Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build high-performance declarative UI with Jetpack Compose. Use when writing Composable functions, optimizing recomposition, hoisting state, or working with LazyColumn and side effects; defer deep-link and navigation routing to android-navigation.
.claude/skills/hoangnguyen0403-android-compose/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 13 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-03 | ✓→✓ | = Same ✓ | 6% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 10% | 0% |
| case-06 | ✓→✓ | = Same ✓ | -7% | 0% |
Role: Android UI Performance Expert. Prioritize frame stability and state management.
onItemClick: (Id) -> Unit).MaterialTheme.colorScheme, no hardcoded hex.See implementation examples for state hoisting patterns.
@Stable or @Immutable.key in LazyColumn items for stable identity.derivedStateOf for frequently updating derived values.See implementation examples for derivedStateOf usage.
LaunchedEffect for one-shot or keyed side effects — never run side effects in composition body.remember.LaunchedEffect, not raw coroutines.remember.@Preview.LazyColumn items use key parameter../gradlew build succeeds.When this skill applies, preserve the following domain terminology or equivalent concrete examples in the answer when relevant:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→fail | 16,263 | 9,052 | -44% | 1 | 1 | 0% | 3,243 | 2,138 | -34% | 0 | 0 | — |
case-01 | fail→pass | 13,223 | 17,302 | +31% | 1 | 1 | 0% | 2,575 | 4,221 | +64% | 0 | 0 | — |
case-02 | fail→pass | 19,488 | 15,133 | -22% | 1 | 1 | 0% | 3,956 | 4,037 | +2% | 0 | 0 | — |
case-03 | pass→pass | 14,227 | 13,277 | -7% | 1 | 1 | 0% | 2,685 | 2,851 | +6% | 0 | 0 | — |
case-04 | pass→pass | 14,101 | 12,896 | -9% | 1 | 1 | 0% | 2,915 | 3,200 | +10% | 0 | 0 | — |
case-06 | pass→pass | 12,210 | 8,851 | -28% | 1 | 1 | 0% | 2,159 | 2,015 | -7% | 0 | 0 | — |
case-07 | pass→pass | 9,611 | 7,910 | -18% | 1 | 1 | 0% | 1,817 | 1,937 | +7% | 0 | 0 | — |
case-08 | pass→pass | 11,181 | 9,110 | -19% | 1 | 1 | 0% | 2,126 | 2,232 | +5% | 0 | 0 | — |
case-09 | pass→pass | 9,066 | 5,166 | -43% | 1 | 1 | 0% | 1,719 | 1,376 | -20% | 0 | 0 | — |
case-10 | pass→pass | 8,902 | 7,068 | -21% | 1 | 1 | 0% | 1,758 | 1,730 | -2% | 0 | 0 | — |
case-11 | pass→pass | 9,459 | 8,163 | -14% | 1 | 1 | 0% | 1,780 | 2,084 | +17% | 0 | 0 | — |
case-12 | pass→pass | 13,556 | 8,300 | -39% | 1 | 1 | 0% | 2,373 | 2,075 | -13% | 0 | 0 | — |
case-13 | pass→pass | 11,514 | 8,631 | -25% | 1 | 1 | 0% | 2,237 | 2,118 | -5% | 0 | 0 | — |
case-14 | pass→pass | 12,912 | 8,970 | -31% | 1 | 1 | 0% | 2,459 | 2,135 | -13% | 0 | 0 | — |
case-15 | pass→pass | 10,740 | 8,086 | -25% | 1 | 1 | 0% | 2,032 | 1,882 | -7% | 0 | 0 | — |
case-16 | pass→pass | 10,317 | 4,053 | -61% | 1 | 1 | 0% | 1,869 | 1,174 | -37% | 0 | 0 | — |
case-17 | pass→pass | 4,569 | 5,694 | +25% | 1 | 1 | 0% | 751 | 1,446 | +93% | 0 | 0 | — |
case-18 | pass→pass | 11,854 | 9,414 | -21% | 1 | 1 | 0% | 2,158 | 1,930 | -11% | 0 | 0 | — |
case-19 | pass→pass | 13,347 | 11,040 | -17% | 1 | 1 | 0% | 2,494 | 2,594 | +4% | 0 | 0 | — |
case-20 | pass→pass | 11,004 | 8,243 | -25% | 1 | 1 | 0% | 1,943 | 1,968 | +1% | 0 | 0 | — |
case-21 | fail→fail | 9,466 | 10,557 | +12% | 1 | 1 | 0% | 1,726 | 2,482 | +44% | 0 | 0 | — |
case-22 | pass→pass | 10,027 | 7,601 | -24% | 1 | 1 | 0% | 1,872 | 1,901 | +2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.