I have no idea how to evaluate this, but I've had it write down the same rule like 3 times and it still keeps managing to fall into this trap of blocking on commands. The fact that it refuses to adhere to my guidelines and rules is probably a model issue, but a better harness could probably overcome the issues.