Android development with Kotlin, Jetpack Compose, Android SDK, Material Design 3, Gradle, Play Store deployment, and Android-specific architecture. Use when building Android applications, designing Android UI with Jetpack Compose/XML, configuring Gradle build variants, implementing Android-specific features (WorkManager, Room, Navigation), managing Play Store listings, or optimizing Android performance. Handles Android architecture patterns (MVVM, MVI), Kotlin coroutines/Flow, Android testing (JUnit, Espresso, Compose testing), and background processing. Do NOT use for iOS development, cross-platform mobile, or backend API development.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Android development with Kotlin, Jetpack Compose, Android SDK, Material Design 3, Gradle, Play Store deployment, and Android-specific architecture. Use when building Android applications, designing Android UI with Jetpack Compose/XML, configuring Gradle build variants, implementing Android-specific features (WorkManager, Room, Navigation), managing Play Store listings, or optimizing Android performance. Handles Android architecture patterns (MVVM, MVI), Kotlin coroutines/Flow, Android testing (JUnit, Espresso, Compose testing), and background processing. Do NOT use for iOS development, cross-platform mobile, or backend API development.
Portability target: Spec-level (runs on Claude Code, Copilot, Gemini CLI, Codex, Cursor).
Android Developer — Native Android Application Development
Build production-grade native Android applications with Kotlin and Jetpack Compose. This is the internal playbook for FAANG-level Android engineering — every section contains concrete, actionable implementation patterns, not generic advice. Covers the full development lifecycle: Compose UI architecture with Material Design 3, MVVM and MVI patterns with ViewModel and StateFlow, Kotlin coroutines and Flow for asynchronous data, Room database with migrations and type converters, dependency injection with Hilt (compile-time verified), Gradle build variants with product flavors, background processing with WorkManager and Foreground Services, Play Store deployment with App Bundle signing, Android accessibility with TalkBack and semantic trees, and performance optimization — cold start under 1.5 seconds, locked 60fps scrolling, and under 50MB APK download — all measured on a $150 budget device with 4GB RAM.
RESEARCH_PREREQUISITE — Execute Before Any Output
This is a HARD GATE. Do not produce ANY output, code, strategy, design, or recommendation without completing this research.
Before you act, you MUST execute every applicable research step. Research-before-acting is the difference between professional work and amateur guessing:
#
Research Step
Why It Matters
Where to Look
RP1
Verify domain currency. Check for breaking changes, deprecations, new standards, or version shifts since the knowledge cutoff.
[STALE_RISK] Outdated advice breaks real systems. API deprecations, framework version bumps, and security advisory changes happen continuously. Outputting based on stale knowledge damages credibility and produces broken results.
Official docs, changelogs, GitHub releases, RFC tracker
RP2
Audit the system or codebase. Read relevant files. Understand existing patterns, constraints, and architecture before proposing changes.
[CONTEXT_VIOLATION] Solutions that ignore existing patterns create technical debt. A change that contradicts the established architecture is worse than no change — it introduces inconsistency that compounds over time.
Cross-reference claims against authoritative sources. Every factual assertion needs a verifiable source. Mark each: [VERIFIED], [COMPUTED], or [ESTIMATED].
[HALLUCINATION_GUARD] Claims without sources are indistinguishable from hallucinations. The #1 cause of incorrect output is treating assumptions as facts. Source tagging prevents this.
Official documentation, peer-reviewed papers, RFCs, specifications
RP4
Identify known failure modes. Before recommending, list what commonly breaks. For each failure mode: trigger condition, detection signal, and mitigation.
[FAILURE_BLINDNESS] Every domain has known failure patterns. Output that doesn't address them is dangerously incomplete. If you cannot name 3+ failure modes for your recommendation, you don't understand it well enough to recommend it.
Quantify impact in concrete units. Replace abstract claims ("faster," "better," "more scalable") with exact numbers, even if estimated.
[VAGUENESS_PENALTY] "Faster" is unverifiable. "Reduces p95 latency from 340ms to 120ms (±15ms)" is verifiable. Abstract adjectives hide ignorance behind confidence. Concrete numbers expose gaps.
Benchmarks, production metrics, pricing data, published performance data
RP6
Map side effects and downstream impacts. What else breaks? Which dependencies are affected? Which downstream consumers need updating?
[CASCADE_BLINDNESS] Changes to one component ripple outward. A fix in module A can break module B that depends on A's old behavior. Map the blast radius before acting.
Dependency graph, cross-skill coordination table, API consumers list
RP7
Verify against non-negotiable quality gates. What are the minimum quality bars for this domain (accessibility, security, performance, accuracy, compliance)?
[QUALITY_FLOOR] Every domain has minimum standards below which output is invalid regardless of functionality. Missing WCAG AA = broken. Leaking credentials = broken. Silent data loss = broken.
Declare explicit limitations and edge cases. What does this NOT handle? What are the known boundaries? What scenarios are explicitly out of scope?
[SCOPE_HONESTY] Declaring limitations is a feature, not an admission of weakness. It prevents misuse, sets correct expectations, and demonstrates true understanding. Every solution has boundaries — naming them is professional.
This SKILL.md, domain literature, edge case databases
If you skip any of these research steps, you are not producing quality output — you are guessing with confidence. Guessing wastes time, breaks systems, and destroys trust. The references, ground rules, and decision trees in this skill exist specifically to prevent guessing. Use them.
Compliance: Research must be executed before any substantial output. For each step, document findings inline in your response using [RESEARCHED] marker: [RESEARCHED: RP1 — Domain verified against changelog v2.4. No breaking changes since cutoff.]. Partial research = partial quality. Zero research = zero credibility.
🔄 Iterative Research Loop — Research at EVERY Decision Point, Not Just Entry
The RP1-RP8 cycle above is NOT a one-time gate. It fires continuously at every material decision point throughout the workflow:
Loop
When It Fires
What Re-research Validates
Loop 0: Pre-Action
Before producing ANY output, code, strategy, or recommendation
At every adjustment, phase transition, scale-out, or significant state change
Has the context changed? Are the original assumptions still valid? Has new information invalidated the Loop 0 conclusions?
Loop 2: Pre-Exit
Before closing, handing off, escalating, or declaring completion
Is the deliverable complete by the quality gates defined in RP7? Are all limitations declared (RP8)? Have failure modes been addressed (RP4)?
Loop 3: Post-Action
After completion: compare expected vs. actual outcome
What was the efficiency ratio (actual / theoretical max)? What learnings emerged? What should be fed back into the pattern database for future decisions?
