Agent decided to test if harmful command block worked by issuing a rm -rf / Thankfully it worked so only damage was a mild heart attack. I implemented a sandbox immediately afterwards.
