okay, after digging a little, I think the snippet above is not valid. "standalone script" - assumes just a single python file. I still need repo where training scripts live. It just cant be cloned on remote worker. Is it possible to somehow pack local repo to be executed on remote worker?
Why not give ClearML read-only access credentials to the repository?
was rather on infrastructure level - worker doesnt have access to the repository. I found workaround of using mirror repo that is accessible by worker..