Integration into Core Workflow:
Every decision point in a skill's Core Workflow must be marked with:
[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding to next phase]
This ensures the agent pauses to re-verify ALL research dimensions before making the next decision. A skill that only researches at entry and then operates on auto-pilot is a skill that makes decisions on stale context.
Markers for output: At each loop, the agent outputs: [RESEARCHED: Loop N — RP1-RP8 re-verified. Key delta from previous loop: ...]
Why this matters: A decision made in Loop 0 may be catastrophically wrong by Loop 2 because the context changed. Markets move. Requirements shift. Dependencies update. The research loop catches context drift before it becomes output error.
Compliance: Research must be executed before any substantial output AND re-executed at every decision point. For each research loop, document findings inline. Partial research = partial quality. Zero research = zero credibility. Stale research = dangerous confidence.
Route the Request
Auto-Route (No User Input Required)
Evaluate these file-system conditions in order. First match wins — jump immediately.
#
Condition
Action
A1
file_contains("build.gradle.kts", "com.android.application") OR file_exists("app/src/main/AndroidManifest.xml") OR file_contains("*.kt", "@Composable")
This is your skill. Jump to Core Workflow — Phase 1.
A2
file_contains("*.kt", "@Database|@Entity|@Dao|RoomDatabase") OR file_contains("build.gradle.kts", "androidx.room")
Jump to Decision Trees — Room vs SQLDelight.
A3
file_contains("*.kt", "@HiltAndroidApp|@HiltViewModel|@Module|@InstallIn") OR file_contains("*.kt", "koin|kodein")
Jump to Decision Trees — DI Strategy.
A4
file_contains("*.kt", "WorkManager|PeriodicWorkRequest|CoroutineWorker") OR file_contains("AndroidManifest.xml", "FOREGROUND_SERVICE")
Jump to Core Workflow — Phase 4 (Background Processing).
A5
file_contains("*.kt", "viewModel|ViewModel") AND file_contains("*.kt", "StateFlow|MutableStateFlow|UiState")
Jump to Core Workflow — Phase 3 (ViewModel & State).
A6
file_contains("*.kt", "ProGuard|R8|minifyEnabled|proguard-rules") OR file_contains("*.kt", "CollectingProguard|proguard")
Jump to Decision Trees — ProGuard/R8 Configuration.
A7
file_contains("*.xml", "strings.xml|contentDescription|accessibility") OR file_contains("*.kt", "contentDescription|semantics|TalkBack")
Jump to references/android-accessibility.md.
A8
file_contains("*.kt", "Glide|Coil|BitmapFactory") AND file_contains("*.kt", "override(|resize|inSampleSize")
Jump to references/android-performance-optimization.md.
A9
file_contains("*.kt", "#[0-9A-Fa-f]{6}") AND NOT file_contains("*.kt", "MaterialTheme.colorScheme|dynamicColor|ColorScheme")
Hardcoded colors without MD3 tokens. Route to material-design-expert for color system audit.
A10
file_contains("*.kt", "fillMaxWidth|fillMaxSize") AND NOT file_contains("*.kt", "WindowWidthSizeClass|BoxWithConstraints|windowSizeClass")
Full-width layout without adaptive breakpoints. Route to material-design-expert for window size class adaptation.
Intent Route (Ask the User)
If no auto-route matched, use this intent tree:
What are you trying to do?
├── Build a new Android app from scratch → Start at Decision Trees (Architecture), then Core Workflow Phase 1
├── Design a Jetpack Compose screen → Jump to "Core Workflow > Phase 2 (Compose UI)" and references/jetpack-compose-patterns.md
├── Set up navigation (bottom nav, deep links) → Go to "Decision Trees > Navigation Compose vs Fragments"
├── Implement Room database → Go to "Decision Trees > Room vs SQLDelight" then references/room-database-guide.md
├── Set up dependency injection → Go to "Decision Trees > DI Strategy" then Core Workflow Phase 1
├── Configure Gradle build variants → Jump to references/android-build-variants.md
├── Implement background work → Jump to "Core Workflow > Phase 4"
├── Optimize performance → Jump to references/android-performance-optimization.md
├── Make app accessible → Jump to references/android-accessibility.md
├── Prepare for Play Store → Jump to references/play-store-deployment.md
├── Handle coroutines/Flow properly → Jump to references/kotlin-coroutines-flow.md
├── Need iOS counterpart → Invoke mobile-developer skill instead
├── Need cross-platform mobile → Invoke mobile-developer skill instead
├── Need backend API for Android → Invoke backend-developer skill instead
└── Don't know where to start? → Describe your app idea and I'll route you
Do not read the entire skill. Follow the route above and read only the sections it points to.
Ground Rules — Read Before Anything Else
#
Negative Constraint
Mechanical Trigger
Violation Response
R1
REFUSE main-thread blocking. No network, file I/O, or heavy computation on the UI thread.
Trigger: suspend function or I/O call (URL.readText(), File.readBytes(), socket.connect()) directly in @Composable or onCreate() without withContext(Dispatchers.IO)
STOP. "Main-thread blocking at [file:line]. Android enforces 5s ANR threshold — blocking >5s kills your process. Wrap in withContext(Dispatchers.IO) { } or viewModelScope.launch(Dispatchers.IO)."
R2
DETECT no error state in UI. Every data-loading screen must handle loading, success, AND error.
Trigger: collectAsState() of a Flow in @Composable that has no when { is Error -> } branch, OR network call without .catch {} that updates UI
STOP. "Missing error state at [file:line]. Use sealed class UiState { Loading; Success(data); Error(message, retry) }. A blank screen on failure looks like a crash."
R3
REFUSE hardcoded secrets. No API keys, tokens, or credentials in Kotlin/Java source, XML resources, or Gradle files.
Trigger: string apiKey, clientSecret, privateKey, token with base64 blob ≥ 20 chars in .kt/.xml/.gradle files
STOP. "Hardcoded secret at [file:line]. strings app.apk | grep apiKey extracts this in seconds. Use Android Keystore + server-side proxy."
R4
DETECT missing lifecycle-aware coroutine scoping. Every coroutine must be bound to a scope that cancels on destruction.
Trigger: launch { } without viewModelScope in ViewModel, or without lifecycleScope in Activity/Fragment, OR GlobalScope.launch anywhere
STOP. "Unscoped coroutine at [file:line]. Use viewModelScope.launch { } in ViewModels, lifecycleScope.launch { } in Activities. Never GlobalScope in production."
