Android App Development Skill

← Back to skills

This skill guides production-grade Android and cross-platform (non-iOS) app development following practices used at big tech companies. It covers the entire development lifecycle — architecture, UI, code quality, testing, error handling, release, and maintenance.

Category: Mobile & Native Apps
Repo: antigravity-awesome-skills
Path: skills/android-dev/SKILL.md
Updated: 6/12/2026, 5:19:41 PM

Android App Development Skill

Overview

This skill guides production-grade Android and cross-platform (non-iOS) app development following practices used at big tech companies. It covers the entire development lifecycle — architecture, UI, code quality, testing, error handling, release, and maintenance.

When to Use This Skill

  • Use when deciding on a tech stack (see §1 Stack Selection)
  • Use when setting up project architecture (see §2 Architecture)
  • Use when designing UI, screens, or a design system (see §3 UI & Design)
  • Use when ensuring code quality, patterns, or APIs (see Best Practices)
  • Use when implementing error handling or debugging crashes (see §5 Error Handling)
  • Use when planning testing strategy (see §6 Testing)
  • Use when configuring build, CI/CD, or release pipelines (see §7 Build & Release)
  • Use when optimizing performance or memory (see §8 Performance)
  • Use when debugging or fixing bugs (see §9 Debugging)
  • Use when following the full development roadmap (see §10 Development Roadmap)
  • Use when needing deep reference for a stack (see references/ directory)

§1 Stack Selection

Choose based on team, requirements, and platform targets. Do not recommend iOS-specific paths.

Native Android — Kotlin + Jetpack Compose

Best for: Android-only apps, hardware-intensive features, best-in-class UX, new projects.

  • Language: Kotlin
  • UI: Jetpack Compose (modern declarative UI)
  • Key libs: Room, Retrofit/Ktor, Hilt, WorkManager, DataStore, Navigation Compose
  • Reference: references/native-android.md

Native Android — Java + XML Views

Best for: Existing Java codebases, teams without Kotlin experience, legacy app maintenance, incremental Kotlin migration.

  • Language: Java (fully supported by Google, not deprecated)
  • UI: XML Layouts (ConstraintLayout, RecyclerView, ViewBinding)
  • Key libs: Room, Retrofit, Hilt, WorkManager, LiveData, ViewModel
  • Java and Kotlin coexist seamlessly in the same project — migrate incrementally
  • Reference: references/java-android.md

Flutter (Dart)

Best for: Android + Web (+ desktop) from one codebase, fast iteration, pixel-perfect custom UI.

  • Language: Dart
  • UI: Flutter Widget tree (Material 3 / Cupertino widgets available but target Material for Android)
  • Key libs: Provider/Riverpod/Bloc, Dio, Drift/Isar, go_router, flutter_local_notifications
  • Reference: references/flutter.md

React Native (JavaScript/TypeScript)

Best for: Web + Android code sharing, JS/TS teams, rich ecosystem.

  • Language: TypeScript (preferred)
  • UI: React Native core components + NativeWind / React Native Paper
  • Key libs: React Navigation, Zustand/Redux Toolkit, React Query, MMKV
  • Reference: references/react-native.md

Kotlin Multiplatform (KMM / Compose Multiplatform)

Best for: Sharing business logic across Android + Desktop + Web while keeping native Android UI.

  • Language: Kotlin everywhere
  • UI: Native Compose on Android; Compose Multiplatform for shared UI
  • Key libs: Ktor, SQLDelight, Koin, kotlinx.serialization, Napier
  • Reference: references/kmm.md

Hybrid (Capacitor / Ionic)

Best for: Web-first teams, simple apps, PWA-like content apps.

  • Language: TypeScript + HTML/CSS
  • UI: Ionic components or custom web UI
  • Avoid for: Heavy animations, native sensor access, high-performance games
  • Reference: references/hybrid.md

Decision Matrix

RequirementNative KotlinNative JavaFlutterRNKMMHybrid
Android-only (new)✅ Best
Android-only (existing Java)⚠️ migrate✅ Best⚠️
Android + Web✅ Best
Android + Desktop⚠️⚠️
Shared business logic onlyN/AN/AN/AN/A✅ BestN/A
Native performance⚠️
JS/TS team✅ Best
Custom pixel-perfect UI⚠️✅ Best⚠️

§2 Architecture

Core Principle: Separation of Concerns

Every production Android project must separate UI, business logic, and data into distinct, independently testable layers.

