As shown in the figure, following the hexagonal architecture pattern, the storage interface seems to be placed in the application layer. But according to the DDD specification, the data storage interface should belong to the domain layer. So what is to be done?
I agree with @Tseng in general. Note that the repository interface is a domain language-specific concept (it's in the ubiquitous language and domain experts must understand the meaning of each repository name and method). So it means it's a port that the domain layer exposes.
Can we use it in the presentation layer? Yes, but it's not recommended. We design an application layer to address some concerns (like opening db connections, starting and committing transactions, etc) that can be reused on many presentation layer methods (separation of concerns). The application layer exposes use case scenarios. The presentation layer initiates the application layer use cases.
In this picture, there are two kinds of repositories. One for the write model (commands that use the domain model) and another for the read model (queries that use another model named read model instead of the domain model to optimize queries).
So this sentence from @Tseng's answer is not quite right:
What "Queries can bypass domain layer" means, is that you don't have to use domain or application services to access the repository
It means that queries (in the CQS/CQRS pattern) use the read model instead of the domain model. We don't use the domain repository in queries. So "bypassing the domain layer" means using another model for queries instead of the domain model.