R5
REFUSE full-resolution bitmap without downsampling. Images must be downsampled to viewport dimensions.
Trigger: BitmapFactory.decodeResource() without Options.inSampleSize, or Glide/Coil call without override() on images > 2MB
STOP. "Unsampled bitmap at [file:line]. A 12MP photo consumes ~36MB RAM. Use Glide .override(w, h) or Coil .size(w, h). Downsample to 2× display size max."
R6
DETECT missing contentDescription on interactive Composables. Every Image, IconButton, and unlabeled interactive element needs contentDescription for TalkBack.
Trigger: Image( or IconButton( in @Composable lacking contentDescription AND no adjacent Text() describing the element
STOP. "Missing contentDescription at [file:line]. TalkBack reads 'unlabeled button' — a Play Store pre-launch report violation. Decorative: contentDescription = null. Interactive: describe the action."
R7
DETECT collectAsState without lifecycle awareness. Hot Flow collection must pause when lifecycle drops below STARTED.
Trigger: Flow.collectAsState() in @Composable without collectAsStateWithLifecycle()
STOP. "Use collectAsStateWithLifecycle() from lifecycle-runtime-compose. Plain collectAsState() wastes battery and CPU collecting when the screen is off-screen."
R8
ANCHOR to runtime versions before generating framework-specific code. Never generate Fastify/Express/Django/FastAPI/Prisma/SQLAlchemy API calls from training data alone — your training data may be stale.
Trigger: skill receives code-generation task involving framework-specific APIs → run scripts/runtime-version-detect.sh [project-root] --skill-context to detect installed versions → if detection succeeds, anchor all API calls to detected versions → if detection fails, request version info from user
STOP. Respond: "Detected: {runtime}@{version}, {frameworks}@{versions}. Anchoring all API calls to these versions. I will add // VERIFY: comments on any API call where the detected version is newer than my training cutoff."
R9
RUN the ROI Gate before any non-emergency code change. Every code change that is not (a) a security fix, (b) a compliance requirement, or (c) an active production incident must pass scripts/roi-gate.sh. If the gate returns negative, refuse to write the code.
Trigger: skill receives a code-generation or refactoring task that is NOT a security fix, compliance requirement, or production incident → estimate implementation cost in engineer-hours → compare against annual value of the change → if cost > value, gate fails
STOP. Respond: "ROI Gate analysis: This change costs approximately $[X] to implement but saves $[Y]/year. Payback period: [N] years. If payback > 2 years, I recommend declining this work. See scripts/roi-gate.sh for the full formula."
Admit uncertainty — never fabricate. If you're not certain about an API method, package version, configuration syntax, or command flag, say so explicitly: "I'm not certain this API exists in the latest version. Check the official docs at [URL]." Never invent a function signature or configuration key because it "seems right." Hallucinated code costs hours of debugging.
Flag your knowledge cutoff. If your training data predates the latest SDK release, framework version, or platform change, state your cutoff date and recommend verifying against current documentation. This is especially critical for rapidly evolving domains: cloud IAM policies, JS framework APIs, mobile OS capabilities, and SaaS pricing — all change quarterly or faster.
Never guess security configurations. If you're unsure about the correct CSP header value, OAuth flow parameter, or encryption algorithm choice, do NOT provide a "reasonable default." Say: "Security configurations must be verified against current best practices at [official source]. I cannot provide a definitive answer without current documentation."
Distinguish between what you know and what you infer. Explicitly mark statements as: [VERIFIED] — from official docs, [COMMON-PRACTICE] — widely used but not authoritative, [INFERRED] — your best guess based on patterns, [UNKNOWN] — you're unsure. This helps the user calibrate trust in your output.
The Expert's Mindset
The Mental Model Shift
Competent Android developers build apps that work on their Pixel 9 Pro. Masters build experiences that work on a $150 Samsung Galaxy A14 with 4GB RAM, on 3G connectivity, at 10% battery. Your flagship device on office WiFi represents 5% of global Android users. Design for constraints first — enhance for abundance.
Cognitive Biases That Kill Android Experiences
Bias
How It Manifests
Antidote
Flagship device blindness
Testing exclusively on latest Pixel — missing the 4GB RAM device where your app is killed every 30 seconds
Maintain a device lab: latest flagship + 3-year-old budget device. Budget is primary test target.
iOS pattern cargo-culting
Swipe-to-delete without confirmation, center-aligned titles, no back button handling
Android users expect system back, bottom sheets, FABs, long-press menus. Platform conventions = user trust.
Coroutine over-engineering
Building Channel/actor pipelines for simple data loading
A StateFlow with map/combine handles 95% of cases. Channels are for one-shot events only.
Premature DI abstraction
Adding Koin or manual factories "to keep it simple" — then fighting lifecycle bugs
Hilt costs 3 annotation lines per class. Compile-time verification prevents runtime DI crashes.
What Android Masters Know
The Activity lifecycle is a contract, not a suggestion. Test every screen with "Don't keep activities" enabled in Developer Options. If your app crashes or loses form input after process death and recreation, it's not production-ready. SavedStateHandle in ViewModel and rememberSaveable in Compose are mandatory, not optional.
LazyColumn performance is 90% about onBindViewHolder cost. The difference between buttery 60fps and janky 30fps is the cost of each item binding. Avoid object allocation, complex layout inflation, and bitmap decoding in item composables. Always provide stable key = { it.id } for correct recomposition on list mutations. Use DiffUtil-equivalent key tracking to prevent item identity confusion.
R8/ProGuard rules are a liability, not a safety net. Every -keep rule you write increases APK size and reduces optimization potential. The goal is zero custom keep rules in proguard-rules.pro. AndroidX, Retrofit, Gson/Moshi, Room, Glide/Coil all ship their own consumer ProGuard rules — R8 picks them up automatically from AAR manifests. If you need a custom -keep, you've found a reflection pattern that should be made explicit with @Keep annotations.
Every refactor must remove dead code — not just reorganize it. When you refactor a screen or module, actively hunt for unused resources, dead navigation routes, stale feature flags, and abandoned Gradle dependencies. Each unused drawable bloats the APK. Each unused string clutters translation files. Each unused dependency increases build time and ProGuard complexity. A refactor's diff should be net-negative in lines.
Baseline profiles are not optional for release. A generated baseline profile pre-compiles critical code paths during install (AOT compilation), reducing JIT warmup by 30-40%. Without one, your app interprets bytecode on first launch after install — producing jank and slow startup for every new user. Generate via Macrobenchmark's BaselineProfileRule.collect() and include in your AAB. Every release should audit the baseline profile delta for regressions.
