I'm struggling with the concept of repositories for queries

[renetitulaer]

Reading about DDD, my conclusion is that DDD is especially interesting for commands, not queries (in CQRS).

Invariants only have meaning in the context of commands so when you create/update/delete entities.

In the context of DDD we have aggregates. In the context of queries we don't. Within DDD, an order has only a reference to the customer id (most often). For queries it makes sense that an order has a reference to the customer so we can display the name of the customer.

When I want to use Ardalis specifications, the entity should implement IAggregateRoot but for queries (when specifications are especially useful), I don't want to know about repositories. Wouldn't it make more sense to decouple specifications from repositories.

Maybe a repository should even only support loading one aggregate. For create/update/delete, a repository applies to one entity. Maybe this should also apply to read, only read one entity by a key (alternate keys allowed).

Next to the repository there should be a factory or QueryService. A QueryService operates on a collection of entities. Which could be aggregates but they don't have to be, for example an Order model with a customer property.

I'm curious how you think about this.