DDD, CQRS, and Vertical Slice Architecture

Vertical Slice Architecture organizes code by feature (or use case), not by technical layer. Each slice is a small, self-contained vertical: one folder, one command or query, one handler, and only the domain types it needs. Combined with CQRS (separate read and write paths) and a thin DDD core, you get low context per file and a structure that is easy for an AI agent (or a new developer) to implement one slice at a time.

One slice = one feature

• Slice = one user-facing capability (e.g. “Place order”, “Get order details”).
• Folder = one slice. No “Controllers”, “Handlers”, “Repositories” at the top level; instead PlaceOrder/, GetOrderDetails/.
• Files = command or query, handler, optional validator, and the minimal domain or persistence it uses. Keep each file small and single-purpose.

This keeps context low: to add or change “Place order”, you open one folder and a few files.

How the slice fits together

Rendering diagram…

Folder layout (low context)

Copy

src/ ├── Features/ │ ├── PlaceOrder/ │ │ ├── PlaceOrderCommand.cs │ │ ├── PlaceOrderHandler.cs │ │ └── PlaceOrderValidator.cs │ ├── GetOrderDetails/ │ │ ├── GetOrderDetailsQuery.cs │ │ └── GetOrderDetailsHandler.cs │ └── CancelOrder/ │ ├── CancelOrderCommand.cs │ └── CancelOrderHandler.cs ├── Domain/ # Shared only when needed │ ├── Order.cs │ └── OrderRepository.cs └── Infrastructure/ # DB, external APIs

Slices do not depend on each other. They depend on a small Domain (entities, value objects, repository interfaces) and Infrastructure implements those interfaces.

CQRS in the slice

• Command = write. One class for the command, one handler. Handler loads aggregate(s), calls domain methods, saves. Returns success/failure or an id.
• Query = read. One class for the query, one handler. Handler loads data (repository or read model), maps to a DTO, returns. No domain logic.

Each slice owns its request and response types. No giant shared “command bus” types; the slice is the boundary.

DDD: thin domain

• Aggregate root = the entity the slice loads and saves (e.g. Order). Keep aggregates small; one root per consistency boundary.
• Value objects = immutable types for concepts (e.g. OrderId, Money). Use them inside the aggregate.
• Repository interface = in Domain or next to the slice; implementation in Infrastructure. Slice only knows “load/save this aggregate”.

Avoid “domain services” that span many slices. Prefer logic inside the aggregate or inside the slice’s handler. That keeps the domain small and the slice self-contained.

Why this helps AI (and humans)

• Low context: One feature = one folder. The agent (or developer) only needs to read a few files to implement or change a slice.
• Small files: One command, one handler, one query, one handler. Easy to generate or edit in one go.
• Clear contract: Command and query types are the API. Handlers are the only place that touches domain or persistence for that use case.
• No big shared layers: No “Application layer” with 50 handlers. Slices stay independent; you can add or regenerate one slice without touching others.

Minimal implementation checklist

1. Create a folder per feature under Features/.
2. Add a command or query type (record/class) and a single handler class.
3. Handler: for commands, load aggregate → call method → save; for queries, load data → map to DTO → return.
4. Put shared entities and repository interfaces in Domain/; implement repositories in Infrastructure/.
5. Wire handlers in the composition root (DI) by feature, not by layer.

This gives you a DDD-CQRS–style design with vertical slices: small, bounded context per feature and small files that are easy for an AI agent to implement or modify one slice at a time.