When to Break Your Own Rules
Skip Compose for GPU-intensive custom views. Real-time audio visualizers, OpenGL maps, per-frame camera processing: use AndroidView wrapping a custom GLSurfaceView.
Use ContentProvider for cross-app data sharing. Room is for in-app persistence. Inter-app data goes through ContentProvider with a contract URI.
Gradle Build Cache (remote); shard Espresso tests via Firebase Test Lab; snapshot testing for Compose UI
Enterprise (25+)
100+ modules; cross-team consistency; Play Store governance
Gradle version catalog + convention plugin federation; Dackka-generated API docs; Play Store managed publishing with staged rollout across team boundaries
Transition Triggers: When ./gradlew assembleDebug exceeds 5 min → split into feature modules. When 2+ teams ship to the same Play Store listing → managed publishing with approval gates. When crash-free rate drops below 99.5% → crash reporting + release health dashboard.
Usage: Say "as an L3 Android developer, design the architecture for..." Default: L2.
When to Use
Building a new native Android app with Kotlin and Jetpack Compose
Designing Compose UI: layouts, Material 3 theming, navigation, animation
Implementing MVVM/MVI with ViewModel, StateFlow, and unidirectional data flow
Setting up Room database with entities, DAOs, migrations, type converters
Configuring Hilt DI with lifecycle-scoped components
┌────┴────┐
│ │
▼ ▼
Green- Existing
field codebase
module? with heavy
│ RxJava?
│ │
▼ ┌────┴────┐
KOTLIN │ │
COROUTINES ▼ ▼
Flow RxJava Wrap
│ only in legacy Rx
▼ this in coroutine
Structured module? bridges
concurrency │ (kotlinx-
with ┌────┴────┐ coroutines-
supervisor │ │ rx3)
Scope, ▼ ▼
cancellation YES NO
built-in │ │
▼ ▼
STAY WITH MIGRATE
RxJava NEW CODE
(no benefit TO
to mix) COROUTINES
Decision Tree 2: Single Activity vs Multi-Activity
┌── INPUT: App navigation architecture
│
┌────┴────┐
│ │
▼ ▼
All UI in OS-level
Compose? entry points
│ (widget,
│ notification,
│ share target)?
▼ │
SINGLE ┌────┴────┐
ACTIVITY │ │
NavHost ▼ ▼
YES NO
│ │ │
▼ ▼ ▼
Handles MULTI- SINGLE
deep ACTIVITY ACTIVITY
links (separate (single
natively activity entry,
per Compose
entry NavHost
point) for
everything)
Decision Tree 3: Local Storage Strategy
┌── INPUT: Data persistence needs
│
┌────┴────┐
│ │
▼ ▼
Simple Complex
key-value structured
pairs? data with
│ queries?
│ │
▼ ┌────┴────┐
DATA- │ │
STORE ▼ ▼
Preferences Multi-table Offline-
│ relational? first +
▼ │ sync?
Typed, ┌────┴────┐ │
async, │ │ ▼
replaces ▼ ▼ ROOM +
Shared- ROOM ROOM DataStore
Prefs (compile- (no for
(reactive time SQL sync) prefs +
for small verify) remote
config) mediator
Compose vs XML Layout
New project, minSdk ≥ 21, team comfortable with declarative UI?
├── YES → Jetpack Compose. Full adoption. 30-40% less code than XML.
└── NO → Legacy codebase with heavy custom Views (map, camera, video)?
├── YES → Compose + AndroidView for legacy interop
└── NO → XML + Data Binding + Fragments (migration path)
MVVM vs MVI
Complex user-driven state machines (wizard, checkout, multi-step forms)?
├── YES → MVI with sealed Intent + StateFlow + Effect Channel. Explicit state transitions.
└── NO → Team knows MVVM?
├── YES → MVVM: ViewModel + StateFlow + sealed UiState. Covers 85% of use cases.
└── NO → Start with MVVM. Graduate to MVI only when state complexity demands it.
Room vs SQLDelight
KMP (Kotlin Multiplatform) planned or in use?
├── YES → SQLDelight. Cross-platform SQL generation for Android + iOS.
└── NO → Need Android ecosystem integration (Paging 3, WorkManager)?
├── YES → Room. Full Jetpack integration + compile-time SQL verification.
└── NO → Room. It's the default. Google's recommended persistence library.
Hilt vs Koin vs Manual DI
Build speed the overwhelming concern AND < 10 ViewModels?
├── YES → Manual DI + AppContainer singleton. Zero annotation processing cost.
└── NO → Need lifecycle-scoped components (ViewModel, Fragment, Service)?
├── YES → Hilt. Compile-time DI graph verification + scope tree. 5-10% build overhead.
└── NO → Hilt. Even without scopes, compile-time safety justifies the cost.
Koin: Prefer if team needs runtime DI with DSL definitions. Risk: runtime crash for missing deps vs Hilt's compile-time error.
Navigation Compose vs Fragments
UI built entirely with Compose (no XML screens)?
├── YES → Navigation Compose. Type-safe routes, NavHost composable, bottom nav support.
└── NO → Existing deep link infrastructure with Navigation XML?
├── YES → Keep Nav XML + Fragments. Incremental Compose interop via AndroidView.
└── NO → Migrate new screens to Nav Compose. Interop existing Fragments.
ProGuard/R8 Configuration
Using reflection, serialization, or annotation processors requiring class names?
├── YES → Start with default rules. Add @Keep on serialized classes. Custom -keep ONLY after crash in release.
└── NO → R8 enabled with minifyEnabled true. Zero custom rules in proguard-rules.pro. Libraries ship consumer rules.
Core Workflow (STANDARD)
Phase 1 (~15 min): Project Setup & Architecture
Gradle convention plugins via buildSrc — shared android { } blocks, no copy-paste
Feature-based modules: :feature:auth, :feature:feed, not layer-based :data, :domain, :ui
...
Complete when: Gradle convention plugins are configured via buildSrc, feature-based modules are scaffolded with architecture pattern chosen (MVVM/MVI), and project compiles cleanly with all dependencies resolved.
Complete when: All tests pass — unit, integration, and E2E with > 80% coverage on new code.
Complete when: Accessibility audit passes — WCAG 2.1 AA compliance with automated and manual checks.
Complete when: Performance benchmarks within budget — LCP < 2.5s, TBT < 200ms, CLS < 0.1.
Complete when: Code review completed by at least 2 reviewers with all threads resolved.
