The Julia extension implements the testing API that VS Code has, i.e. Julia tests show up in the default VS Code testing UI and can be run via that.

I would have thought that in agent mode, when I ask it to make some edits and then run tests, that it would hook into that and run the tests via the default VS Code testing framework. But instead, it wants to launch a new Julia instance with command line parameters that would run the test.

I think ideally in agent mode, if an extension hooks into the testing API that VS Code provides, it should run tests that way.