Why not use a client-server infrastructure for tests? The server sends the test code, the client runs the code, sends the output to the server and this replies pass/not pass.

One could even make zero-knowledge test development this way.