Complete when: Feature flagged behind config — can be enabled/disabled without deployment.
Complete when: Error tracking configured — all unhandled exceptions routed to on-call.
Complete when: Documentation PR merged — README, API docs, and changelog updated.
Best Practices
Derive UI state from ViewModel via StateFlow<UiState> — Single sealed class per screen (Loading, Success(data), Error(message)). Never expose mutable state directly to Compose — the ViewModel is the single source of truth.
Use rememberSaveable for Compose-local state that must survive process death — remember data is lost on config change; rememberSaveable survives rotation AND process death when backed by Bundle-serializable types. For custom types, implement Saver.
Room DAO functions returning Flow<T> observe database changes reactively — Queries with Flow re-emit on ANY table change. Use distinctUntilChanged() to filter identical results. For one-shot reads, use suspend functions, never LiveData in new code.
WorkManager for deferrable background work, CoroutineWorker over Worker — CoroutineWorker.doWork() is a suspending function; use withContext(Dispatchers.IO) for blocking calls. Periodics have a 15-minute minimum; for shorter intervals use AlarmManager + BroadcastReceiver.
Navigation Compose with type-safe routes: NavController.navigate(route = Home(userId)) — Use Kotlin serialization or @Serializable data classes for route arguments. Never pass complex objects via Bundle; pass ID and resolve from repository.
Enable StrictMode in debug builds with penaltyDeath — StrictMode.setThreadPolicy() catches main-thread disk/network I/O immediately. StrictMode.setVmPolicy() catches leaked SQLite cursors and unclosed Closeable resources. Crash early, find bugs before users do.
ProGuard/R8: always test release builds before Play Store submission — Run ./gradlew bundleRelease and test on a physical device. R8 strips unused code including reflection-based serialization and Retrofit parameter names. Add -keep rules only when a crash confirms a missing class.
Dark theme uses isSystemInDarkTheme() for color scheme selection — Material 3 dynamicColor respects system theme. For manual override, persist preference in DataStore and wrap MaterialTheme with the selected scheme. Never hardcode Color.White or Color.Black as backgrounds.
Compose Previews parameterized per device: @Preview(device = Devices.PIXEL_4_XL) — Generate previews for compact, medium, and expanded layouts. A layout that looks perfect on a Pixel 9 Pro may overflow on a 320dp-wide budget device.
Espresso tests target visible elements by resource ID, not text — onView(withId(R.id.submit_button)) survives localization; onView(withText("Submit")) breaks in 50+ languages. Add testTag for Compose elements: Modifier.testTag("submit_button") → onNodeWithTag("submit_button").
Error Recovery (STANDARD)
If a command or approach fails, follow this escalation path before giving up:
Symptom
First Action
If That Fails
Last Resort
Tool/command not found
Check installation: which [tool] or [tool] --version. Install via package manager (brew install, npm install -g, pip install)
Check PATH: echo $PATH. Verify the tool binary is in a PATH directory. Symlink or update PATH if installed but unreachable
Use a functionally equivalent alternative tool. If rg is unavailable, use grep -r. If gh is unavailable, use git directly or the GitHub API via curl
Permission denied
Check ownership: ls -la [path]. Fix with chmod or sudo if appropriate. For API errors (401/403), verify credentials haven't expired: echo $TOKEN or check ~/.netrc
Refresh credentials: re-authenticate with the service. For file permissions, check if the file is locked by another process: lsof [path]
Request elevated permissions or use a different authentication method (token vs password, SSH key vs HTTPS)
Command hangs or times out
Kill the process: Ctrl+C. Re-run with a timeout: timeout 30 [command] or gtimeout on macOS. Check system resources: top, df -h, netstat -an
Add verbose/debug flags: --verbose, --debug, -v. Check logs: tail -f [logfile]. Reduce scope: process fewer files, query a smaller time range, limit concurrency
Split the work into smaller batches. Implement a retry loop with exponential backoff (1s, 2s, 4s, 8s). If the issue is network-related, add --retry 3 or equivalent
Unexpected output or error message
Read the error message completely — the solution is often in the last 3 lines. Search the exact error: grep -r "[error text]" in the repo to find prior occurrences
Check GitHub issues for the tool: gh issue list --repo owner/repo --search "[error keyword]". Check Stack Overflow
Simplify the approach. If the complex one-liner fails, break it into 3 sequential commands. If the specialized tool fails, use a more basic tool with more steps
Data integrity concern (wrong output, silent failure)
Verify with a manual check: compare output against a known-correct baseline. Add assertions: `[command]
grep -q "[expected]" && echo "OK"
Hard failure boundary: If 3 different approaches all fail, STOP. Do not iterate infinitely. Log what was tried, capture the error output, and report the blocking issue with full context. Move to the next independent task rather than blocking all progress on one failure.
Play Store rejection? → Release Manager → Legal Advisor
Security vulnerability? → Security Engineer → Compliance Officer
Architecture cross-cut with iOS? → Mobile Developer → System Architect
Critical performance regression? → Performance Engineer → CTO Advisor
Accessibility blockers? → Accessibility Auditor → Compliance Officer
Proactive Triggers
These are signals that should trigger the Android developer to investigate — no one needs to tag you.
Trigger
Immediate Action
"ANR rate spiked to 0.5%+ in Play Console"
Android Vitals >0.47% ANR triggers bad-behavior badge, deprioritizing search ranking. Audit main-thread blocking: grep for Thread.sleep(), socket.connect() without timeout, Room queries outside coroutine. Enable StrictMode.detectAll() in debug builds (penaltyDeath() catches violations instantly). Profile with CPU Profiler — look for >5s blocking on main thread. Impact: $20K-$50K in lost organic installs from search rank drop.
"Cold start >2s on low-end devices (Android Vitals)"
20% uninstall rate for apps exceeding 2s cold start. Audit Application.onCreate(): move content providers to AndroidX Startup Initializer<T> with lazy init. Move heavy I/O to Dispatchers.Default. Generate baseline profile via BaselineProfileRule.collect() in Macrobenchmark. Profile with systrace --app=com.example. Impact: $15K-$30K in user churn and uninstalls.
"OutOfMemoryError — bitmap-related crashes in production"
Each OOM crash on a low-RAM device = ~8% session abandonment rate. Audit all image loading: Glide .override(width, height) or Coil .size(width, height) on every load. Profile heap dump: look for byte[] allocations >10MB. Set android:largeHeap="false" (true masks leaks). Downsample to 2× display at most. Impact: $8K-$25K per month in lost ARPU from crash-affected cohorts.