Recommended Architecture: Clean Architecture + MVI/MVVM

app/
├── ui/              # Composables / Activities / Fragments / Screen states
├── presentation/    # ViewModels, UI State, UI Events
├── domain/          # Use cases, domain models, repository interfaces
├── data/            # Repository impl, remote (API), local (DB), mappers
└── di/              # Dependency injection modules

Data flow (unidirectional):

User Action → ViewModel/Store → Use Case → Repository → Data Source
                    ↓
             UI State (sealed class / StateFlow)
                    ↓
             Composable / View renders state

Key Architecture Patterns by Stack

Native (MVVM + MVI):

  • StateFlow / SharedFlow for reactive state
  • sealed class UiState + sealed class UiEvent
  • Hilt for DI, coroutines + Flow for async
  • Repository pattern wrapping Room + Retrofit

Flutter (BLoC or Riverpod):

  • Bloc or Cubit for business logic isolation
  • AsyncNotifierProvider (Riverpod) for data + state
  • Repositories as abstract classes with impl injected

React Native (Redux Toolkit or Zustand):

  • RTK Query or React Query for server state
  • Zustand slices for client state
  • Custom hooks to encapsulate business logic per feature

KMM:

  • Shared commonMain holds domain + data layers
  • expect/actual for platform-specific implementations
  • Kotlin coroutines + Flow bridged to platform (StateFlow on Android)

Module Structure (Multi-module for large apps)

:app            # Entry point, DI wiring
:core:ui        # Design system, shared composables
:core:network   # API client, interceptors
:core:database  # Room / SQLDelight setup
:feature:home
:feature:profile
:feature:settings

§3 UI & Design

Design System First

Before writing screens, define:

  1. Color tokens — Primary, secondary, surface, on-surface, error; light + dark variants
  2. Typography scale — Display, headline, title, body, label (Material 3 type system)
  3. Spacing scale — 4dp grid system (4, 8, 12, 16, 24, 32, 48dp)
  4. Shape tokens — Corner radii per component family
  5. Component library — Button, TextField, Card, BottomSheet, TopAppBar, etc.

Jetpack Compose UI Rules

  • Use MaterialTheme tokens; never hardcode colors/dimensions
  • CompositionLocal for theme, locale, haptics
  • remember / rememberSaveable correctly (saveable for UI state surviving rotation)
  • Extract large composables into sub-composables; each function ≤ 80 lines
  • Use LazyColumn/LazyVerticalGrid for lists; never Column with forEach for large data
  • Side effects only in LaunchedEffect, DisposableEffect, SideEffect
  • Avoid state hoisting anti-patterns: hoist state to the lowest common ancestor

Accessibility (Non-Negotiable)

  • All interactive elements: contentDescription or semantics { }
  • Min touch target: 48×48dp
  • TalkBack compatibility tested before every release
  • Dynamic text size support (sp not dp for text)
  • Color contrast ratio ≥ 4.5:1 (WCAG AA)

Navigation

  • Native: Navigation Compose with typed NavHost and SafeArgs equivalent
  • Flutter: go_router with named routes and guards
  • RN: React Navigation v7 with typed NavigationProp
  • Deep link handling registered for every screen that can be externally opened
  • Back stack managed deliberately — don't push duplicates, use popUpTo / launchSingleTop

Responsive & Adaptive UI

  • Support all screen sizes: phones, foldables, tablets (WindowSizeClass)
  • Test at 320dp, 360dp, 411dp, 600dp+, 840dp+ widths
  • Foldable hinge awareness via WindowInfoTracker
  • Edge-to-edge display + WindowInsets handling required for Android 15+

Best Practices

Language Standards

Kotlin:

