Files
architecture/old2/skills/ddd/SKILL.md
2026-01-15 17:28:06 +01:00

273 lines
7.5 KiB
Markdown

---
name: ddd
description: >
Domain-Driven Design concepts: bounded contexts, aggregates, commands, events,
and tactical patterns. Use when analyzing domain models, identifying bounded
contexts, or mapping features to DDD patterns.
user-invocable: false
---
# Domain-Driven Design (DDD)
Strategic and tactical patterns for modeling complex domains.
## Strategic DDD: Bounded Contexts
### What is a Bounded Context?
A **bounded context** is a boundary within which a domain model is consistent. Same terms can mean different things in different contexts.
**Example:** "Order" means different things in different contexts:
- **Sales Context**: Order = customer purchase with payment and shipping
- **Fulfillment Context**: Order = pick list for warehouse
- **Accounting Context**: Order = revenue transaction
### Identifying Bounded Contexts
Look for:
1. **Different language**: Same term means different things
2. **Different models**: Same concept has different attributes/behavior
3. **Different teams**: Natural organizational boundaries
4. **Different lifecycles**: Entities created/destroyed at different times
5. **Different rate of change**: Some areas evolve faster than others
**From vision/manifesto:**
- Identify personas → each persona likely interacts with different contexts
- Identify core domain concepts → group related concepts into contexts
- Identify capabilities → capabilities often align with contexts
**From existing code:**
- Look for packages/modules that cluster related concepts
- Identify seams where code is loosely coupled
- Look for translation layers between subsystems
- Identify areas where same terms mean different things
### Context Boundaries
**Good boundaries:**
- Clear interfaces between contexts
- Each context owns its data
- Contexts communicate via events or APIs
- Minimal coupling between contexts
**Bad boundaries:**
- Shared database tables across contexts
- Direct object references across contexts
- Mixed concerns within a context
### Common Context Patterns
| Pattern | Description | Example |
|---------|-------------|---------|
| **Core Domain** | Your unique competitive advantage | Custom business logic |
| **Supporting Subdomain** | Necessary but not differentiating | User management |
| **Generic Subdomain** | Common problems, use off-the-shelf | Email sending, file storage |
## Tactical DDD: Building Blocks
### Aggregates
An **aggregate** is a cluster of entities and value objects treated as a unit for data changes.
**Rules:**
- One entity is the **aggregate root** (only entity referenced from outside)
- All changes go through the root
- Enforce business invariants within the aggregate
- Keep aggregates small (2-3 entities max when possible)
**Example:**
```
Order (root)
├── OrderLine
├── ShippingAddress
└── Payment
```
External code only references `Order`, never `OrderLine` directly.
**Identifying aggregates:**
- What entities always change together?
- What invariants must be enforced?
- What is the transactional boundary?
### Commands
**Commands** represent intent to change state. Named with imperative verbs.
**Format:** `[Verb][AggregateRoot]` or `[AggregateRoot][Verb]`
**Examples:**
- `PlaceOrder` or `OrderPlace`
- `CancelSubscription` or `SubscriptionCancel`
- `ApproveInvoice` or `InvoiceApprove`
**Commands:**
- Are handled by the aggregate root
- Either succeed completely or fail
- Can be rejected (return error)
- Represent user intent or system action
### Events
**Events** represent facts that happened in the past. Named in past tense.
**Format:** `[AggregateRoot][PastVerb]` or `[Something]Happened`
**Examples:**
- `OrderPlaced`
- `SubscriptionCancelled`
- `InvoiceApproved`
- `PaymentFailed`
**Events:**
- Are immutable (already happened)
- Can be published to other contexts
- Enable eventual consistency
- Create audit trail
### Value Objects
**Value Objects** are immutable objects defined by their attributes, not identity.
**Examples:**
- `Money` (amount + currency)
- `EmailAddress`
- `DateRange`
- `Address`
**Characteristics:**
- No identity (two with same values are equal)
- Immutable (cannot change, create new instance)
- Can contain validation logic
- Can contain behavior
**When to use:**
- Concept has no lifecycle (no create/update/delete)
- Equality is based on attributes, not identity
- Can be shared/reused
### Entities
**Entities** have identity that persists over time, even if attributes change.
**Examples:**
- `User` (ID remains same even if name/email changes)
- `Order` (ID remains same through lifecycle)
- `Product` (ID remains same even if price changes)
**Characteristics:**
- Has unique identifier
- Can change over time
- Identity matters more than attributes
## Mapping Features to DDD Patterns
### Process
For each feature from vision:
1. **Identify the bounded context**: Which context does this belong to?
2. **Identify the aggregate(s)**: What entities/value objects are involved?
3. **Identify commands**: What actions can users/systems take?
4. **Identify events**: What facts should be recorded when commands succeed?
5. **Identify value objects**: What concepts are attribute-defined, not identity-defined?
### Example: "User can place an order"
**Bounded Context:** Sales
**Aggregate:** `Order` (root)
- `OrderLine` (entity)
- `ShippingAddress` (value object)
- `Money` (value object)
**Commands:**
- `PlaceOrder`
- `AddOrderLine`
- `RemoveOrderLine`
- `UpdateShippingAddress`
**Events:**
- `OrderPlaced`
- `OrderLineAdded`
- `OrderLineRemoved`
- `ShippingAddressUpdated`
**Value Objects:**
- `Money` (amount, currency)
- `Address` (street, city, zip, country)
- `Quantity`
## Refactoring to DDD
When existing code doesn't follow DDD patterns:
### Identify Misalignments
**Anemic domain model:**
- Entities with only getters/setters
- Business logic in services, not entities
- **Fix:** Move behavior into aggregates
**God objects:**
- One entity doing too much
- **Fix:** Split into multiple aggregates or value objects
**Context leakage:**
- Same model shared across contexts
- **Fix:** Create context-specific models with translation layers
**Missing boundaries:**
- Everything in one module/package
- **Fix:** Identify bounded contexts, separate into modules
### Refactoring Strategies
**Extract bounded context:**
```markdown
As a developer, I want to extract [Context] into a separate module,
so that it has clear boundaries and can evolve independently
```
**Extract aggregate:**
```markdown
As a developer, I want to extract [Aggregate] from [GodObject],
so that it enforces its own invariants
```
**Introduce value object:**
```markdown
As a developer, I want to replace [primitive] with [ValueObject],
so that validation is centralized and the domain model is clearer
```
**Introduce event:**
```markdown
As a developer, I want to publish [Event] when [Command] succeeds,
so that other contexts can react to state changes
```
## Anti-Patterns
**Avoid:**
- Aggregates spanning multiple bounded contexts
- Shared mutable state across contexts
- Direct database access across contexts
- Aggregates with dozens of entities (too large)
- Value objects with identity
- Commands without clear aggregate ownership
- Events that imply future actions (use commands)
## Tips
- Start with strategic DDD (bounded contexts) before tactical patterns
- Bounded contexts align with team/organizational boundaries
- Keep aggregates small (single entity when possible)
- Use events for cross-context communication
- Value objects make impossible states impossible
- Refactor incrementally - don't rewrite everything at once