Yes, it is somewhat tough to do. My implementation captures it, and tries to avoid accidentally doing anything bad by only trusting some specific types of output that might be mounted (so it is sure it isn’t causing a loop with stdout/stderr). There is no great perfect answer in the JVM to do thread specific I/O capture. I have a first attempt at it in the PR, and in the Keplin code base. But it needs more time, thoughts and churning.