budget`, `14b77d3`, tester-army/e2e#1073) is ported. An input (a tap or other node action, a point action, typing, a key, `back`, a keyboard dismissal, and the node `LongPressAsync`, `HoverAsync`, `DragToAsync`, and `SwipeAsync` the port has in place of upstream's `longPress`, `hover`, `dragTo`, and `swipe` actions) runs its device commands under `InputBudget` (`DeviceSupport.cs`, upstream's `support.ts`): with no budget left the input is not sent and fails with `OPERATION_TIMEOUT`, an input that does not answer in time fails with `ACTION_MAY_HAVE_COMMITTED`, one sent with almost no budget left still gets 250 ms to answer, a hold (a long press or timed swipe) at least as long as the budget left is refused with `NOT_ACTIONABLE`, and each command of a multi-command input (a key on an unfocused field) is sent on its own. Reads, launches, artifacts, and the `Device` fixture calls (including `device.dismissKeyboard`, which passes no budget, as upstream's `dismissKeyboard(signal)` does) are unbounded. Differences: upstream reads the budget per call, from `operation.timeoutMs`, which is the call's `timeout` or `actionTimeout`. The port's session calls carry no timeout, so every input gets `DeviceSession.ActionTimeout`, which `MobileStartInfo.ActionTimeout` sets (`EngineStartOptions.ActionTimeout` fills it), as the web engine's operations get the action timeout. `DeviceSession.InputBudget` is upstream's `inputBudget` and `DeviceSession.SendInputAsync` its `sendInput`. `InputBudget`'s `track` is an `Action` that registers the task (upstream's returns the promise it is given); see [In-flight commands](#in-flight-commands). The `OperationContext.signal` doc change in `docs/reference/engine.mdx` and `contract.ts` has no counterpart: the port has no `OperationContext`. Tests: in `Engine/ObservationTests`, 'ends %s, hung on %s, at its budget as possibly committed' (a theory of the same seven entries; the screen swipe is a `perform` on the root), 'sends nothing once a settling target spent the budget, and no keystroke once a focus tap used it up', and 'refuses a long press that holds longer than its timeout leaves, before sending it' (also on the node `LongPressAsync`); in `Support/SupportHelpersTests`, 'sends an input with what is left of its budget, and reports one cut off as possibly committed'. These run on real timers of a few hundred milliseconds, where upstream's fake timers do not wait. 'lets a locate outlive its budget' was already ported. -- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the label before it is the name, and the child is left out. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change; it is tracked in #850. +- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the child is left out, and the label before it is the name unless the node has a non-empty `ContentDescription`, which stays the name (as for any node) and leaves the label as `Text`. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change; it is tracked in #850. - Upstream's WebDriver element-type fix (`fix(mobile): read XCUIElementType class names as their XCTest types`, `3f9a7d8`, tester-army/e2e#1133) is ported. `SnapshotProjection.NormalizeKind` is upstream's `normalizeKind` in `nodes.ts`: it drops `^XCUIElementType(?=[A-Z])` (`XcuiElementTypePrefix`, a .NET `[GeneratedRegex]` for upstream's inline regex) before it maps the type, so the `XCUIElementTypeButton` that agent-device's WebDriver runtime reports on hosted Appium clouds reads as `Button`. `DeviceSession.KeyboardShowing` is upstream's `keyboardShowing` in `surface.ts`: it routes node types through `NormalizeKind`, so an `XCUIElementTypeKeyboard` counts as a shown keyboard. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Nodes/SnapshotProjectionTests`, 'reads the XCUIElementType class names a WebDriver session reports as the native XCTest types'; in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'reads an XCUIElementTypeKeyboard from a WebDriver session as a shown keyboard'. - Upstream's held-device skip (`mobile: skip booted devices another agent-device session holds`, `9e53c40`, tester-army/e2e#1379) is ported. `BootedDevices` (`DevicePool.cs`, upstream's class in `pool.ts`) skips a booted device whose inventory entry has a `claimedBy.session` that is not one of this target's own slot sessions, logs the skipped ones, and fails `prepare` with a non-retryable `ENGINE_FAILURE` when every booted device is held. `BootedDevice` is upstream's interface of that name, as a record. `LifecycleTests.Skips_a_booted_device_another_agent_device_session_holds_keeps_one_its_own_slot_session_holds_and_fails_when_every_booted_one_is_held` ports the `lifecycle` `it` of `engine.test.ts`, whole. - Not ported: the tool-pixels test (needs the full runner and a scripted model), and the `device.openApp` check of an unknown permission name. From 0e5ea8ad31d0dca8846c48ce161c249dccc56780 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 18:05:05 -0300 Subject: [PATCH 4/7] Document the Compose Multiplatform placeholder in SKILL.md and the mobile guide Co-Authored-By: Claude Sonnet 5.5 --- docs-site/src/content/docs/get-started/mobile.mdx | 2 +- skills/e2e-dotnet/SKILL.