compose-ui-testing-patterns — Claude Code skill
Guides Jetpack Compose UI testing for semantics, screenshots, keyboard input, focus, and interaction state.
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a claude-codeWhat compose-ui-testing-patterns does
This skill picks the smallest test shape that proves the requested Compose behavior. It favours plain UI composables with controlled state for text, conditional content, and callbacks, while keeping navigation, lifecycle, and DI checks in integration tests.
For semantic behaviour, it uses onNodeWithText, enabled-state assertions, and absence checks. Test tags are reserved for nodes without stable visible text or with repeated labels. Building a ViewModel or dependency graph for a layout check is a poor fit here.
The guidance also covers keyboard input, focus, MutableInteractionSource, and screenshot tests. Rather than Thread.sleep, it calls for Compose synchronization through waitForIdle, runOnIdle, or bounded waitUntil.
What it helps with
- assert text and button state in Compose
- verify a callback after a click
- test keyboard-driven focus navigation
- record a visual contract with a screenshot test
- check hover, pressed, or focused state
Example prompts for the agent
- Write a Compose UI test for
ProfileScreenusing controlledProfileUiStateand assertonSaveClick. - Review this Compose test: does the state assertion need
runOnIdle? - Create a screenshot test for fixed state and describe how to verify the baseline artifact.
How to install compose-ui-testing-patterns
1Skills CLI (Vercel)
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a claude-code2Manually: put the files in place
~/.claude/skills/compose-ui-testing-patterns/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/compose-ui-testing-patterns/
1Skills CLI (Vercel)
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a codex2Manually: put the files in place
.agents/skills/compose-ui-testing-patterns/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a cursor2Manually: put the files in place
.cursor/skills/compose-ui-testing-patterns/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install chrisbanes/skills compose-ui-testing-patterns2Manually: put the files in place
.github/skills/compose-ui-testing-patterns/SKILL.md1Skills CLI (Vercel)
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a gemini-cli2Platform CLI
gemini skills install https://github.com/chrisbanes/skills --path skills/compose-ui-testing-patternsManually: put the files in place
.devin/skills/compose-ui-testing-patterns/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about compose-ui-testing-patterns
How do I install compose-ui-testing-patterns in Claude Code?
Run npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a claude-code, then run npx skills add chrisbanes/skills --skill compose-ui-testing-patterns -a claude-code again. The skill is located at ~/.claude/skills/compose-ui-testing-patterns/SKILL.md.
What does compose-ui-testing-patterns do?
It guides Jetpack Compose test work across semantic assertions, callbacks, keyboard input, focus, interaction state, and screenshot evidence. Its main rule is to test the smallest UI contract that proves the behaviour.
When should I use an integration test instead of a plain Compose UI test?
Use an integration test when the behaviour depends on navigation, lifecycle, DI, or platform wiring. For layout, state branches, and callback wiring, the skill favours testing the plain UI composable.
Source text
SKILL.md · Apache-2.0
---
name: compose-ui-testing-patterns
description: Use when writing or reviewing Jetpack Compose UI tests, screenshot tests or baseline-recording evidence, previews, semantics assertions, fake image loading, keyboard input, focus assertions, interaction state (hover/pressed/focused), or tests for plain state-driven UI composables.
---
# Compose: UI testing patterns
## Core principle
Test the smallest UI contract that proves the behavior. Prefer plain state-driven UI tests with callbacks. Add integration only when lifecycle, navigation, DI, or platform behavior is the thing under test.
## Procedure
1. State the behavior and test concern the task asks you to prove.
2. Inspect the existing test against that concern and choose the smallest
sufficient seam from the table below.
3. Keep focused edits within the requested test concern. Do not move test-only
helpers into production or broaden production APIs unless that production
boundary is itself under test.
4. Drive controlled state or input, synchronize through Compose when needed,
and assert an observable semantic, visual, or callback result.
5. Finish with no edit when the existing test already uses the narrowest valid
seam and proves the requested behavior.
## Test target choice
| What you need to prove | Test shape |
|---|---|
| Text, button, loading/error branch, conditional content | Plain UI Compose test |
| Callback wiring from click/input | Plain UI Compose test |
| Focus navigation or keyboard behavior | Compose test with key input |
| Visual layout, clipping, elevation, typography, image composition | Screenshot test |
| State holder updates UI correctly | State holder/unit test plus one wiring smoke test |
| Hover, pressed, focused, dragged interaction state | Plain UI test with MutableInteractionSource |
| Navigation, lifecycle, DI integration | Integration test |
## Prefer plain UI tests
If the screen has a state holder/UI split, test the plain UI composable:
```kotlin
composeTestRule.setContent {
ProfileScreen(
state = ProfileUiState(name = "Ada", canSave = true),
onNameChange = {},
onSaveClick = { saved = true },
onBackClick = {},
)
}
composeTestRule.onNodeWithText("Ada").assertIsDisplayed()
composeTestRule.onNodeWithText("Save").performClick()
assertThat(saved).isTrue()
```
This avoids constructing ViewModels, components, repositories, navigation, and dependency graphs for layout behavior.
