SKILL.md
Kotlin Multiplatform: Platform Abstraction Decisions
Expert guidance for KMP architecture in Amethyst - deciding what to share vs keep platform-specific.
When to Use This Skill
Making platform abstraction decisions:
- "Should I create expect/actual or keep Android-only?"
- "Can I share this ViewModel logic?"
- "Where does this crypto/JSON/network implementation belong?"
- "This uses Android Context - can it be abstracted?"
- "Is this code in the wrong module?"
- Preparing for iOS/web/wasm targets
- Detecting incorrect placements
Abstraction Decision Tree
Central question: "Should this code be reused across platforms?"
Follow this decision path (< 1 minute):
Q: Is it used by 2+ platforms?
├─ NO → Keep platform-specific
│ Example: Android-only permission handling
│
└─ YES → Continue ↓
Q: Is it pure Kotlin (no platform APIs)?
├─ YES → commonMain
│ Example: Nostr event parsing, business rules
│
└─ NO → Continue ↓
Q: Does it vary by platform or by JVM vs non-JVM?
├─ By platform (Android ≠ iOS ≠ Desktop)
│ → expect/actual
│ Example: Secp256k1Instance (uses different security APIs)
│
├─ By JVM (Android = Desktop ≠ iOS/web)
│ → jvmAndroid
│ Example: Jackson JSON parsing (JVM library)
│
└─ Complex/UI-related
→ Keep platform-specific
Example: Navigation (Activity vs Window too different)
Final check:
Q: Maintenance cost of abstraction < duplication cost?
├─ YES → Proceed with abstraction
└─ NO → Duplicate (simpler)
Real Examples from Codebase
Crypto → expect/actual:
// commonMain - expect declaration
expect object Secp256k1Instance {
fun signSchnorr(data: ByteArray, privKey: ByteArray): ByteArray
}
// androidMain - uses Android Keystore
// jvmMain - uses Desktop JVM crypto
// iosMain - uses iOS Security framework
Why: Each platform has different security APIs.
JSON parsing → jvmAndroid:
// quartz/build.gradle.kts
val jvmAndroid = create("jvmAndroid") {
api(libs.jackson.module.kotlin)
}
Why: Jackson is JVM-only, works on Android + Desktop, not iOS/web.
Navigation → platform-specific:
- Android:
MainActivity(Activity + Compose Navigation)
- Desktop:
Window+ sidebar + MenuBar
Why: UI paradigms fundamentally different.
Mental Model: Source Sets as Dependency Graph
Think of source sets as a dependency graph, not folders.
┌─────────────────────────────────────────────┐
│ commonMain = Contract (pure Kotlin) │
│ - Business logic, protocol, data models │
│ - No platform APIs │
└────────────┬────────────────────────────────┘
│
├──────────────────────┬────────────────────
│ │
▼ ▼
┌───────────────────┐ ┌──────────────────┐
│ jvmAndroid │ │ iosMain │
│ JVM libs shared │ │ iOS common │
│ - Jackson │ │ │
│ - OkHttp │ └────┬─────────────┘
└───┬───────────┬───┘ │
│ │ │
▼ ▼ ├─→ iosArm64Main
┌─────────┐ ┌──────────┐ └─→ iosSimulatorArm64Main
│android │ │jvmMain │
│Main │ │(Desktop) │
└─────────┘ └──────────┘
Future: jsMain, wasmMain
Key insight: jvmAndroid is NOT a platform - it's a shared JVM layer.
The jvmAndroid Pattern
Unique to Amethyst. Shares JVM libraries between Android + Desktop.
When to Use jvmAndroid
Use jvmAndroid when:
- ✅ JVM-specific libraries (Jackson, OkHttp, url-detector)
- ✅ Android implementation = Desktop implementation (same JVM)
- ✅ Library doesn't work on iOS/web
Do NOT use jvmAndroid for:
- ❌ Pure Kotlin code (use commonMain)
- ❌ Platform-specific APIs (use androidMain/jvmMain)
- ❌ Code that should work on all platforms
Example from quartz/build.gradle.kts
// Must be defined BEFORE androidMain and jvmMain
val jvmAndroid = create("jvmAndroid") {
dependsOn(commonMain.get())
dependencies {
api(libs.jackson.module.kotlin) // JSON parsing - JVM only
api(libs.url.detector) // URL extraction - JVM only
implementation(libs.okhttp) // HTTP client - JVM only
}
}
// Both depend on jvmAndroid
jvmMain { dependsOn(jvmAndroid) }
androidMain { dependsOn(jvmAndroid) }
Why Jackson in jvmAndroid, not commonMain?
