# Android Compose Architecture Review

> Analyze Jetpack Compose UI hierarchies and suggest MVVM/MVI or other architectural improvements. Use when reviewing existing Compose code, creating new Compose components, analyzing composable structure, or when the user asks about Compose architecture patterns. Best for code review and refactoring guidance.

- Skill: `desquared/android-compose-architecture-review` (Agent Skill)
- Install (CLI): `npx skillmds@latest add desquared/android-compose-architecture-review`
- Raw SKILL.md: https://api.skillmd.com/api/skills/desquared/android-compose-architecture-review/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: desquared (https://skillmd.com/u/desquared)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/desquared/android-compose-architecture-review

---


# Compose Architecture Review

> **When to Use This Skill:**
> - Reviewing EXISTING Compose screens/ViewModels
> - Code review for architectural violations
> - Refactoring guidance for specific files
> - Validating state management patterns
>
> **When to Use `project-architecture-analyst-android` Agent Instead:**
> - Planning NEW features (uncertain file placement)
> - Understanding project-wide architecture
> - Major changes affecting multiple modules
> - Need to discover existing patterns before implementing

## Review Checklist

### Separation of Concerns
- [ ] Composables: presentation only (no business logic, network, DB)
- [ ] ViewModels: state + business logic (StateFlow, events)
- [ ] Repository/UseCase: data access separated from UI

### State Management
- `remember` - composable-local state only
- `rememberSaveable` - survive config changes
- `ViewModel` with `StateFlow/MutableStateFlow` - screen state
- `collectAsStateWithLifecycle()` - lifecycle-aware collection
- `derivedStateOf` - computed state from other state
- CompositionLocal - shared read-only state
- Use unidirectional data flow: state flows down, events flow up
- Prefer stateless composables (receive value + onChange callbacks)
- Use plain state-holder classes for complex state not tied to lifecycle
- Prefer side effects via `LaunchedEffect` (async work), `DisposableEffect` (cleanup), `SideEffect` (framework integration), `produceState`, `rememberCoroutineScope`, or `snapshotFlow` rather than arbitrary side effects in composable bodies
- Use `SavedStateHandle` for persistent state across process death

#### State Lifecycle Issues
- [ ] No leaked ViewModels (use viewModel() or hiltViewModel())
- [ ] Proper cleanup in ViewModel.onCleared()
- [ ] StateFlow collected with lifecycle awareness
- [ ] No unnecessary state hoisting

### Dependency Injection
- [ ] Dependencies injected (check project: Koin, Hilt, Dagger, or manual)
- [ ] No hardcoded singletons in composables
- [ ] ViewModels receive dependencies via constructor

### Composable Design
- [ ] Composables are small (single responsibility)
- [ ] Complex screens broken into components
- [ ] Reusable components extracted
- [ ] Stateless composables preferred (state hoisting)

## Common Issues

| Issue                                      | Fix |
|--------------------------------------------|-----|
| Business logic in composable               | Extract to ViewModel |
| Network call in composable                 | Move to ViewModel with LaunchedEffect |
| Singleton in composable                    | Inject via DI or CompositionLocal |
| 200+ line composable                       | Break into smaller components |
| State not hoisted                          | Hoist to parent or ViewModel |
| Side effects or network calls directly in composable | Wrap in `LaunchedEffect(key) { ... }` or move to ViewModel |
| Stateful composable when stateless would suffice | Hoist state and pass value + lambda |

## State Hoisting Pattern

Composable receives: `value: T, onValueChange: (T) -> Unit`  
ViewModel holds: `MutableStateFlow<T>`

## Severity

- 🔴 Critical: Architecture violation
- 🟡 Improvement: Better pattern available
- 🟢 Enhancement: Optional optimization