Agent Skills: Kotlin Expert

Kotlin Expert

Advanced Kotlin patterns for AmethystMultiplatform. Covers Flow state management, sealed hierarchies, immutability, DSL builders, and inline functions with real codebase examples.

Mental Model

Kotlin in Amethyst:

State Management (Hot Flows)
    ├── StateFlow<T>           # Single value, always has value, replays to new subscribers
    ├── SharedFlow<T>          # Event stream, configurable replay, multiple subscribers
    └── MutableStateFlow<T>    # Private mutable, public via .asStateFlow()

Type Safety (Sealed Hierarchies)
    ├── sealed class           # State variants with data (AccountState.LoggedIn/LoggedOut)
    └── sealed interface       # Generic result types (SignerResult<T>)

Compose Performance (@Immutable)
    ├── @Immutable             # 173+ event classes - prevents recomposition
    └── data class             # Structural equality, copy(), immutable by convention

DSL Patterns
    ├── Builder classes        # Fluent APIs (TagArrayBuilder)
    ├── Lambda receivers       # inline fun tagArray { ... }
    └── Method chaining        # return this

Performance
    ├── inline fun             # Eliminate lambda overhead
    ├── reified type params    # Runtime type info (OptimizedJsonMapper)
    └── value class            # Zero-cost wrappers (NOT USED yet in Amethyst)

Delegation:

• kotlin-coroutines agent: Deep async (structured concurrency, channels, operators)
• kotlin-multiplatform skill: expect/actual, source sets
• This skill: Amethyst Kotlin idioms, state patterns, type safety

1. Flow State Management

StateFlow: State that Changes

Mental model: StateFlow is a "hot" observable state holder. Always has a value, new collectors immediately get current state.

Amethyst pattern:

// AccountManager.kt:48-50
class AccountManager {
    private val _accountState = MutableStateFlow<AccountState>(AccountState.LoggedOut)
    val accountState: StateFlow<AccountState> = _accountState.asStateFlow()

    fun login(key: String) {
        _accountState.value = AccountState.LoggedIn(...)
    }
}

Key principles:

1. Private mutable, public immutable: _accountState (MutableStateFlow) private, accountState (StateFlow) public
2. Always has value: Initial value required (LoggedOut)
3. Single value: Replays ONE most recent value to new subscribers
4. Hot: Stays in memory, all collectors share same instance

See: AccountManager.kt:48-50, RelayConnectionManager.kt:49-52

SharedFlow: Event Streams

Mental model: SharedFlow is a "hot" broadcast stream for events. Configurable replay buffer, doesn't require initial value.

Amethyst pattern:

// RelayConnectionManager.kt:52-53
val connectedRelays: StateFlow<Set<NormalizedRelayUrl>> = client.connectedRelaysFlow()
val availableRelays: StateFlow<Set<NormalizedRelayUrl>> = client.availableRelaysFlow()

When to use StateFlow vs SharedFlow:

| Scenario | Use StateFlow | Use SharedFlow | |----------|---------------|----------------| | UI state | ✅ Current screen data, login status | ❌ | | One-time events | ❌ | ✅ Navigation, snackbars, toasts | | Always has value | ✅ | ❌ Optional | | Replay count | 1 (latest only) | Configurable (0, 1, n) | | Backpressure | Conflates (drops old) | Configurable buffer |

Best practice:

// State: Use StateFlow
private val _uiState = MutableStateFlow(UiState.Loading)
val uiState: StateFlow<UiState> = _uiState.asStateFlow()

// Events: Use SharedFlow
private val _navigationEvents = MutableSharedFlow<NavEvent>(replay = 0)
val navigationEvents: SharedFlow<NavEvent> = _navigationEvents.asSharedFlow()

Flow Anti-Patterns

❌ Exposing mutable state:

val accountState: MutableStateFlow<AccountState>  // BAD: Can be mutated externally

✅ Expose immutable:

val accountState: StateFlow<AccountState> = _accountState.asStateFlow()  // GOOD

❌ SharedFlow for state:

val loginState = MutableSharedFlow<LoginState>()  // BAD: State might get lost

✅ StateFlow for state:

val loginState = MutableStateFlow(LoginState.LoggedOut)  // GOOD: Always has value

See: references/flow-patterns.md for comprehensive examples.