- Jackson is JVM-specific library
- Works on Android (runs on JVM)
- Works on Desktop (runs on JVM)
- Does NOT work on iOS (not JVM) or web (not JVM)
Web/wasm consideration: For future web support, consider migrating from Jackson → kotlinx.serialization (see Target-Specific Guidance).
What to Abstract vs Keep Platform-Specific
Quick decision guidelines based on codebase patterns:
Always Abstract
- Crypto (Secp256k1, encryption, signing)
- Core protocol logic (Nostr events, NIPs)
- Why: Needed everywhere, platform security APIs vary
Often Abstract
- I/O operations (file reading, caching)
- Logging (platform logging systems differ)
- Serialization (if using kotlinx.serialization)
- Why: Commonly reused, platform implementations available
Sometimes Abstract
- Business logic: YES - state machines, data processing
- ViewModels: YES - state + business logic shareable (StateFlow/SharedFlow)
- Screen layouts: NO - platform-native (Window vs Activity)
- Why: ViewModels contain platform-agnostic state; Screens render differently per platform
Rarely Abstract
- Complex UI components (composables with heavy platform dependencies)
- Why: Platform paradigms can differ significantly
Never Abstract
- Navigation (Activity vs Window fundamentally different)
- Permissions (Android vs iOS APIs incompatible)
- Platform UX patterns
- Why: Too platform-specific, abstraction creates leaky APIs
Evidence from shared-ui-analysis.md
Component
Shared?
Rationale
PubKeyFormatter, ZapFormatter
✅ YES
Pure Kotlin, no platform APIs
TimeAgoFormatter
⚠️ ABSTRACTED
Needs StringProvider for localized strings
ViewModels (state + logic)
✅ YES
StateFlow/SharedFlow platform-agnostic, Compose Multiplatform lifecycle compatible
Screen layouts (Scaffold, nav)
❌ NO
Window vs Activity, sidebar vs bottom nav fundamentally different
Image loading (Coil)
⚠️ ABSTRACTED
Coil 3.x supports KMP, needs expect/actual wrapper
expect/actual Mechanics
When to use: Code needed by 2+ platforms, varies by platform.
Pattern Categories from Codebase
Objects (singletons):
// 24 expect declarations found, common pattern:
expect object Secp256k1Instance { ... }
expect object Log { ... }
expect object LibSodiumInstance { ... }
Classes (instantiable):
expect class AESCBC { ... }
expect class DigestInstance { ... }
Functions (utilities):
expect fun platform(): String
expect fun currentTimeSeconds(): Long
See references/expect-actual-catalog.md for complete catalog with rationale.
Target-Specific Guidance
Android, JVM (Desktop), iOS - Current Primary Targets
Status: Mature patterns, stable APIs
Android (androidMain):
- Uses Android framework (Activity, Context, etc.)
- secp256k1-kmp-jni-android (
0.23.0inlibs.versions.toml) for crypto
- AndroidX libraries
Desktop JVM (jvmMain):
- Uses Compose Desktop (Window, MenuBar, etc.)
- secp256k1-kmp-jni-jvm (same
0.23.0line) for crypto
- Pure JVM libraries
iOS (iosMain):
- Mature target — actively built and tested
- Architecture targets: iosArm64, iosSimulatorArm64, iosX64 (plus macosArm64 for host tooling)
- Platform APIs via platform.posix, Security framework
Web, wasm - Future Targets
Status: Not yet implemented, consider for future-proofing
Constraints to know:
- ❌ No platform.posix (file I/O different)
- ❌ No JVM libraries (Jackson, OkHttp won't work)
- ❌ Different async model (JS event loop vs threads)
Future-proofing tips:
- Prefer pure Kotlin in commonMain
- Use kotlinx.* libraries:
- kotlinx.serialization instead of Jackson
- ktor instead of OkHttp (ktor supports web)
- kotlinx.datetime instead of custom date handling
- Avoid platform.posix for file operations
- Test abstractions work without JVM assumptions
Example migration path:
// Current: jvmAndroid (JVM-only)
api(libs.jackson.module.kotlin)
// Future: commonMain (all platforms)
api(libs.kotlinx.serialization.json)
Integration: When to Invoke Other Skills
Invoke gradle-expert
Trigger gradle-expert skill when encountering:
- Dependency conflicts (e.g., secp256k1-android vs secp256k1-jvm version mismatch)
- Build errors related to source sets
- Version catalog issues (libs.versions.toml)
- "Duplicate class" errors
- Performance/build time issues
Example trigger:
Error: Duplicate class found: fr.acinq.secp256k1.Secp256k1
→ Invoke gradle-expert for dependency conflict resolution.
