One of the possible solutions was to publish the test projects, and the run tests from the published dll. However, it was not possible to reproduce the issues encountered in the last release so it would not be useful. I'm guessing the issue here is that the test project is referencing code from the applications, so it doesn't get trimmed in tests but would be trimmed when publishing.

For the API there is another solution - run the tests against the docker image. It will mean that either a new test project needs to be created, or the existing ones will need to be modified since they use the application DI container for setup/asserts.

As for the desktop app, I'm not sure.

I think that, at least for now, automated UI tests will be too much. However, instead of testing against the in-memory DesignTimeGnomeshadeClient, the desktop app tests could be run against a docker container built from the current commit. Need to try it out to know which approach would be the best - just testing the view models or moving to headless testing; testing the project as is, or testing a published .dll.