In theory, that is the benefit of having an agent that is limited to only doing the tests, and an agent that only does the coding, and have them run separately,

In theory, that is the benefit of having an agent that is limited to only doing the tests, and an agent that only does the coding, and have them run separately, that way to fix a test, you don't change the test, etc...