We talked about this in the past. Currently there is some work on a Squirrel-based console, which in fact can allow such things. Even without it, bypasses are possible to get things executed on the AI's side. But what you get in return is a way to make an AI do something it isn't doing on its own.. which is a nice description for cheating

So we decided not to add such feature, although it can be very useful for debugging. Downside of adding 'debug only' stuff, when you remove it, people complain.
To give a small example: once the Windows build had a 'feature' to get the screen in double-size (so pixels were 2x2 instead of 1x1). This was meant as debug-tool only. When it was removed, as it was Windows only, and therefor silly, people complained. In fact, people are still complaining they miss this feature.
So the lesson? