What is the purpose of this?
Attackers do kill those agents, absolutely, but you'd have to be root to kill the agent with this method (and ideally any method), at which point endpoint instrumentation and forensic artifacts are unreliable anyway. Any logs that someone were looking at that said "attacker killed me" could just be deleted/ rewritten. The attacker could even rewrite them on the fly and ship them off the box so you're none the wiser :)
For example you have a "curl" command in a script that is waiting for a server that doesn't respond, killing the process with the kill command may cause the script to fail. If you want the script to continue anyway, you can use that trick.
Not a security bug or anything. Just fun.