  • Prefer data class, sealed class, object, enum class appropriately
  • No !! null assertions — use ?.let, ?: return, requireNotNull with message
  • Coroutines: always specify CoroutineScope + Dispatcher explicitly; never GlobalScope
  • Use @Stable / @Immutable on Compose state classes for smart recomposition

Java:

  • @NonNull / @Nullable annotations on every method param and return type
  • Never call methods on unchecked objects — null-check explicitly or use Objects.requireNonNull
  • Always null binding reference in Fragment's onDestroyView() to prevent memory leaks
  • Use ExecutorService (not AsyncTask — deprecated) for background work; or LiveData + Room's built-in threading
  • Prefer ListAdapter + DiffUtil over manual notifyDataSetChanged() in RecyclerView
  • Use ViewBinding — never findViewById

Dart (Flutter):

  • Null safety required — no ! without explicit null check above
  • Immutable state objects with copyWith
  • const constructors on all stateless widgets

TypeScript (RN):

  • strict: true in tsconfig always
  • Zod or io-ts for runtime type validation of API responses
  • No any — use unknown and narrow

Dependency Management

  • Pin all dependency versions in build.gradle.kts / pubspec.yaml / package.json
  • Audit dependencies monthly for security vulnerabilities
  • Avoid transitive dependency conflicts — use dependency resolution strategies
  • Keep dependency count minimal — every added lib is a maintenance burden

Code Review Checklist (PR gate)

  • New public APIs have KDoc / DartDoc / JSDoc
  • No hardcoded strings — use string resources / l10n
  • No hardcoded dimensions or colors outside design tokens
  • No blocking I/O on main thread
  • No memory leaks (no Activity context stored in singletons)
  • Coroutine scopes / streams properly cancelled / disposed
  • Feature flag guarding any non-trivial feature

§5 Error Handling

The Golden Rule

Never let exceptions propagate to the user silently or crash the app.

Error Classification

TypeStrategy
Network errorsRetry with exponential backoff; show retry UI
Auth errors (401/403)Refresh token → re-request → logout if fails
Validation errorsShow inline field errors immediately
Data parsing errorsLog + fallback to cached/default state
Unexpected crashesCatch at top-level; show error screen + report
Background task failuresRetry via WorkManager; notify user if critical

Result / Either Pattern (Kotlin)

sealed class AppResult<out T> {
    data class Success<T>(val data: T) : AppResult<T>()
    data class Error(val exception: AppException) : AppResult<Nothing>()
}

sealed class AppException(msg: String) : Exception(msg) {
    class NetworkException(msg: String) : AppException(msg)
    class AuthException(msg: String) : AppException(msg)
    class ParseException(msg: String) : AppException(msg)
    class UnknownException(msg: String) : AppException(msg)
}

Use AppResult<T> as return type for all repository + use case functions. ViewModels map to UiState.Error.

Crash Reporting

  • Integrate Firebase Crashlytics or Sentry from day one
  • Set user identifiers and custom keys before crash occurs
  • Non-fatal exceptions logged for all caught errors
  • ANR monitoring enabled
  • Crash-free sessions target: ≥ 99.5%

Offline / Network Resilience

  • Cache-first strategy: show stale data, fetch fresh in background
  • Room / Drift / MMKV as single source of truth
  • Expose network state via ConnectivityManager and reflect in UI
  • All network calls wrapped with timeout + retry policy

§6 Testing

Testing Pyramid

         /\
        /E2E\        ← 10%  (UI tests: Espresso, Maestro, Appium)
       /------\
      / Integr \     ← 20%  (Repository, DB, API contract tests)
     /----------\
    /    Unit    \   ← 70%  (ViewModels, Use Cases, Utilities)
   /--------------\

Unit Tests (70%)

  • Every ViewModel, UseCase, Repository, Mapper tested
  • Native: JUnit5 + MockK + Turbine (Flow testing) + Kotest assertions
  • Flutter: flutter_test + mocktail
  • RN: Jest + @testing-library/react-native + msw for API mocking
  • Coverage target: ≥ 80% on domain + presentation layers

Integration Tests (20%)

  • Room DB tests with in-memory database
  • Retrofit/Ktor tests with MockWebServer (OkHttp)
  • Repository tests verifying cache + remote coordination
  • API contract tests against real staging endpoint