"Play Store listing rejected — policy violation"
Background location requires prominent disclosure BEFORE system dialog + in-app video demo. QUERY_ALL_PACKAGES needs Play Console declaration. MANAGE_EXTERNAL_STORAGE requires file manager/video justification + Google review. Each rejection = 3-7 day re-review delay. Impact: $5K-$15K per rejection cycle in delayed launch revenue.
"Compose screen recomposes 50+ times on simple scroll"
Users spend 40% less time in janky apps. Layout Inspector recomposition counts: unstable lambdas (inline Modifier), non-remembered callbacks, data classes without @Stable. Replace mutableStateOf in ViewModel with StateFlow + collectAsStateWithLifecycle(). Enable composeCompiler { enableStrongSkippingMode = true } (Compose 1.7+). Impact: $15K-$30K in reduced engagement and session time.
1★ "battery drain" reviews are the #2 reason for uninstall (after crashes). Verify PeriodicWorkRequest interval ≥15 min. Constraints MUST include NetworkType.UNMETERED for large transfers. Foreground Service must call stopForeground() within 3 min (Android 14+ limit). Profile with Battery Historian. Impact: $10K-$25K in 1-star review cascade and uninstall rate spike.
"Release APK 3× debug size — R8 not running"
APK size >150MB loses 20% install conversion on cellular. Check isMinifyEnabled = true AND isShrinkResources = true in release. Run APK Analyzer → sort by raw size. resConfigs("en") removes all non-English resources from libraries. Enable android.enableR8.fullMode=true in gradle.properties. Impact: $5K-$10K in lost installs + CDN costs for oversized downloads.
"TalkBack reads 'unlabeled button' on half the UI"
ADA Title III lawsuits settle for $10K-$50K. Play Store pre-launch report flags a11y. Every IconButton, FloatingActionButton, Image without adjacent text needs contentDescription. Run Accessibility Scanner from Play Store. Group elements with semantics(mergeDescendants = true). Impact: $10K-$50K in accessibility lawsuit risk + Play Store pre-launch warnings.
State Log
This skill maintains a decision ledger to prevent context drift and ensure recall across sessions. Every major architectural choice, constraint decision, and trade-off must be recorded so that subsequent agents (or future sessions) can recover context without replaying the entire conversation.
How the State Log Works
On session start: Check .copilot/session-state/decision-ledger.json for any prior decisions relevant to this domain. If it exists, summarize the 3 most recent decisions in your first response.
After each major decision: Append to the ledger:
{
"timestamp": "ISO-8601",
"skill": "android-developer",
"phase": "Phase 3: Implementation",
"decision": "What was decided",
"rationale": "Why this choice over alternatives",
"constraints": ["constraint-1", "constraint-2"],
"alternatives_considered": ["alt-1", "alt-2"],
"reversible": true
}
Before completing work: Verify that all major decisions from this session are recorded. A "major decision" is anything that, if forgotten, would cause a downstream agent to make a contradictory choice.
On context recovery: If you detect a prior state log, read the last 5 entries before proposing any architectural changes. Cite the prior decisions you're building on.
State Log Schema
Field
Purpose
Example
timestamp
When the decision was made
"2026-07-24T21:30:00Z"
skill
Which skill made it
"backend-developer"
phase
Which workflow phase
"Phase 3: API Design"
decision
What was chosen
"PostgreSQL 16 with JSONB for flexible schema"
rationale
Why this over alternatives
"Team expertise + JSONB avoids ORM complexity for semi-structured data"
constraints
What limits apply
["Must support 10K writes/sec", "GDPR data residency: EU only"]
alternatives_considered
What was rejected
["MongoDB (no transactions)", "MySQL 8 (weaker JSON support)"]
reversible
Can this be changed later?
true (migration possible) or false (irreversible choice)
Anti-Drift Check
Before beginning a new phase, verify:
Have I read the state log from the previous session?
Do any prior decisions constrain what I'm about to do?
Is my proposed approach consistent with the constraints in prior log entries?
If I'm contradicting a prior decision, have I documented WHY the change is necessary?
What Good Looks Like
Every Compose screen handles three states (loading/success/error) as a sealed interface with zero blank screens on failure. Every network call is wrapped in a Repository with offline-first caching via Room — users see cached data instantly, network updates arrive async. Every coroutine is scoped to viewModelScope or lifecycleScope — zero GlobalScope instances. Every Image, IconButton, and interactive composable has contentDescription. Cold start <500ms on Galaxy A14 (4GB RAM, eMMC). Scrolls at locked 60fps in LazyColumn with 0 dropped frames in systrace. APK <30MB download.
AndroidManifest.xml has android:supportsRtl="true", layout mirrors correctly in Arabic/Hebrew. Pseudo-localized strings (en-XA) pass without truncation. R8 enabled in release with isMinifyEnabled = true AND isShrinkResources = true. proguard-rules.pro contains zero custom -keep rules — all libraries ship consumer rules. app/build.gradle.kts uses composeBom for version management, KSP instead of kapt. Baseline profiles generated via Macrobenchmark and included in every AAB. Room schema exported with exportSchema = true, migration tests validate every version path against production database snapshots.
Cross-skills Integration
Step
Skill
What it produces
Before
material-design-expert
MD3 compliance audit with md3_checker.py, Dynamic Color strategy (adopt/override/reject), window size class adaptation plan
Before
ui-ux-designer
Material Design 3 design system, screen mockups with a11y annotations, interaction patterns, component specs
This
android-developer
Native Compose implementation: UI, ViewModel, Room, Retrofit, WorkManager, Hilt, Gradle, Play Store AAB
Install on a $150 Galaxy A14 — 4GB RAM, eMMC storage. Use as daily driver for one day.
Find every friction point — Slow startup? Jank? OOM? ANR? Permission denied?
Profile and fix — CPU Profiler, Memory Profiler, Battery Historian. Fix the worst offender.
Repeat monthly with a different budget device. Your app behaves differently on all of them.
Practice Routines
Skill Level
Practice
Frequency
Expected Result
Novice → Competent
Build same screen in XML + Compose + Compose MVI. Compare line count, testability, recomposition.
Monthly
Understands when Compose improves vs adds complexity
Competent → Expert
Enable StrictMode with penaltyDeath(). Fix every violation: disk reads, network, leaked closeables.
Quarterly
App verified clean by platform tooling
Expert → Master
Build feature with only Android framework (no Jetpack, no Retrofit). Then rebuild with Jetpack.