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/docs-site/src/content/docs/get-started/mobile.mdx b/docs-site/src/content/docs/get-started/mobile.mdx index 6effa1a92..4e1dd3c04 100644 --- a/docs-site/src/content/docs/get-started/mobile.mdx +++ b/docs-site/src/content/docs/get-started/mobile.mdx @@ -82,7 +82,7 @@ public sealed class GreetingTests : MobileAppTest } ``` -`GetByTestId` finds a control by its accessibility identifier. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. +`GetByTestId` finds a control by its accessibility identifier. A field's hint is its `Placeholder`, shown or not, so `GetByPlaceholder` finds it. Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The projection reads that text as the field's `Placeholder` and leaves the child out of the tree, so `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. ```csharp await Device.HomeAsync(); diff --git a/skills/e2e-dotnet/SKILL.md b/skills/e2e-dotnet/SKILL.md index d16f17516..37fb42d86 100644 --- a/skills/e2e-dotnet/SKILL.md +++ b/skills/e2e-dotnet/SKILL.md @@ -35,7 +35,7 @@ trip. The installed tool prints the same text with `e2e guide ` | Running tests: `dotnet test` filters, the `e2e` tool, `report.json` for scripts, exit codes, CI, test filters (not wired into `dotnet test`: the `ITestFilter` seam and the `run.changes` and `run.filters` report fields) | `running` | [references/running.md](references/running.md) | | Handling a failed run: the trace page, `report.json` for scripts, `E2ESession.Report()` for a session you run yourself, error codes and their fixes, a visible browser, replay misses, phase timings, the agent step table, every model call | `debugging` | [references/debugging.md](references/debugging.md) | | Exploring an app toward a goal without a test file, with `e2e explore` or `Explore.ExploreAsync` | `explore` | [references/explore.md](references/explore.md) | -| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app | `mobile` | [references/mobile.md](references/mobile.md) | +| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (its drawn example text is the field's `Placeholder`, not part of its label or a child text) | `mobile` | [references/mobile.md](references/mobile.md) | | Testing a Windows desktop app: `WinAppEngine`, `WinAppTest`, the `Window` fixture, the `windows` target in `e2e explore` and `e2e mcp`, the interactive-desktop requirement, known limits | `windows` | [references/windows.md](references/windows.md) | | Asked to bug bash or hunt for bugs: proving each bug with a repro test | `bug-bash` | [references/bug-bash.md](references/bug-bash.md) | From fc5bb089d46dda72b7cbc0a20a2062f35e86e814 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 18:11:05 -0300 Subject: [PATCH 5/7] Scope the Compose Multiplatform placeholder claim in the guide and SKILL.md Co-Authored-By: Claude Sonnet 5.5 --- docs-site/src/content/docs/get-started/mobile.mdx | 2 +- skills/e2e-dotnet/SKILL.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/docs-site/src/content/docs/get-started/mobile.mdx b/docs-site/src/content/docs/get-started/mobile.mdx index 4e1dd3c04..ef72852c8 100644 --- a/docs-site/src/content/docs/get-started/mobile.mdx +++ b/docs-site/src/content/docs/get-started/mobile.mdx @@ -82,7 +82,7 @@ public sealed class GreetingTests : MobileAppTest } ``` -`GetByTestId` finds a control by its accessibility identifier. A field's hint is its `Placeholder`, shown or not, so `GetByPlaceholder` finds it. Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The projection reads that text as the field's `Placeholder` and leaves the child out of the tree, so `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. +`GetByTestId` finds a control by its accessibility identifier. A field's hint is its `Placeholder`, shown or not, so `GetByPlaceholder` finds it. Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The projection reads that text as the field's `Placeholder` and leaves the child out of the tree only for an empty, editable, non-secure iOS field that reports no placeholder of its own, whose single child is a `StaticText` with no children and whose text ends the label after `, ` with a name before it. A field of any other shape keeps its label and children as reported. For a field of that shape, `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. ```csharp await Device.HomeAsync(); diff --git a/skills/e2e-dotnet/SKILL.md b/skills/e2e-dotnet/SKILL.md index 37fb42d86..79f548564 100644 --- a/skills/e2e-dotnet/SKILL.md +++ b/skills/e2e-dotnet/SKILL.md @@ -35,7 +35,7 @@ trip. The installed tool prints the same text with `e2e guide ` | Running tests: `dotnet test` filters, the `e2e` tool, `report.json` for scripts, exit codes, CI, test filters (not wired into `dotnet test`: the `ITestFilter` seam and the `run.changes` and `run.filters` report fields) | `running` | [references/running.md](references/running.md) | | Handling a failed run: the trace page, `report.json` for scripts, `E2ESession.Report()` for a session you run yourself, error codes and their fixes, a visible browser, replay misses, phase timings, the agent step table, every model call | `debugging` | [references/debugging.md](references/debugging.md) | | Exploring an app toward a goal without a test file, with `e2e explore` or `Explore.ExploreAsync` | `explore` | [references/explore.md](references/explore.md) | -| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (its drawn example text is the field's `Placeholder`, not part of its label or a child text) | `mobile` | [references/mobile.md](references/mobile.md) | +| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (only an empty, editable, non-secure field with no placeholder of its own, whose single childless `StaticText` ends its label after `, `, has that drawn example text as its `Placeholder` instead of part of its label or a child text) | `mobile` | [references/mobile.md](references/mobile.md) | | Testing a Windows desktop app: `WinAppEngine`, `WinAppTest`, the `Window` fixture, the `windows` target in `e2e explore` and `e2e mcp`, the interactive-desktop requirement, known limits | `windows` | [references/windows.md](references/windows.md) | | Asked to bug bash or hunt for bugs: proving each bug with a repro test | `bug-bash` | [references/bug-bash.md](references/bug-bash.md) | From 093b565d37e29d61c1292ac1b25154ece964850b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Dar=C3=ADo=20Kondratiuk?= Date: Fri, 30 Oct 2026 18:20:21 -0300 Subject: [PATCH 6/7] State the same guarded Compose Multiplatform placeholder conditions everywhere Co-Authored-By: Claude Sonnet 5.5 --- COMPATIBILITY.md | 2 +- docs-site/src/content/docs/get-started/mobile.mdx | 2 +- skills/e2e-dotnet/SKILL.md | 2 +- skills/e2e-dotnet/references/mobile.md | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-) diff --git a/COMPATIBILITY.md b/COMPATIBILITY.md index 8c8843a5e..fae88611a 100644 --- a/COMPATIBILITY.md +++ b/COMPATIBILITY.md @@ -308,7 +308,7 @@ The port installs the skill with `npx skills add hardkoded/e2e-dotnet` and regis - Upstream's `device.unstable_raw` (`feat(mobile): device.unstable_raw exposes the test's agent-device session`, `c3c4238`, tester-army/e2e#1380) is `Device.UnstableRaw`. Upstream's `DeviceSession` interface (`device.ts`) is `RawDeviceSession`, a .NET-only name, because the port's `DeviceSession` is the surface (upstream's `AgentDeviceSurface`) and a public type of that name exists. Its `client` is the session's `IDeviceClient` (the port's stand-in for `AgentDeviceClient`), `selection` is `DeviceSelection`, and `platform` is `MobilePlatform`. `BoundDevice` (`surface.ts`) is internal, set by the boot from the `id` and `device` of the `devices.boot` answer (`DeviceSession.BoundDevice`, upstream's `boundDevice`), and cleared when the session closes. Reading `UnstableRaw`, or `Client` of a handle taken earlier, after the session closed is `INVALID_STATE`. The client is `DeviceCommands.Translate`, which refuses a method it does not list, so `DeviceMethods.AppsList` (`apps.list`, the daemon's `apps` command) is added for upstream's reference example; the other methods of upstream's client stay unsupported until code calls them. Tests: the three `it` of `describe('unstable_raw')` in `Engine/DeviceFixtureUnstableRawTests`; the first is `Partial_`, since the port's fixture has no map of declared operations for `expect(declared).not.toHaveProperty('unstable_raw')`. Not ported: `unstable_raw reaches the test session through agent-device` (`apps/mobile-benchmark/tests/device-fixture.e2e.ts`), which needs the mobile benchmark app, a React Native app the port lacks. - Upstream's known-viewport fix (`fix(mobile): keep the known viewport when the app reads smaller than the screen`, `7674a2c`, tester-army/e2e#1291) is ported. `DeviceSession.Project` learns a snapshot's viewport only when it is plainly the screen's: the first one the session reads, the one it knows, or that one rotated (`SameScreen`, upstream's `sameScreen`; two sizes within a point are one screen, since rects from the tree carry fractions). `ViewportForAsync` (upstream's `viewportFor`) checks a tree of any other size against the device's screenshot, in either orientation, before it replaces the known one: a screenshot reports the portrait size even in landscape, so it confirms a size but cannot pick the orientation. The size it checks against is `DeviceSupport.LogicalScreenSize` (upstream's `logicalScreenSize`): the screenshot's logical dimensions, else its pixel dimensions over its density. A simulator's screenshot reports those; a real device's reports neither, so there the tree's size is taken as it comes. A screenshot that fails during that check fails the observation, as upstream's does. `TraceScreen` and `MaskedScreenshotAsync` size the trace screen and the secure-field masks by the known viewport, not the tree's. On iOS, a swipe from a sheet's grabber can pull down Notification Center, and the app's tree behind it reported a third of the screen (130x281 on a 390x844 device), which misplaced every rect and tap after it. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Engine/ObservationTests`, 'keeps the known viewport for an app drawn smaller than the screen, and follows a rotation' (new) and 'measures the screen in the viewport the snapshot publishes, and in its tree when it publishes none' (split into two harnesses, as upstream's); in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'drags at the centre of a screen rotated since the last observation' (new), since the keyboard drag measures its start through `ViewportForAsync` too. - Upstream's input budget (`fix(mobile): bound each input by its budget`, `14b77d3`, tester-army/e2e#1073) is ported. An input (a tap or other node action, a point action, typing, a key, `back`, a keyboard dismissal, and the node `LongPressAsync`, `HoverAsync`, `DragToAsync`, and `SwipeAsync` the port has in place of upstream's `longPress`, `hover`, `dragTo`, and `swipe` actions) runs its device commands under `InputBudget` (`DeviceSupport.cs`, upstream's `support.ts`): with no budget left the input is not sent and fails with `OPERATION_TIMEOUT`, an input that does not answer in time fails with `ACTION_MAY_HAVE_COMMITTED`, one sent with almost no budget left still gets 250 ms to answer, a hold (a long press or timed swipe) at least as long as the budget left is refused with `NOT_ACTIONABLE`, and each command of a multi-command input (a key on an unfocused field) is sent on its own. Reads, launches, artifacts, and the `Device` fixture calls (including `device.dismissKeyboard`, which passes no budget, as upstream's `dismissKeyboard(signal)` does) are unbounded. Differences: upstream reads the budget per call, from `operation.timeoutMs`, which is the call's `timeout` or `actionTimeout`. The port's session calls carry no timeout, so every input gets `DeviceSession.ActionTimeout`, which `MobileStartInfo.ActionTimeout` sets (`EngineStartOptions.ActionTimeout` fills it), as the web engine's operations get the action timeout. `DeviceSession.InputBudget` is upstream's `inputBudget` and `DeviceSession.SendInputAsync` its `sendInput`. `InputBudget`'s `track` is an `Action` that registers the task (upstream's returns the promise it is given); see [In-flight commands](#in-flight-commands). The `OperationContext.signal` doc change in `docs/reference/engine.mdx` and `contract.ts` has no counterpart: the port has no `OperationContext`. Tests: in `Engine/ObservationTests`, 'ends %s, hung on %s, at its budget as possibly committed' (a theory of the same seven entries; the screen swipe is a `perform` on the root), 'sends nothing once a settling target spent the budget, and no keystroke once a focus tap used it up', and 'refuses a long press that holds longer than its timeout leaves, before sending it' (also on the node `LongPressAsync`); in `Support/SupportHelpersTests`, 'sends an input with what is left of its budget, and reports one cut off as possibly committed'. These run on real timers of a few hundred milliseconds, where upstream's fake timers do not wait. 'lets a locate outlive its budget' was already ported. -- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): when an editable, non-secure iOS field has no value and no placeholder of its own, and its only child is a childless `StaticText` whose text ends the field's label after `, `, that text is the field's `Placeholder`, the child is left out, and the label before it is the name unless the node has a non-empty `ContentDescription`, which stays the name (as for any node) and leaves the label as `Text`. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change; it is tracked in #850. +- Upstream's Compose Multiplatform text fields on iOS (`mobile reads Compose Multiplatform text fields on iOS`, `75ae2ab`, tester-army/e2e#1427) are ported in part. `SnapshotProjection.ProjectSnapshot` reads the placeholder an empty iOS field folds into its label (upstream's `foldedPlaceholder` in `nodes.ts`, `FoldedPlaceholder` here): for an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text, that text is the field's `Placeholder`, the child is left out of the tree, and the name is the label before `, ` (a non-empty `ContentDescription` stays the name, and that label is then the `Text`). A field of any other shape keeps its label and children as reported. The private record `Folded` is a .NET-only name for upstream's inline `{ label, placeholder, child }` object. Test: 'reads the placeholder an empty Compose Multiplatform field folds into its label' in `Nodes/SnapshotProjectionTests`. Not ported: `mayTakeText` and `NodeBinding.takesText` (`nodes.ts`, `surface.ts`), with the three `it`s that use them in `describe('input freshness')` of `engine.test.ts` ('acts on an unclassified iOS field', 'rejects an unclassified iOS element', 'rejects an unclassified iOS element that gained a control before typing'). The port's `fill` and `clear` have no check of the node's role and send the text to any node with a ref, so an iOS `other` is already typed into; the check upstream widens does not exist here. Adding it would be a new refusal, not this change; it is tracked in #850. - Upstream's WebDriver element-type fix (`fix(mobile): read XCUIElementType class names as their XCTest types`, `3f9a7d8`, tester-army/e2e#1133) is ported. `SnapshotProjection.NormalizeKind` is upstream's `normalizeKind` in `nodes.ts`: it drops `^XCUIElementType(?=[A-Z])` (`XcuiElementTypePrefix`, a .NET `[GeneratedRegex]` for upstream's inline regex) before it maps the type, so the `XCUIElementTypeButton` that agent-device's WebDriver runtime reports on hosted Appium clouds reads as `Button`. `DeviceSession.KeyboardShowing` is upstream's `keyboardShowing` in `surface.ts`: it routes node types through `NormalizeKind`, so an `XCUIElementTypeKeyboard` counts as a shown keyboard. Upstream records this only in its changeset; its skill references and docs say nothing about it, so `SKILL.md` and `docs/` have nothing to mirror. Tests: in `Nodes/SnapshotProjectionTests`, 'reads the XCUIElementType class names a WebDriver session reports as the native XCTest types'; in `Engine/DeviceFixtureDismissingAKeyboardThatHasNoDismissKeyTests`, 'reads an XCUIElementTypeKeyboard from a WebDriver session as a shown keyboard'. - Upstream's held-device skip (`mobile: skip booted devices another agent-device session holds`, `9e53c40`, tester-army/e2e#1379) is ported. `BootedDevices` (`DevicePool.cs`, upstream's class in `pool.ts`) skips a booted device whose inventory entry has a `claimedBy.session` that is not one of this target's own slot sessions, logs the skipped ones, and fails `prepare` with a non-retryable `ENGINE_FAILURE` when every booted device is held. `BootedDevice` is upstream's interface of that name, as a record. `LifecycleTests.Skips_a_booted_device_another_agent_device_session_holds_keeps_one_its_own_slot_session_holds_and_fails_when_every_booted_one_is_held` ports the `lifecycle` `it` of `engine.test.ts`, whole. - Not ported: the tool-pixels test (needs the full runner and a scripted model), and the `device.openApp` check of an unknown permission name. diff --git a/docs-site/src/content/docs/get-started/mobile.mdx b/docs-site/src/content/docs/get-started/mobile.mdx index ef72852c8..eea2c097c 100644 --- a/docs-site/src/content/docs/get-started/mobile.mdx +++ b/docs-site/src/content/docs/get-started/mobile.mdx @@ -82,7 +82,7 @@ public sealed class GreetingTests : MobileAppTest } ``` -`GetByTestId` finds a control by its accessibility identifier. A field's hint is its `Placeholder`, shown or not, so `GetByPlaceholder` finds it. Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The projection reads that text as the field's `Placeholder` and leaves the child out of the tree only for an empty, editable, non-secure iOS field that reports no placeholder of its own, whose single child is a `StaticText` with no children and whose text ends the label after `, ` with a name before it. A field of any other shape keeps its label and children as reported. For a field of that shape, `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. +`GetByTestId` finds a control by its accessibility identifier. A field's hint is its `Placeholder`, shown or not, so `GetByPlaceholder` finds it. Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. For an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text (`Name, Ada Lovelace`), that text is the field's `Placeholder`, the child is left out of the tree, and the name is the label before `, ` (a non-empty `ContentDescription` stays the name, and that label is then the `Text`). A field of any other shape keeps its label and children as reported. So `GetByPlaceholder("Ada Lovelace")` finds such a field and `GetByText("Ada Lovelace")` does not. `Device` is the device fixture: `HomeAsync`, `OpenAppAsync`, `ForegroundAppAsync`, and `SetAppearanceAsync(Appearance.Dark)`. ```csharp await Device.HomeAsync(); diff --git a/skills/e2e-dotnet/SKILL.md b/skills/e2e-dotnet/SKILL.md index 79f548564..190f4ea5c 100644 --- a/skills/e2e-dotnet/SKILL.md +++ b/skills/e2e-dotnet/SKILL.md @@ -35,7 +35,7 @@ trip. The installed tool prints the same text with `e2e guide ` | Running tests: `dotnet test` filters, the `e2e` tool, `report.json` for scripts, exit codes, CI, test filters (not wired into `dotnet test`: the `ITestFilter` seam and the `run.changes` and `run.filters` report fields) | `running` | [references/running.md](references/running.md) | | Handling a failed run: the trace page, `report.json` for scripts, `E2ESession.Report()` for a session you run yourself, error codes and their fixes, a visible browser, replay misses, phase timings, the agent step table, every model call | `debugging` | [references/debugging.md](references/debugging.md) | | Exploring an app toward a goal without a test file, with `e2e explore` or `Explore.ExploreAsync` | `explore` | [references/explore.md](references/explore.md) | -| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (only an empty, editable, non-secure field with no placeholder of its own, whose single childless `StaticText` ends its label after `, `, has that drawn example text as its `Placeholder` instead of part of its label or a child text) | `mobile` | [references/mobile.md](references/mobile.md) | +| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (for an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text, that text is the field's `Placeholder`, the child is left out, and the name is the label before `, ` unless `ContentDescription` is non-empty; any other field keeps its label and children) | `mobile` | [references/mobile.md](references/mobile.md) | | Testing a Windows desktop app: `WinAppEngine`, `WinAppTest`, the `Window` fixture, the `windows` target in `e2e explore` and `e2e mcp`, the interactive-desktop requirement, known limits | `windows` | [references/windows.md](references/windows.md) | | Asked to bug bash or hunt for bugs: proving each bug with a repro test | `bug-bash` | [references/bug-bash.md](references/bug-bash.md) | diff --git a/skills/e2e-dotnet/references/mobile.md b/skills/e2e-dotnet/references/mobile.md index 0b9d25e70..b9f374b4e 100644 --- a/skills/e2e-dotnet/references/mobile.md +++ b/skills/e2e-dotnet/references/mobile.md @@ -163,7 +163,7 @@ Limits of the preview: the tests attach to the running app, so `Device.InstallAp ## Known limits -- Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. The node's `Placeholder` is that text, its name is the label before it, and the child is left out of the tree, so `GetByPlaceholder("Ada Lovelace")` finds the field and `GetByText("Ada Lovelace")` does not. +- Compose Multiplatform reports no placeholder on iOS: an empty field appends the text it draws to its label (`Name, Ada Lovelace`) over one `StaticText` child. For an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text (`Name, Ada Lovelace`), that text is the field's `Placeholder`, the child is left out of the tree, and the name is the label before `, ` (a non-empty `ContentDescription` stays the name, and that label is then the `Text`). A field of any other shape keeps its label and children as reported. So `GetByPlaceholder("Ada Lovelace")` finds such a field and `GetByText("Ada Lovelace")` does not. - On iOS, a plain React Native `View` has no children in the tree, so `Filter(has: ...)