UI / E2E Tests (10%)

  • Espresso for critical user journeys (login, checkout, core action)
  • Maestro for cross-platform E2E flows (recommended for Flutter + RN too)
  • Run on real device farm (Firebase Test Lab / BrowserStack) before release
  • Smoke test suite runs on every PR; full E2E suite nightly

Test Data Management

  • Use factories / builders for test data, never copy-paste objects
  • Hermetic tests: never share mutable state between test cases
  • Fakes over mocks for complex dependencies (repositories, data sources)

§7 Build & Release

Build Variants

debug       → dev API, logging on, no minification, debuggable
staging     → staging API, logging on, minified, not debuggable
release     → prod API, logging off, minified, signed

Gradle Best Practices (Native)

  • build.gradle.kts only — no Groovy DSL in new projects
  • Version catalog (libs.versions.toml) for all dependency versions
  • buildConfig for environment-specific constants
  • Baseline profiles for startup performance
  • R8 full mode enabled in release; maintain proguard rules in version control

CI/CD Pipeline

PR Opened
  └─ lint + unit tests + build debug APK          [< 5 min]

Merge to main
  └─ unit + integration tests + staging build     [< 15 min]
  └─ deploy to Firebase App Distribution (QA)

Release tag
  └─ full test suite + E2E on device farm         [< 45 min]
  └─ build release AAB
  └─ upload to Play Console (internal track)
  └─ promote: internal → closed testing → open → production

Recommended CI: GitHub Actions, Bitrise, or CircleCI.

Play Store Release Strategy

  • Always release to internal → closed → open testing before production
  • Use staged rollouts: 5% → 20% → 50% → 100% with 24-48h monitoring
  • Monitor Crashlytics + ANR rate + rating before expanding rollout
  • Never skip staged rollout for significant changes

App Signing

  • Upload key (Play App Signing): stored in CI secrets, never committed
  • Use Google Play App Signing for distribution key management
  • Document key recovery procedure in team runbook

§8 Performance

Startup Performance

  • App startup time target: cold start < 1s, warm start < 500ms
  • Use App Startup library for initializing libraries lazily
  • Baseline profiles generated + committed to repo
  • Heavy initialization moved off main thread

UI Performance

  • Target: 60fps (90/120fps on supported devices); zero jank
  • Measure with Android Studio Profiler + FrameMetrics API
  • Avoid allocation in draw() / onMeasure() / composition
  • Use derivedStateOf in Compose to avoid unnecessary recompositions
  • Image loading: Coil (Compose) / Glide / Picasso — never load full-res in thumbnails

Memory

  • No Activity / Context references in ViewModels or singletons
  • WeakReferences for listeners stored beyond their owner's lifecycle
  • Bitmap recycling and memory cache sizing
  • Heap dump + leak detection via LeakCanary in debug builds (always)

Network

  • HTTP caching headers respected
  • Image CDN + WebP format
  • Gzip/Brotli compression verified
  • Request batching where applicable
  • Connection pooling configured

Battery

  • Background work only via WorkManager with appropriate constraints
  • Location updates: request only needed accuracy level; stop when backgrounded
  • Wakelocks used sparingly with explicit release

§9 Debugging & Bug Fixing

Debugging Process

  1. Reproduce reliably — document exact steps, device, OS version, account state
  2. Isolate — is it UI, business logic, network, or persistence?
  3. Instrument — add targeted logs / breakpoints, NOT shotgun logging
  4. Hypothesize — form 1-3 specific hypotheses before touching code
  5. Fix the root cause — never patch symptoms; trace back to the source
  6. Regression test — write a test that fails before fix, passes after
  7. Document — comment explaining why the fix works, not just what it does

Common Android Bug Patterns

BugLikely CauseFix
ANRMain thread I/O / long computationMove to coroutine/Dispatcher.IO
Memory leakContext stored in singletonUse applicationContext; WeakRef
Crash on rotationViewModel not used; state not savedrememberSaveable / ViewModel
UI lagRecomposition loopsderivedStateOf, stable params
Blank screen after API callError swallowed silentlyCheck error state propagation
Deep link not workingManifest intent-filter missingVerify adb shell am start test
Push notification silentBackground restrictionsTest on real devices across OEMs