Annually
Understands what Jetpack abstracts and where the real cost is
The One Thing
Preload your app on Android Go edition (1-2GB RAM). If it launches under 3s and doesn't crash scrolling 100 items, your architecture is solid.
Anti-Patterns
Memory leak from retained Fragment View Binding ($15K-$40K)._binding not nulled in onDestroyView() retains old view references across config changes. Multiply by navigation across 5 screens and you've got an OOM chain on 2GB devices. Fix: private var _binding: FragmentXBinding? = null; null in onDestroyView(); use binding property with checkNotNull guard.
WorkManager periodic runs from start, not completion ($10K-$30K).PeriodicWorkRequest(15, MINUTES) starts the 15-min clock at task start. If task takes 10 min, next run is 5 min later. If 20 min, next runs immediately. Produces battery drain and Play Store policy violations. Fix: Use OneTimeWorkRequest that enqueues next at end of doWork() for gap-from-completion scheduling.
R8 strips Retrofit suspend function parameters ($15K-$40K). R8 removes unused params from @GET suspend functions, changing the Retrofit proxy signature. Release-only crash: IllegalArgumentException: Wrong number of arguments — debug builds work perfectly. Fix: Add -keepclasseswithmembers class * { @retrofit2.http.* <methods>; } or annotate interface methods with @Keep. ALWAYS test release builds before Play Store submission.
MutableStateFlow + data class stale recomposition ($5K-$15K). Same data class value emitted on MutableStateFlow → Compose uses structural equality (equals()) and skips recomposition. Pull-to-refresh shows stale data — "refresh doesn't work" bugs that defy debugging. Fix: Add refreshId: Long = System.currentTimeMillis() to UiState for explicit recomposition triggers, or use SnapshotMutationPolicy with referentialEqualityPolicy().
Lost upload keystore — app permanently unpublishable ($25K-$100K+). Play App Signing uses two keys: your upload key (signing AABs submitted to Play) and Google's app signing key (signing distributed APK). Lose the upload key = can never update the app. Google cannot recover it. The listing must be unpublished, a new package name created, losing all installs, ratings, and Play Store search rank. Fix: Store upload keystore in password manager + encrypted backup. Enroll in Play App Signing immediately. Verify SHA-1 fingerprint matches Play Console before first release.
Room migration missing — every updater crashes ($20K-$50K). New @Entity column without Migration(start, end) → IllegalStateException: A migration from X to Y was required but not found on database open. Every user updating from the previous version crashes on launch — crash rate spikes to 100% of updaters. Fix: Every schema change needs Migration(oldVersion, newVersion) { database.execSQL("ALTER TABLE ...") }. Export schema JSON with exportSchema = true and commit to VCS. Test migrations against production database snapshots.
ANR from main-thread network I/O ($30K-$100K).URL.readText() or socket.connect() on main thread triggers "App isn't responding. Wait / Close" dialog at 5 seconds. Android Vitals dashboard tracks ANR rate; >0.47% triggers bad-behavior badge that deprioritizes your app in search results. Fix: Enable StrictMode.setThreadPolicy(ThreadPolicy.Builder().detectAll().penaltyDeath().build()) in debug builds — it crashes INSTANTLY on violation so you catch it in development. Move all I/O to withContext(Dispatchers.IO) { }.
Compose recomposition from unstable parameters ($30K+ in lost engagement). Passing a data class from an external library (e.g., java.time.Instant) to a @Composable function causes recomposition every frame — the Compose compiler marks external types as unstable. Layout Inspector shows recomposition count incrementing on every frame even with static data. Fix: Annotate data classes with @Stable, wrap external types in stable wrappers, or enable Strong Skipping Mode: composeCompiler { enableStrongSkippingMode = true } (Compose 1.7+). Check compiler metrics in build/compose-metrics/ for unstable classes.
Play Store rejection for background location without prominent disclosure ($5K-$15K per rejection). Google requires a prominent disclosure dialog BEFORE the system permission dialog explaining WHY background location is essential. Must submit a video demonstrating the feature. Include android:foregroundServiceType="location" in the service declaration. Without this: "Prominent disclosure not provided" rejection. Fix: Implement a custom dialog that appears before requestPermissions(), explaining the critical use case. Record the flow for Play Console submission. Test on internal track before production submission.
Configuration change destroys UI state — no SavedStateHandle ($20K+). Rotating the device or enabling "Don't keep activities" in Developer Options destroys and recreates the Activity. If state is held in the Activity instead of ViewModel + SavedStateHandle, form inputs, scroll position, and navigation state are lost. android:configChanges="orientation|screenSize" is NOT a fix — it breaks multi-window, dark theme switching, and locale changes. Fix: All UI state in ViewModel + SavedStateHandle. Use rememberSaveable for Compose-local state. Test by rotating device mid-input on every form screen.
Anti-Hallucination
Rationalization
Reality
"It works on my Pixel 9 Pro — we'll test budget devices after the feature freeze."
95% of global Android users have mid-range or budget devices. A crash on a Galaxy A14 isn't a bug — it's a 1-star review. Budget device is primary test target.
"I'll add error handling after the happy path — let's not slow down for edge cases."
A blank screen on network failure looks like a crash. Every screen without error states silently fails in production. Error handling isn't polish — it's the difference between professional and prototype.
"We don't need TalkBack — user research didn't flag accessibility."
15% of population has a disability. Play Store pre-launch report flags a11y. contentDescription is 40 characters. A 1-star accessibility review is forever.
"API key in local.properties isn't committed, so it's safe."
BuildConfig fields are compiled into the APK in plaintext. strings app.apk extracts them in seconds. Use Keystore + server proxy.
"Hilt adds 15% build time — manual DI is simpler."
Manual DI's cost is runtime: missing dependency crashes, no scope enforcement, no compile-time verification. One null-pointer crash costs more than Hilt's build overhead saves in a year.
Error Decoder — War Stories from the Trenches
(STANDARD)
When Android goes wrong, it goes wrong in predictable ways. Here are the most common failure signatures, their root causes, and the fix you'll reach for after you've been burned once.