2. Sealed Hierarchies

Sealed Classes: State Variants

Mental model: Sealed classes represent a closed set of variants that share common data/behavior.

Amethyst pattern:

// AccountManager.kt:36-46
sealed class AccountState {
    data object LoggedOut : AccountState()

    data class LoggedIn(
        val signer: NostrSigner,
        val pubKeyHex: String,
        val npub: String,
        val nsec: String?,
        val isReadOnly: Boolean
    ) : AccountState()
}

// Usage
when (state) {
    is AccountState.LoggedOut -> showLogin()
    is AccountState.LoggedIn -> showFeed(state.pubKeyHex)
}  // Exhaustive - compiler enforces all cases

Key principles:

1. Closed hierarchy: All subclasses known at compile-time
2. Exhaustive when: Compiler ensures all cases handled
3. Shared data: Sealed class can hold common properties
4. Single inheritance: Subclass can't extend another class

When to use:

• Modeling UI states (Loading, Success, Error)
• Login states (LoggedOut, LoggedIn)
• Result types with different data per variant

Sealed Interfaces: Generic Result Types

Mental model: Sealed interfaces for contracts with multiple implementations that need generics or multiple inheritance.

Amethyst pattern:

// SignerResult.kt:25-46
sealed interface SignerResult<T : IResult> {
    sealed interface RequestAddressed<T : IResult> : SignerResult<T> {
        class Successful<T : IResult>(val result: T) : RequestAddressed<T>
        class Rejected<T : IResult> : RequestAddressed<T>
        class TimedOut<T : IResult> : RequestAddressed<T>
        class ReceivedButCouldNotPerform<T : IResult>(
            val message: String?
        ) : RequestAddressed<T>
    }
}

// Usage with generics
fun handleResult(result: SignerResult<SignResult>) {
    when (result) {
        is SignerResult.RequestAddressed.Successful -> processEvent(result.result.event)
        is SignerResult.RequestAddressed.Rejected -> showRejected()
        is SignerResult.RequestAddressed.TimedOut -> showTimeout()
    }
}

Key principles:

1. Multiple inheritance: Subtype can implement other interfaces
2. Variance: Supports out/in modifiers for generics
3. No constructor: Can't hold state directly (subtypes can)
4. Nested hierarchies: Can create sub-sealed hierarchies

Sealed Class vs Sealed Interface

| Feature | Sealed Class | Sealed Interface | |---------|--------------|------------------| | Constructor | ✅ Can hold common state | ❌ No constructor | | Inheritance | ❌ Single parent only | ✅ Multiple interfaces | | Generics | ❌ No variance | ✅ Covariance/contravariance | | Use case | State variants | Result types, contracts |

Decision tree:

Need to hold common data in base?
    YES → sealed class
    NO → sealed interface

Need generics with variance (out/in)?
    YES → sealed interface
    NO → Either works

Subtypes need multiple inheritance?
    YES → sealed interface
    NO → Either works

Amethyst examples:

• sealed class AccountState - state variants with different data
• sealed interface SignerResult<T> - generic result types with variance

See: references/sealed-class-catalog.md for all sealed types in quartz.

3. Immutability & Compose Performance

@Immutable Annotation

Mental model: @Immutable tells Compose "this value never changes after construction." Compose can skip recomposition if @Immutable object reference doesn't change.

Amethyst pattern:

// TextNoteEvent.kt:51-63
@Immutable
class TextNoteEvent(
    id: HexKey,
    pubKey: HexKey,
    createdAt: Long,
    tags: Array<Array<String>>,
    content: String,
    sig: HexKey
) : BaseThreadedEvent(id, pubKey, createdAt, KIND, tags, content, sig) {
    // All properties immutable (val), no mutable state
}

Key principles:

1. All properties immutable: Only val, never var
2. No mutable collections: Use ImmutableList, Array, not MutableList
3. Deep immutability: Nested objects also immutable
4. Compose optimization: Skips recomposition if reference equals

Why it matters:

// Without @Immutable
@Composable
fun NoteCard(note: TextNoteEvent) {  // Recomposes every time parent recomposes
    Text(note.content)
}

// With @Immutable
@Composable
fun NoteCard(note: TextNoteEvent) {  // Only recomposes if note reference changes
    Text(note.content)
}

173+ @Immutable classes in quartz - all events immutable for Compose performance.

