Ddd Playbook

Domain-Driven Design (DDD)

Domain-Driven Design tackles complex software by putting the business domain — its language, rules, and invariants — at the center of the design, instead of letting the database schema, the framework, or the UI drive the model.

Apply this whenever you model a domain or write code around business logic. Follow the decision rules in this file; open a reference file when you need depth on a specific area. The rules matter more than any single example — understand why each exists so you can apply it to cases the examples don't spell out.

Prime directive: model the domain, and protect it

• Put business rules and invariants inside the domain model itself, not scattered across controllers, services, or SQL. A model that only holds data while the logic lives elsewhere is an anemic domain model — the most common DDD failure, and the main thing DDD exists to prevent.
• Speak the ubiquitous language: use the exact terms domain experts use, in the code (class, method, variable names) and in conversation. If the business says "policy", the class is Policy, not InsuranceRecord. A gap between code and language is a defect waiting to happen.
• Keep the domain pure: the domain layer expresses business concepts and must not depend on frameworks, persistence, web, or messaging concerns. Those belong at the edges.

Layered architecture (4 layers)

Organize code into four layers. Dependencies point inward, toward the domain; the domain depends on nothing outside itself.

1. Interfaces — the inbound adaptors: entry points where the outside world drives the context (REST/GraphQL controllers, CLI, message/event listeners, schedulers). Translate external input into application calls and the result back out. No business logic.
2. Application — orchestrates use cases (application services, often split into command services and query services): loads aggregates, invokes domain behavior, manages transactions and security. Coordinates, but holds no business rules itself.
3. Domain — the heart: entities, value objects, aggregates, domain events, domain services, and the repository and service interfaces (ports). All business rules and invariants live here. Depends on nothing but itself.
4. Infrastructure — the outbound adaptors: the technical implementations the context uses to reach the outside (persistence/ORM and repository implementations, messaging, external API clients). Implements the ports the inner layers declare.

The inner layers declare ports (interfaces) and the adaptors implement them — an inbound adaptor brings a request into the context, an outbound adaptor lets it reach out. This is the classic DDD layered model; it does not require hexagonal, onion, or "clean" architecture — those are compatible refinements, but the non-negotiables are domain purity and the inward dependency rule.

Bounded contexts

A bounded context is an explicit boundary within which a model and its ubiquitous language stay consistent. The same word can mean different things in different contexts (a "Customer" in Sales ≠ in Support); don't force one model across the whole system.

Tactical building blocks — how to decide

When modeling, choose the right block deliberately:

• Value Object — no identity; defined entirely by its attributes; immutable. Use liberally: money, date ranges, addresses, quantities, identifiers. Prefer a value object over a primitive whenever a concept carries rules (e.g., Money bundles amount + currency and forbids mixing currencies). They make invalid states unrepresentable.
• Entity — has a distinct identity that persists through changes to its attributes (a Customer stays the same customer even if their name changes). Use when identity and a lifecycle matter.
• Aggregate — a cluster of entities and value objects treated as one consistency unit, accessed only through its aggregate root. Key rules, each with a reason:
  • Keep aggregates small — large ones cause contention and load too much data.
  • One aggregate = one transaction. Don't modify two aggregates in the same transaction; it couples their consistency.
  • Reference other aggregates by identity (ID), never by holding their object — this keeps boundaries and transactions clean.
  • Enforce the aggregate's invariants inside the root, so it is always internally consistent.
• Domain Event — a statement that something meaningful happened in the domain (e.g., OrderPlaced). Use it to achieve eventual consistency across aggregates and to decouple side effects from the action that caused them.
• Domain Service — stateless domain logic that doesn't naturally belong to a single entity or value object (e.g., a transfer between two accounts). Keep it in the domain layer; don't confuse it with an application service.
• Repository — collection-like access to aggregates by their root. Define one repository per aggregate root, with the interface in the domain layer and the implementation in infrastructure.
• Factory — encapsulates complex creation of an aggregate or value object when a plain constructor would be unclear or would leak rules.

CQRS

Command Query Responsibility Segregation separates the model that changes state (commands) from the model that reads it (queries). It comes in two strengths, and conflating them is a common source of over-engineering:

• The light form — split the application layer along the command/query line (command services and query services). One model, one store, no eventual consistency. It is cheap, it keeps write orchestration from tangling with read orchestration, and it is a reasonable default.
• The full form — give each side its own model: a write model (the aggregates, enforcing invariants) and a separate read model shaped for how the data is queried, kept up to date from domain events. It buys queries that span aggregates and independent scaling; it costs projection machinery and eventual consistency. Treat it as a deliberate choice per bounded context, not a default.

How to approach a DDD task

1. Establish the language — clarify the domain terms with the user; use them verbatim in the model.
2. Locate the bounded context — which context are we in, and what is its model?
3. Find the aggregates and their invariants — what must always be true, and what is the consistency boundary?
4. Model tactically — choose value objects, entities, and aggregate roots; push rules into them; keep the domain pure.
5. Place each piece in the right layer — rules in domain, orchestration in application, I/O in interfaces, technical detail in infrastructure.
6. Implement for the stack — follow the stack's implementation idioms; see the routing below.

Strategic design — essentials

Two more strategic decisions shape where the effort goes:

• Subdomains — distinguish the core (your competitive advantage — invest most here), supporting, and generic (buy/reuse) subdomains, so effort goes where it matters.
• Context Mapping — define the relationships between bounded contexts (e.g., an Anti-Corruption Layer to protect your model from an external one).

References

• Tactical patterns in depth — the full catalog with detailed rules, trade-offs, and worked examples, including CQRS depth (read models, eventual consistency, relationship to event sourcing). Read references/tactical-patterns.md.
• Strategic concepts in depth — read references/strategic-design.md.
• The modeling process and tools — EventStorming, Domain Message Flow, the Bounded Context Canvas, the context-map pattern catalog, and the modeling recipe. Read references/modeling-process.md.

Implementing the model

The rules above are stack-agnostic, but the idioms that express them — keeping persistence out of the domain, publishing domain events, shaping the inbound adaptor — differ. When writing code, use the companion skill for the project's stack:

• Spring Boot / Java → the ddd-spring-boot skill
• Angular (frontend, DDD-adapted) → the ddd-angular skill

Each of them carries the same core design rules as this file, so it stands on its own; reach for this skill alongside them when the work is modeling rather than coding. If no companion skill matches the project's stack, apply the rules above idiomatically for that technology.