Logging Standards

  • Production: Firebase Crashlytics only (no Log.d in release builds)
  • Debug/Staging: Timber with debug tree
  • Log levels: ERROR (crashes), WARN (recoverable), INFO (key events), DEBUG (dev only)
  • Never log PII — mask emails, phone numbers, tokens in logs

OEM-Specific Issues

  • Test on Samsung, Xiaomi/MIUI, OnePlus/OxygenOS, Huawei (no GMS) for critical flows
  • Background restrictions vary widely by OEM — test push, alarms, background sync
  • Maintain a physical or cloud device farm with top market-share devices

§10 Development Roadmap

Follow this phase structure for any new Android project:

Phase 0 — Foundation (Week 1-2)

  • Stack decision documented with rationale
  • Module structure defined
  • Design system tokens defined (colors, type, spacing, shapes)
  • CI pipeline running (lint + unit tests + build)
  • Crash reporting integrated (Crashlytics/Sentry)
  • Analytics baseline integrated (Firebase/Amplitude)
  • API contract / mock server set up
  • DI framework configured
  • Navigation skeleton implemented
  • Flavor/build variant config complete

Phase 1 — Core Features (Weeks 3-8)

  • Auth flow (login, register, token refresh, logout)
  • Core screen shells with real navigation
  • Network layer (client, interceptors, error handling)
  • Local persistence layer (DB schema + DAOs)
  • Repository layer wiring remote + local
  • ViewModels + UI states for each feature
  • Unit tests for all ViewModels + use cases
  • Feature flags infrastructure

Phase 2 — Polish (Weeks 9-12)

  • Design QA pass against Figma/spec
  • Accessibility audit (TalkBack, contrast, touch targets)
  • Dark mode implementation + verification
  • Localization (strings externalized, RTL support if needed)
  • Loading, empty, error states on every screen
  • Deep link handling
  • Widget / notification implementation
  • Offline mode verification

Phase 3 — Hardening (Weeks 12-14)

  • Performance profiling (startup, scroll, memory)
  • E2E test suite on device farm (Firebase Test Lab)
  • Security review (certificate pinning, biometrics, secure storage)
  • Proguard / R8 rules verified
  • Crash-free rate ≥ 99.5% on staging
  • Play Store listing, screenshots, privacy policy

Phase 4 — Release

  • AAB signed and uploaded to internal track
  • Staged rollout plan defined
  • Monitoring dashboard set up (Crashlytics, Play Console vitals)
  • Rollback plan documented
  • On-call rotation assigned

Phase 5 — Post-Launch (Ongoing)

  • Crash-free rate monitored daily
  • ANR rate < 0.47% (Play Store threshold)
  • App rating monitored; negative reviews triaged weekly
  • Dependency updates reviewed monthly
  • OS beta testing with each new Android release

Limitations

  • This skill is scoped to Android and Android-adjacent delivery paths; it does not cover iOS-only architecture, App Store release operations, or Apple platform UI guidance.
  • Version numbers, Play Console policy thresholds, and recommended libraries can change; verify release-critical details against current Android, Google Play, and library documentation before shipping.
  • Code snippets are architecture patterns, not complete applications; adapt package names, dependency versions, permissions, privacy disclosures, and security controls to the actual project.
  • The guidance does not replace device QA, accessibility review, security review, legal/privacy review, or store compliance checks for a production release.

Additional Resources

For stack-specific deep dives, read:

  • references/native-android.md — Kotlin, Compose, Room, Hilt, Coroutines
  • references/java-android.md — Java, XML Views, ViewBinding, LiveData, Retrofit, Room, Hilt, migration path
  • references/flutter.md — Dart, BLoC/Riverpod, Drift, go_router
  • references/react-native.md — TypeScript, RN architecture, Hermes, New Architecture
  • references/kmm.md — KMM shared modules, SQLDelight, Ktor, Compose Multiplatform
  • references/hybrid.md — Capacitor, Ionic, PWA considerations

Related skills