Data Classes & Immutability

Pattern:

@Immutable
data class RelayStatus(
    val url: NormalizedRelayUrl,
    val connected: Boolean,
    val error: String? = null
) {
    // Implicit: equals(), hashCode(), copy(), toString()
}

// Usage
val oldStatus = RelayStatus(url, connected = false)
val newStatus = oldStatus.copy(connected = true)  // Immutable update

Key principles:

1. Structural equality: equals() compares properties, not reference
2. copy(): Create modified copies without mutating
3. All properties in constructor: For proper equals()/hashCode()
4. Prefer val: Make properties immutable

kotlinx.collections.immutable

Pattern:

import kotlinx.collections.immutable.ImmutableList
import kotlinx.collections.immutable.persistentListOf
import kotlinx.collections.immutable.toImmutableList

// Instead of List (which could be mutable internally)
val relays: ImmutableList<String> = persistentListOf("wss://relay1.com", "wss://relay2.com")

// Add returns new instance
val updated = relays.add("wss://relay3.com")  // relays unchanged, updated has 3 items

When to use:

• Compose state that needs collection
• Publicly exposed collections
• Shared state across threads

See: references/immutability-patterns.md

4. DSL Builders

Type-Safe Fluent APIs

Mental model: DSL (Domain-Specific Language) builders use lambda receivers and method chaining to create readable, type-safe APIs.

Amethyst pattern:

// TagArrayBuilder.kt:23-90
class TagArrayBuilder<T : IEvent> {
    private val tagList = mutableMapOf<String, MutableList<Tag>>()

    fun add(tag: Array<String>): TagArrayBuilder<T> {
        if (tag.isEmpty() || tag[0].isEmpty()) return this
        tagList.getOrPut(tag[0], ::mutableListOf).add(tag)
        return this  // Method chaining
    }

    fun remove(tagName: String): TagArrayBuilder<T> {
        tagList.remove(tagName)
        return this  // Method chaining
    }

    fun build() = tagList.flatMap { it.value }.toTypedArray()
}

// Inline function with lambda receiver (line 90)
inline fun <T : Event> tagArray(initializer: TagArrayBuilder<T>.() -> Unit = {}): TagArray =
    TagArrayBuilder<T>().apply(initializer).build()

Usage:

val tags = tagArray<TextNoteEvent> {
    add(arrayOf("e", eventId, relay, "reply"))
    add(arrayOf("p", pubkey))
    remove("a")  // Remove address tags
}

Key patterns:

1. Method chaining: Return this from mutator methods
2. Lambda receiver: TagArrayBuilder<T>.() -> Unit - lambda has this: TagArrayBuilder<T>
3. inline function: Eliminates lambda overhead
4. apply(): Executes lambda with receiver, returns receiver

DSL Pattern Template

class MyBuilder {
    private val items = mutableListOf<Item>()

    fun add(item: Item): MyBuilder {
        items.add(item)
        return this
    }

    fun build(): Result = Result(items.toList())
}

inline fun myDsl(init: MyBuilder.() -> Unit): Result =
    MyBuilder().apply(init).build()

// Usage
val result = myDsl {
    add(Item("foo"))
    add(Item("bar"))
}

Why inline?

• Eliminates lambda object allocation
• Enables reified type parameters
• Better performance for frequently-called DSLs

See: references/dsl-builder-examples.md for more patterns.

5. Inline Functions & reified

inline fun: Eliminate Overhead

Mental model: inline copies function body to call site. No lambda object created, direct code insertion.

