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)

Use When	Skip When
Complex business domain with many rules	Simple CRUD, few business rules
Long-lived system (years of maintenance)	Prototype, MVP, throwaway code
Team of 5+ developers	Solo developer or small team (1-2)
Multiple entry points (API, CLI, events)	Single entry point, simple API
Need to swap infrastructure (DB, broker)	Fixed infrastructure, unlikely to change
High test coverage required	Quick scripts, internal tools

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

Pattern Boundaries

Pattern	Primary Question	Use It For	Do Not Treat As
DDD	How do we model a complex business domain?	Ubiquitous language, bounded contexts, aggregates, value objects	A folder structure by itself
Hexagonal Architecture	How does the application interact with the outside world?	Ports, driver adapters, driven adapters, testable application core	A mandate for six sides or one exact package layout
Clean Architecture	Which direction should dependencies point?	Inward dependency rule, use case boundaries, framework independence	A universal four-folder template
Onion Architecture	How do we keep the domain model central?	Domain-centered layers and dependency inversion	A separate requirement when Clean/Hexagonal already solve the local problem
CQRS	Do reads and writes need different models?	Bounded contexts with divergent read/write workloads	A default application architecture
Event Sourcing	Do we need state from a complete event history?	Audit, temporal queries, replayable workflows	A persistence default for CRUD systems

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:

Code	Layer	Why
Business invariant ("order needs items to confirm")	Domain (entity method)	It's a rule, not orchestration
Input format validation (JSON shape, required fields)	Adapter (controller/DTO)	Protocol concern, not business rule
Transaction begin/commit	Application	Use case = transaction boundary
ORM entity / table model	Infrastructure	Map to domain objects; never let ORM entities BE domain entities
Domain ↔ DB mapping	Infrastructure (mapper)	Persistence detail
Authorization ("is user allowed?")	Application (policy) or adapter middleware	Domain stays auth-agnostic; encode role RULES in domain only if they're business rules
Clock, UUID generation	Port in domain/application; adapter in infrastructure	Keeps domain deterministic and testable
Reacting to a domain event	Application (event handler)	Side effects = orchestration
Query joining many tables for a screen	Read model (application interface, infrastructure impl)	Don't force it through aggregates

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

Pattern	Purpose	Layer	Key Rule
Entity	Identity + behavior	Domain	Equality by ID
Value Object	Immutable data	Domain	Equality by value, no setters
Aggregate	Consistency boundary	Domain	Only root is referenced externally
Domain Event	Record of change	Domain	Past tense naming (OrderPlaced)
Repository	Persistence abstraction	Domain (port)	Per aggregate, not per table
Domain Service	Stateless logic	Domain	When logic doesn't fit an entity
Application Service	Orchestration	Application	Coordinates domain + infra

Anti-Patterns (CRITICAL)

Anti-Pattern	Problem	Fix
Anemic Domain Model	Entities are data bags, logic in services	Move behavior INTO entities
Repository per Entity	Breaks aggregate boundaries	One repository per AGGREGATE
Leaking Infrastructure	Domain imports DB/HTTP libs	Domain has ZERO external deps
God Aggregate	Too many entities, slow transactions	Split into smaller aggregates
Skipping Use Cases	Controllers call repositories directly in a use-case architecture	Route through application use cases
CRUD Thinking	Modeling data, not behavior	Model business operations
Premature CQRS	Adding complexity before needed	Start with simple read/write, evolve
Cross-Aggregate TX	Multiple aggregates in one transaction	Use domain events for consistency

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:

Before you...	Read
Write code in any layer, wire dependency injection, or decide 3-layer vs 4-layer	references/LAYERS.md
Split a system into services/contexts, integrate with a legacy or third-party system (ACL), run Event Storming	references/DDD-STRATEGIC.md
Model an entity, value object, aggregate, repository, domain service, or factory	references/DDD-TACTICAL.md
Define ports/adapters, name interfaces, or lay out a ports-first structure	references/HEXAGONAL.md
Add commands/queries, domain vs integration events, outbox, sagas, or evaluate CQRS/Event Sourcing	references/CQRS-EVENTS.md
Write unit/integration/architecture tests for any layer	references/TESTING.md
Answer a quick "which pattern/which layer" question without deep-diving	references/CHEATSHEET.md

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