abstract-factory

name	abstract-factory
description	Apply, review, or refactor code using the Abstract Factory design pattern. Use this skill when the problem matches the pattern's structural forces, especially in Kotlin/JVM backend systems. Do not force the pattern when a simpler language or framework feature is sufficient.

Abstract Factory

Objective

Create families of related or compatible objects without coupling clients to their concrete classes.

Trigger conditions

Use this skill when at least two of these signals are present:

• Several products must vary together as a coherent family.
• Mixing implementations from different families would be invalid.
• The application supports multiple platforms, providers, brands, or environments.

Do not select the pattern from its name alone. First identify the design forces, the axis of change, the ownership of state, and the required lifecycle.

Do not use when

• Only one product varies.
• Families are not required to remain mutually compatible.
• Adding a new product type is more frequent than adding a new family.

Canonical participants

AbstractFactory, ConcreteFactory, AbstractProducts, ConcreteProducts, Client.

Decision workflow

1. Describe the concrete design problem without naming a pattern.
2. Identify what varies and what must remain stable.
3. Establish the client-facing contract.
4. Assign each responsibility to a participant.
5. Prefer composition over inheritance unless inheritance is intrinsic to the pattern.
6. Preserve domain invariants and dependency direction.
7. Implement the smallest viable version.
8. Add tests proving substitutability, collaboration, and failure behavior.
9. Compare the result with a simpler alternative.
10. Record why this pattern is justified.

Kotlin implementation guidance

• Prefer immutable values and constructor injection.
• Use sealed interface or sealed class where the implementation set is closed.
• Use fun interface for single-operation strategies, factories, handlers, or observers.
• Use coroutines for asynchronous boundaries; do not expose blocking behavior as suspend.
• Avoid Java-style ceremony when Kotlin language features already encode the intent.
• Keep framework annotations at adapters or composition roots when possible.
• Make concurrency, cancellation, idempotency, and error semantics explicit.

Reference implementation

interface Button { fun render(): String }
interface Dialog { fun render(): String }

interface UiFactory {
    fun button(): Button
    fun dialog(): Dialog
}

class DarkUiFactory : UiFactory {
    override fun button() = object : Button {
        override fun render() = "dark button"
    }
    override fun dialog() = object : Dialog {
        override fun render() = "dark dialog"
    }
}

The example is intentionally minimal. Adapt naming and boundaries to the domain; do not copy it as production architecture without evaluating lifecycle, failures, observability, and concurrency.

Review checklist

• The problem contains the forces this pattern is intended to resolve.
• The abstraction is stable and meaningful to clients.
• Concrete implementations remain substitutable.
• Domain logic has not leaked into infrastructure wrappers.
• Error handling and lifecycle semantics are explicit.
• Tests cover each concrete participant and the client contract.
• The implementation is simpler than the conditional or coupling it replaces.
• A standard Kotlin, Spring, or library mechanism would not be clearer.

Testing strategy

• Contract-test every interchangeable implementation.
• Test client behavior against fakes through the abstraction.
• Verify invalid state, ordering, and failure propagation.
• Add concurrency tests where instances or shared state cross coroutine boundaries.
• Use integration tests only for adapter/framework behavior; keep pattern semantics unit-testable.

Common failure modes

• Pattern-first design: selecting a named pattern before understanding the problem.
• Ceremony without variability: interfaces and classes that have only one permanent implementation.
• Leaky abstractions: clients depend on concrete implementation details.
• Hidden operational semantics: latency, retries, transactions, or thread safety are obscured.
• Misplaced business rules: orchestration wrappers become the actual domain model.

Output expected from the agent

When applying this skill, produce:

1. A force analysis.
2. A pattern-fit verdict: APPLY, CONSIDER, or REJECT.
3. The proposed participants mapped to domain names.
4. A minimal implementation plan.
5. Kotlin code or a patch.
6. Tests.
7. Trade-offs and rejected alternatives.
8. Migration and rollback notes when refactoring existing code.

Related patterns and alternatives

Factory Method, Builder, Prototype

Source basis

Derived from the classic GoF pattern catalog and the public pattern description at: https://refactoring.guru/design-patterns/abstract-factory.

Do not reproduce proprietary diagrams or paid content. Use the source for conceptual orientation and create original domain-specific explanations and examples.