Pattern:

// Without inline
fun <T> measureTime(block: () -> T): T {
    val start = System.currentTimeMillis()
    val result = block()  // Lambda object allocated
    println("Time: ${System.currentTimeMillis() - start}ms")
    return result
}

// With inline
inline fun <T> measureTime(block: () -> T): T {
    val start = System.currentTimeMillis()
    val result = block()  // No allocation, code inlined
    println("Time: ${System.currentTimeMillis() - start}ms")
    return result
}

Benefits:

1. Zero overhead: No lambda object allocation
2. Non-local returns: Can return from outer function inside lambda
3. reified enabled: Access to type parameter at runtime

reified: Runtime Type Access

Mental model: reified makes generic type T available at runtime. Only works with inline.

Amethyst pattern:

// OptimizedJsonMapper.kt:48
expect object OptimizedJsonMapper {
    inline fun <reified T : OptimizedSerializable> fromJsonTo(json: String): T
}

// Usage
val event: TextNoteEvent = OptimizedJsonMapper.fromJsonTo(jsonString)
// Compiler inlines and passes TextNoteEvent::class info

Without reified:

// Would need to pass class explicitly
fun <T> fromJson(json: String, clazz: KClass<T>): T {
    return when (clazz) {
        TextNoteEvent::class -> parseTextNote(json) as T
        // ...
    }
}

val event = fromJson(json, TextNoteEvent::class)  // Verbose

With reified:

inline fun <reified T> fromJson(json: String): T {
    return when (T::class) {  // Can access T::class!
        TextNoteEvent::class -> parseTextNote(json) as T
        // ...
    }
}

val event = fromJson<TextNoteEvent>(json)  // Clean

noinline & crossinline

noinline: Prevent specific lambda from being inlined

inline fun foo(
    inlined: () -> Unit,
    noinline notInlined: () -> Unit  // Can be stored, passed around
) {
    inlined()
    someFunction(notInlined)  // Can pass to non-inline function
}

crossinline: Lambda can't do non-local returns

inline fun foo(crossinline block: () -> Unit) {
    launch {
        block()  // OK: crossinline allows lambda in different context
    }
}

6. Value Classes (Opportunity)

Mental model: value class is a compile-time wrapper with zero runtime overhead. Single property, no boxing.

Not currently used in Amethyst - potential optimization.

Pattern:

@JvmInline
value class EventId(val hex: String)

@JvmInline
value class PubKey(val hex: String)

// Type safety without runtime cost
fun fetchEvent(eventId: EventId): Event {
    // eventId.hex accessed without wrapper object
}

val id = EventId("abc123")
fetchEvent(id)  // Type safe
// fetchEvent(PubKey("xyz"))  // Compile error!

When to use:

• Type safety for primitives (IDs, hex strings, timestamps)
• High-frequency allocations (event processing)
• Clear domain types without overhead

Restrictions:

• Single property only
• Must be val
• Can't have init block with logic
• Inline at compile-time, may box in some cases

Amethyst opportunity:

// Current (String everywhere, no type safety)
fun fetchEvent(id: String): Event  // Could pass wrong string

// With value class
@JvmInline value class EventId(val hex: String)
@JvmInline value class PubKeyHex(val hex: String)
@JvmInline value class Bech32(val encoded: String)

fun fetchEvent(id: EventId): Event  // Type safe, zero cost

Common Patterns

Pattern: StateFlow State Management

class MyViewModel {
    private val _state = MutableStateFlow(State.Initial)
    val state: StateFlow<State> = _state.asStateFlow()

    fun loadData() {
        viewModelScope.launch {
            _state.value = State.Loading
            val result = repository.getData()
            _state.value = when (result) {
                is Success -> State.Success(result.data)
                is Error -> State.Error(result.message)
            }
        }
    }
}

sealed class State {
    data object Initial : State()
    data object Loading : State()
    data class Success(val data: List<Item>) : State()
    data class Error(val message: String) : State()
}

Pattern: Sealed Result with Generics

sealed interface Result<out T> {
    data class Success<T>(val value: T) : Result<T>
    data class Error(val exception: Exception) : Result<Nothing>
    data object Loading : Result<Nothing>
}

// Use with variance
fun <T> fetchData(): Result<T> = ...

val userResult: Result<User> = fetchData()
val itemResult: Result<List<Item>> = fetchData()