Flags to Raise
Platform code in commonMain:
// ❌ INCORRECT - Android API in commonMain
expect fun getContext(): Context // Context is Android-only!
→ Flag: "Android API in commonMain won't compile on other platforms"
Duplicated business logic:
// ❌ INCORRECT - Same logic in both
// androidMain/.../CryptoUtils.kt
fun validateSignature(...) { ... }
// jvmMain/.../CryptoUtils.kt
fun validateSignature(...) { ... } // Duplicated!
→ Flag: "Business logic duplicated, should be in commonMain or expect/actual"
Reinventing wheel - suggest KMP alternatives:
- Custom date/time → kotlinx.datetime
- OkHttp → ktor (supports web)
- Jackson → kotlinx.serialization
- Custom UUID → kotlinx.uuid (when stable)
Common Pitfalls
1. Over-Abstraction
Problem: Creating expect/actual for UI components
// ❌ BAD
expect fun NavigationComponent(...)
Why: Navigation paradigms too different (Activity vs Window)
Fix: Keep platform-specific, accept duplication
2. Under-Sharing
Problem: Duplicating business logic across platforms
// ❌ BAD - duplicated in androidMain and jvmMain
fun parseNostrEvent(json: String): Event { ... }
Why: Bug fixes need to be applied twice, tests duplicated
Fix: Move to commonMain (pure Kotlin) or create expect/actual
3. Leaky Abstractions
Problem: Platform code in commonMain
// commonMain - ❌ BAD
import android.content.Context // Won't compile on iOS!
Fix: Use expect/actual or dependency injection
4. Premature Abstraction
Problem: Creating expect/actual before second platform needs it
// ❌ BAD - only used on Android currently
expect fun showNotification(...)
Why: Wrong abstraction boundaries, wasted effort
Fix: Wait until iOS actually needs it, then abstract
5. Wrong Source Set
Problem: JVM libraries in commonMain
// commonMain - ❌ BAD
import com.fasterxml.jackson.databind.ObjectMapper
Why: Jackson won't compile on iOS/web
Fix: Move to jvmAndroid or migrate to kotlinx.serialization
Quick Reference
Code Type
Recommended Location
Reason
Pure Kotlin business logic
commonMain
Works everywhere
Nostr protocol, NIPs
commonMain
Core logic, no platform APIs
JVM libs (Jackson, OkHttp)
jvmAndroid
Android + Desktop only
Crypto (varies by platform)
expect in commonMain, actual in platforms
Different security APIs per platform
I/O, logging
expect in commonMain, actual in platforms
Platform implementations differ
State (business logic)
commonMain or commons/jvmAndroid
Reusable StateFlow patterns
ViewModels
commons/commonMain/viewmodels/
StateFlow/SharedFlow + logic shareable, Compose MP lifecycle compatible
UI formatters (pure)
commons/commonMain
Reusable, no dependencies
UI components (simple)
commons/commonMain
Cards, buttons, dialogs
Screen layouts
Platform-specific
Window vs Activity, sidebar vs bottom nav
Navigation
Platform-specific only
Activity vs Window too different
Permissions
Platform-specific only
APIs incompatible
Platform UX (menus, etc.)
Platform-specific only
Native feel required
See Also
- references/abstraction-examples.md - Good/bad abstraction examples with rationale
- references/source-set-hierarchy.md - Visual hierarchy with Amethyst examples
- references/expect-actual-catalog.md - All 24 expect/actual pairs with "why abstracted"
- references/target-compatibility.md - Platform constraints and future-proofing
Scripts
scripts/validate-kmp-structure.sh- Detect incorrect placements, validate source sets
scripts/suggest-kmp-dependency.sh- Suggest KMP library alternatives (ktor, kotlinx.serialization, etc.)