## Semantics first
Assert semantics when behavior is semantic:
- Text exists: `onNodeWithText`.
- Button is enabled/disabled: `assertIsEnabled`, `assertIsNotEnabled`.
- Content is selected/focused/toggled: use semantics assertions.
- Content is absent: `assertDoesNotExist`.
Use test tags for nodes that have no stable user-visible text or where multiple nodes share text. Do not use tags as the first choice for all assertions; user-visible semantics are usually stronger.
## Callback testing
Use simple counters or captured values:
```kotlin
var selectedId: String? = null
composeTestRule.setContent {
ItemList(
items = listOf(ItemUi("movie-1", "Movie")),
onItemClick = { selectedId = it },
)
}
composeTestRule.onNodeWithText("Movie").performClick()
assertThat(selectedId).isEqualTo("movie-1")
```
For plain captured callback values, a direct assertion after the action is usually enough. Use `runOnIdle` when the assertion needs Compose to finish applying snapshot state, recomposition, or queued UI work before reading the result.
## Keep UI tests deterministic
For layout, branch, and callback behavior, render controlled state with `setContent` instead of constructing the production app graph. Production DI, repositories, lifecycle observers, and background effects add asynchronous work that is irrelevant to a plain UI contract and can make the test flaky.
Do not use `Thread.sleep` to wait for Compose. Drive the UI to a known state, then use semantic assertions and Compose synchronization (`waitForIdle`, `runOnIdle`, or a bounded `waitUntil` for a real asynchronous condition). If you set `mainClock.autoAdvance = false`, advance the clock (`mainClock.advanceTimeBy`) to the time you intend to observe before asserting, not between checks of one frame; otherwise leave auto-advance on. Reserve full-app integration for behavior that actually depends on navigation, lifecycle, DI, or platform wiring.
## Interaction state with MutableInteractionSource
When interaction state is the concern, read
[the interaction-state procedure](references/interaction-state.md) completely
before writing or reviewing the test. It requires direct, injected interaction
state rather than fragile pointer simulation.
## Keyboard and focus
For keyboard, TV, and desktop UI, drive navigation with the same input model users use (keys/D-pad), not clicks alone. Assert focused semantics, not colors or scale; reserve screenshots for visual focus treatment.
If the inspected composable exposes no input surface that can change the state,
name the missing host or parent interaction seam before proposing a key-driven
test; do not assume a selector or control exists in the supplied UI.
Details—focus graph, `FocusRequester`, restoration, key handlers, and test patterns: [`compose-focus-navigation`](../compose-focus-navigation/SKILL.md).
## Screenshot tests
Use screenshots for visual contracts that semantics cannot prove:
- Layout spacing/alignment.
- Themed colors, typography, elevation, shadows.
- Image composition, gradients, overlays.
- Focus highlight appearance.
- Loading skeletons or dense visual states.
Keep screenshot state deterministic:
- Use fixed state data.
- Freeze clocks or animation progress when possible.
- Replace network/image loading with fake or preview handlers.
- Avoid asserting dynamic text such as current time unless controlled.
### When screenshot defaults change
- Keep an intentionally changed named preset or default in the test and update
its baseline. Do not substitute a raw value to retain the old image; preserve
assertions and tolerances unless independently justified.
- Keep an explicit fixed value when that fixed resolution or geometry is the
contract.
- Verify recording separately from comparison: inspect the expected artifact
paths and intentional baseline diff. Read the exact output path from the test,
confirm that artifact exists, and name that literal path in the verification
evidence. When a verification log reports a pass, state what passed and then
check recording independently; a passing comparison does not prove that
recording wrote the artifact. If the artifact or baseline diff is
unavailable, report that gap rather than claiming recording succeeded. Keep
tool-specific commands in the repository runbook. In a findings-only review,
state what evidence is missing; do not add remediation or rerun steps.
## Fake images and platform services
When image content is irrelevant, fake the loader and assert the requested model if that is the behavior. The exact hook depends on your image library; a project helper might look like this:
```kotlin
val requestedModels = mutableListOf<Any?>()
// Example helper, not a Compose API.
setContentWithFakeImageLoader { request ->
requestedModels += request.data
errorPainter()
}
```
When image appearance matters, provide a deterministic local painter/bitmap instead of network data.
© chrisbanes · Apache-2.0 · chrisbanes/skills