You'll notice
Shell commands are refused while the owning agent has a kill still settling, rather than being run anyway.
Before spawning a shell command, Claude Code now checks whether the owning agent has a kill pending loop settlement in progress, and refuses to run the command if so instead of executing it.