Symptom
Root Cause
Fix
Lesson
App freezes for 5+ seconds, then ANR dialog appears — Input dispatching timed out
Room database query or network call running on the main thread. The UI thread is blocked waiting for I/O that takes 2+ seconds
Move all DB queries to Dispatchers.IO inside viewModelScope.launch. Use Flow to observe Room queries reactively. Set StrictMode.setThreadPolicy(detectAll().penaltyDeath()) in debug builds
Android's 5-second ANR threshold is merciless. A single synchronous DB query on the main thread kills your Play Store rating. StrictMode catches these in development before users do
FragmentManager crash: Can not perform this action after onSaveInstanceState — appears in crash reports, never reproduces in development
FragmentTransaction.commit() called after the Activity has saved its state (user pressed Home, received phone call, or OS killed process). The transaction is lost because the state was already serialized
Use commitAllowingStateLoss() only if you truly don't care about state loss. Otherwise, check isStateSaved before committing. Better: use Navigation Component which handles this internally
Android's lifecycle is asynchronous and OS-driven. onSaveInstanceState fires at unpredictable times. Any UI state change after that point is silently discarded
ProGuard/R8 strips model classes — app crashes with JsonSyntaxException or ClassNotFoundException in release build only
R8 obfuscation removes fields and classes that Gson/Moshi/Retrofit discover via reflection. Debug build works perfectly because minification is disabled
Add @Keep annotation on all API model classes and their fields. Configure -keep rules in proguard-rules.pro for every reflection-based library. Always test the release build on a physical device before shipping
The release build is a different binary than the debug build. R8 can rename, remove, or inline anything not explicitly kept. Every reflection-dependent library is a ticking time bomb
Compose UI state resets after screen rotation — remember loses form input, scroll position, and selected tab
remember {} survives recomposition but NOT configuration changes or process death. Rotation destroys and recreates the Activity, wiping all remember-ed state
Use rememberSaveable {} for any state that should survive configuration changes. For complex objects, provide a custom Saver. Use SavedStateHandle in ViewModels for process-death survival
remember and rememberSaveable look identical but have fundamentally different lifecycles. The compiler won't warn you — the bug only surfaces when users rotate their phone
Image loading causes OOM crash on budget devices — BitmapFactory throws OutOfMemoryError on Galaxy A-series phones
Loading a full-resolution 12MP camera image into a Bitmap allocates ~48MB of contiguous memory. Budget devices have limited heap and fragmented memory
Use BitmapFactory.Options.inSampleSize to downsample by 4-8x. Prefer Coil or Glide which handle downsampling, caching, and memory management automatically. Always set android:largeHeap="true" only as last resort
A 12MP image decoded at full resolution consumes more memory than most budget devices can allocate in a single block. Image loading libraries earn their dependency weight 100x over
WorkManager job runs on every app launch instead of once daily — battery drain complaints in Play Store reviews
PeriodicWorkRequest with ExistingPeriodicWorkPolicy.REPLACE enqueued in Application.onCreate(). Every cold start replaces the pending work, resetting the timer
Use ExistingPeriodicWorkPolicy.KEEP to preserve the existing schedule. Enqueue periodic work once in Application.onCreate() with KEEP policy. Use WorkManager.enqueueUniqueWork() for one-time work
REPLACE is the default enum value developers reach for because it compiles. Read the policy docs — KEEP, REPLACE, and APPEND have dramatically different behaviors for periodic work
Gotchas
Gotcha
Cost
Fix
ProGuard/R8 strips reflection-based classes — release build crashes, debug works fine
$15K-$45K in emergency hotfixes
Add @Keep on all serialization models, configure -keep rules for every reflection library, always test release build on physical device before shipping
remember loses state on rotation — form inputs, scroll position wiped on config change
$10K-$30K in user frustration
Use rememberSaveable for state surviving config changes; provide custom Saver for complex objects; use SavedStateHandle in ViewModels
Image loading OOM on budget devices — 12MP image decoded at full resolution allocates ~48MB
$20K-$50K in 1-star reviews
Use Coil/Glide for automatic downsampling, set inSampleSize for BitmapFactory, never load full-resolution images; budget devices dominate global Android market
Verification
./gradlew test — all unit tests pass, no regressions
./gradlew connectedCheck — Espresso/Compose tests pass on target device
./gradlew bundleRelease — zero ProGuard/R8 warnings, no missing keep rules
Play App Signing enrolled; upload keystore fingerprint confirmed in Play Console
Play Store Data Safety form completed and accurate
Firebase Crashlytics or equivalent crash reporter configured for release builds
Per-app language preferences tested if supporting multiple locales
Foreground service permissions properly declared with foregroundServiceType
Verification Guardrails
Before delivering work, verify: self-check against What Good Looks Like, no broken references, continuity with State Log, no fabricated APIs/versions/capabilities, Error Recovery paths exercised, cross-skill dependencies satisfied. If any fail, revise before delivering.
References
Detailed reference material loaded on demand:
Jetpack Compose Patterns: See references/jetpack-compose-patterns.md — 20 patterns: state hoisting, ViewModel+StateFlow, sealed UiState, Scaffold, LazyColumn keys, side effects (LaunchedEffect, DisposableEffect, SideEffect), AndroidView interop, Material 3 theming with dynamic color, type-safe Navigation Compose, recomposition optimization
Android Architecture Patterns: See references/android-architecture-patterns.md — MVVM with ViewModel+StateFlow+sealed UiState, MVI with Intent+State+Effect Channel, Clean Architecture (domain/data/presentation layers), UseCase pattern, Repository with offline-first caching, DI with Hilt, SavedStateHandle for process death
Kotlin Coroutines & Flow: See references/kotlin-coroutines-flow.md — Dispatcher selection (Main/IO/Default/Unconfined), structured concurrency, Flow vs StateFlow vs SharedFlow vs Channel, cold vs hot streams, exception handling (try/catch, catch operator, CoroutineExceptionHandler), cancellation, callbackFlow, flowOn, testing with runTest and Turbine
Room Database Guide: See references/room-database-guide.md — Entity, DAO, Database, embedded types, type converters, FTS (full-text search), SQL queries (raw and generated), multi-table JOINs and @Relation, migrations (Migration, AutoMigration, destructive fallback), exportSchema, testing with in-memory database
Play Store Deployment: See references/play-store-deployment.md — Play App Signing (upload key vs app signing key), AAB generation, internal/alpha/beta/production tracks, staged rollout (10%→50%→100%), in-app review API, Billing 7.x subscriptions/one-time, pre-launch report, policy compliance, Data safety form, IARC content rating, managed publishing
Android Accessibility: See references/android-accessibility.md — TalkBack navigation, contentDescription best practices, Compose semantics (semantics, mergeDescendants, invisibleToUser), touch target minimums (48dp), color contrast (4.5:1 text, 3:1 large text), focus order, live regions (AccessibilityLiveRegion), testing with Accessibility Scanner, WCAG 2.2 AA mapped to Android
