clean-ddd-hexagonal

Clean Architecture + DDD + Hexagonal

Backend architecture combining DDD tactical patterns, Clean Architecture dependency rules, and Hexagonal ports/adapters for maintainable, testable systems.

This skill is an opinionated synthesis of several related architecture traditions. It is not a single canonical architecture model. Use the original source that matches the design question you are answering: DDD for domain modeling, Hexagonal Architecture for ports/adapters, Clean Architecture for dependency direction, Onion Architecture for domain-centered layering, and CQRS/Event Sourcing only for specific read/write or temporal requirements.

When to Use (and When NOT to)

Start simple. Evolve complexity only when needed. Most systems don't need full CQRS or Event Sourcing.

Pattern Boundaries

CRITICAL: The Dependency Rule

Dependencies point inward only. Outer layers depend on inner layers, never the reverse.

Infrastructure → Application → Domain
   (adapters)     (use cases)    (core)

Violations to catch:

• Domain importing database/HTTP libraries
• In this architecture style, controllers calling repositories directly instead of application use cases
• Entities depending on application services

Design validation: "Create your application to work without either a UI or a database" — Alistair Cockburn. If you can run your domain logic from tests with no infrastructure, your boundaries are correct.

Quick Decision Trees

"Where does this code go?"

Where does it go?
├─ Pure business logic, no I/O           → domain/
├─ Orchestrates domain + has side effects → application/
├─ Talks to external systems              → infrastructure/
├─ Defines HOW to interact (interface)    → port (domain or application)
└─ Implements a port                      → adapter (infrastructure)

Sharp edges — the placements LLMs most often get wrong:

Litmus test for anemic domain models: if an application service reads state out of an entity, decides, then writes state back (if (order.status === 'draft') order.status = 'confirmed'), move that logic into the entity as order.confirm(). Handlers should read like a script: load aggregate → call one behavior method → save → publish.

"Is this an Entity or Value Object?"

Entity or Value Object?
├─ Has unique identity that persists → Entity
├─ Defined only by its attributes    → Value Object
├─ "Is this THE same thing?"         → Entity (identity comparison)
└─ "Does this have the same value?"  → Value Object (structural equality)

"Should this be its own Aggregate?"

Aggregate boundaries?
├─ Must be consistent together in a transaction → Same aggregate
├─ Can be eventually consistent                 → Separate aggregates
├─ Referenced by ID only                        → Separate aggregates
└─ >10 entities in aggregate                    → Split it

Rule: One aggregate per transaction. Cross-aggregate consistency via domain events (eventual consistency).

Directory Structure

src/
├── domain/                    # Core business logic (NO external dependencies)
│   ├── {aggregate}/
│   │   ├── entity              # Aggregate root + child entities
│   │   ├── value_objects       # Immutable value types
│   │   ├── events              # Domain events
│   │   ├── repository          # DDD repository interface (driven port)
│   │   └── services            # Domain services (stateless logic)
│   └── shared/
│       └── errors              # Domain errors
├── application/               # Use cases / Application services
│   ├── {use-case}/
│   │   ├── command             # Command/Query DTOs
│   │   ├── handler             # Use case implementation
│   │   └── port                # Driver port interface
│   └── shared/
│       └── unit_of_work        # Transaction abstraction
├── infrastructure/            # Adapters (external concerns)
│   ├── persistence/           # Database adapters
│   ├── messaging/             # Message broker adapters
│   ├── http/                  # REST/GraphQL adapters (DRIVER)
│   └── config/
│       └── di                  # Dependency injection / composition root
└── main                        # Bootstrap / entry point

Port placement: This skill defaults to a DDD-centered layout where aggregate repository interfaces live beside the aggregate in domain/. A stricter Hexagonal layout may instead put driven ports under application/ports/driven/. Pick one convention per codebase and keep the dependency rule intact.

Presentation layer: Driver adapters (REST/gRPC/CLI) live under infrastructure/ in this default layout. Some codebases lift them into a fourth top-level presentation/ layer instead (references/LAYERS.md shows that variant). Use one home for controllers, not both.

Event publishing: Saving an aggregate and then publishing its events to a broker are two writes; a crash between them silently drops events. When events must reach other services reliably, write them to an outbox table in the same transaction as the aggregate — see the outbox pattern in references/CQRS-EVENTS.md.

DDD Building Blocks

Anti-Patterns (CRITICAL)

Implementation Order

1. Discover the Domain — Event Storming, conversations with domain experts
2. Model the Domain — Entities, value objects, aggregates (no infra)
3. Define Ports — Repository interfaces, external service interfaces
4. Implement Use Cases — Application services coordinating domain
5. Add Adapters last — HTTP, database, messaging implementations

DDD is collaborative. Modeling sessions with domain experts are as important as the code patterns.

Reference Documentation

Read the matching file before doing the task in the left column:

Sources

Primary Sources

• Hexagonal Architecture — Alistair Cockburn (2005)
• Domain-Driven Design: The Blue Book — Eric Evans (2003)
• The Clean Architecture — Robert C. Martin (2012)
• Onion Architecture — Jeffrey Palermo (2008)
• Implementing Domain-Driven Design — Vaughn Vernon (2013)

Primary Pattern References

• CQRS — Martin Fowler
• Event Sourcing — Martin Fowler
• Repository Pattern — Martin Fowler (PoEAA)
• Unit of Work — Martin Fowler (PoEAA)
• Bounded Context — Martin Fowler
• Transactional Outbox — microservices.io
• Effective Aggregate Design — Vaughn Vernon

Implementation Guides

• Microsoft: DDD + CQRS Microservices
• Domain Events — Udi Dahan

Supplemental Syntheses

• Clean Architecture: Standing on the Shoulders of Giants — Herberto Graça
• Explicit Architecture — Herberto Graça (opinionated synthesis, not canonical source)
• Get Your Hands Dirty on Clean Architecture — Tom Hombergs