` on it and queries scoped to it match nothing, while `Filter(hasNot: ...)` keeps every such row, including one you meant to exclude. Scope to the screen's `ScrollView` (`Device.Locator("role=scroll-area")`), or give the children test ids. - `Locator.PressAsync` takes `Enter`, `Space`, or one character, sent through the soft keyboard. A field that is not focused is tapped first, for `Enter` as for a character, so the keyboard is up. Other named keys and every modifier fail with `UNSUPPORTED_CAPABILITY`. - Android needs `adb` from the Android SDK Platform-Tools on `PATH`, locally and for the Limrun tunnel. From 1f1a288e1ecd9db9ad174a4be32ea6dfda0e70be Mon Sep 17 00:00:00 2001 From: Dario Kondratiuk Date: Sat, 31 Oct 2026 07:05:58 -0300 Subject: [PATCH 7/7] SKILL: state that a non-empty ContentDescription stays the name and the folded label becomes Text Co-Authored-By: Claude Sonnet 5.5 Claude-Session: https://claude.ai/code/session_01Ls2e21UaPC6a4UrcK62Jy7 --- skills/e2e-dotnet/SKILL.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/skills/e2e-dotnet/SKILL.md b/skills/e2e-dotnet/SKILL.md index 190f4ea5c..3e8b12eb7 100644 --- a/skills/e2e-dotnet/SKILL.md +++ b/skills/e2e-dotnet/SKILL.md @@ -35,7 +35,7 @@ trip. The installed tool prints the same text with `e2e guide ` | Running tests: `dotnet test` filters, the `e2e` tool, `report.json` for scripts, exit codes, CI, test filters (not wired into `dotnet test`: the `ITestFilter` seam and the `run.changes` and `run.filters` report fields) | `running` | [references/running.md](references/running.md) | | Handling a failed run: the trace page, `report.json` for scripts, `E2ESession.Report()` for a session you run yourself, error codes and their fixes, a visible browser, replay misses, phase timings, the agent step table, every model call | `debugging` | [references/debugging.md](references/debugging.md) | | Exploring an app toward a goal without a test file, with `e2e explore` or `Explore.ExploreAsync` | `explore` | [references/explore.md](references/explore.md) | -| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (for an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text, that text is the field's `Placeholder`, the child is left out, and the name is the label before `, ` unless `ContentDescription` is non-empty; any other field keeps its label and children) | `mobile` | [references/mobile.md](references/mobile.md) | +| Testing an iOS or Android app: `MobileEngine`, the `Device` fixture, picking a simulator (without `MobileOptions.Device`, `MobileEngine` skips booted devices another agent-device session holds, and `prepare` fails with `ENGINE_FAILURE` when every booted one is held), video of the device screen, `Device.UnstableRaw` (the test's agent-device session: name, device id and name, selection, and a raw `IDeviceClient` whose calls are not steps, replayed, or redacted; unstable; `INVALID_STATE` after the session is disposed), Limrun hosted devices, a MAUI app through the DevFlow agent, known limits, `Locator.PressAsync("Enter")` tapping an unfocused field first, and the agent's `open_app` returning to the app under test when it names no app, a Compose Multiplatform field on iOS (for an empty (no value), editable, non-secure iOS field that reports no placeholder of its own, whose only child is a `StaticText` with no children, and whose label is a non-empty name, then `, `, then that child's text, that text is the field's `Placeholder`, the child is left out, and the name is the label before `, `; a non-empty `ContentDescription` stays the name, and that label is then the `Text`; any other field keeps its label and children) | `mobile` | [references/mobile.md](references/mobile.md) | | Testing a Windows desktop app: `WinAppEngine`, `WinAppTest`, the `Window` fixture, the `windows` target in `e2e explore` and `e2e mcp`, the interactive-desktop requirement, known limits | `windows` | [references/windows.md](references/windows.md) | | Asked to bug bash or hunt for bugs: proving each bug with a repro test | `bug-bash` | [references/bug-bash.md](references/bug-bash.md) |