Pattern: Immutable Event Builder

@Immutable
data class Event(
    val id: String,
    val kind: Int,
    val content: String,
    val tags: ImmutableList<Tag>
) {
    companion object {
        fun builder() = EventBuilder()
    }
}

class EventBuilder {
    private var id: String = ""
    private var kind: Int = 1
    private var content: String = ""
    private val tags = mutableListOf<Tag>()

    fun id(value: String) = apply { id = value }
    fun kind(value: Int) = apply { kind = value }
    fun content(value: String) = apply { content = value }
    fun tag(tag: Tag) = apply { tags.add(tag) }

    fun build() = Event(id, kind, content, tags.toImmutableList())
}

// Usage
val event = Event.builder()
    .id("abc")
    .kind(1)
    .content("Hello")
    .tag(Tag.P("pubkey"))
    .build()

Delegation Guide

When to delegate:

| Topic | Delegate To | This Skill Covers | |-------|-------------|-------------------| | Structured concurrency, channels | kotlin-coroutines agent | Flow state patterns only | | expect/actual, source sets | kotlin-multiplatform skill | Platform-agnostic Kotlin | | General Compose patterns | compose-expert skill | @Immutable for performance | | Build configuration | gradle-expert skill | - |

Ask kotlin-coroutines agent for:

• Advanced Flow operators (flatMapLatest, combine, zip)
• Channel patterns
• Structured concurrency (supervisorScope, coroutineScope)
• Error handling in coroutines

This skill teaches:

• StateFlow/SharedFlow state management
• Sealed hierarchies
• @Immutable for Compose
• DSL builders
• Inline/reified patterns

Anti-Patterns

❌ Mutable public state:

val accountState: MutableStateFlow<AccountState>  // BAD

✅ Immutable public interface:

val accountState: StateFlow<AccountState> = _accountState.asStateFlow()

❌ Sealed class for generic results:

sealed class Result<T> {  // BAD: Can't use variance
    data class Success<T>(val value: T) : Result<T>()
}

✅ Sealed interface for generics:

sealed interface Result<out T> {  // GOOD: Covariance
    data class Success<T>(val value: T) : Result<T>
}

❌ Mutable properties in @Immutable class:

@Immutable
data class Event(
    var content: String  // BAD: var breaks immutability
)

✅ All val:

@Immutable
data class Event(
    val content: String
)

❌ Passing class explicitly when reified available:

inline fun <T> parse(json: String, clazz: KClass<T>): T  // BAD

✅ Use reified:

inline fun <reified T> parse(json: String): T  // GOOD

Quick Reference

Flow Decision Tree

Need to expose state?
    YES → StateFlow (always has value, single latest)
    NO → Need events? → SharedFlow (optional replay, broadcast)

Need to mutate?
    Internal only → MutableStateFlow (private)
    Expose publicly → StateFlow via .asStateFlow()

Sealed Decision Tree

Need common data in base type?
    YES → sealed class
    NO → sealed interface

Need generics with variance?
    YES → sealed interface
    NO → Either works

Need multiple inheritance?
    YES → sealed interface
    NO → Either works

Inline Decision Tree

Passing lambda to function?
    Called frequently? → inline (performance)
    Need reified? → inline (required)
    Need to store/pass lambda? → regular fun (can't inline)

Resources

Official Docs

• StateFlow and SharedFlow | Android Developers
• Sealed Classes | Kotlin Docs
• Inline Functions | Kotlin Docs

Bundled References

• references/flow-patterns.md - StateFlow/SharedFlow examples from AccountManager, RelayManager
• references/sealed-class-catalog.md - All sealed types in quartz
• references/dsl-builder-examples.md - TagArrayBuilder, other DSL patterns
• references/immutability-patterns.md - @Immutable usage, data classes, collections

Codebase Examples

• AccountManager.kt:36-50 - sealed class AccountState, StateFlow pattern
• RelayConnectionManager.kt:44-52 - StateFlow state management
• SignerResult.kt:25-46 - sealed interface with generics
• TextNoteEvent.kt:51-63 - @Immutable event class
• TagArrayBuilder.kt:23-90 - DSL builder pattern, inline function
• OptimizedJsonMapper.kt:48 - inline fun with reified

Version: 1.0.0 Last Updated: 2025-12-30 Codebase Reference: AmethystMultiplatform commit 258c4e011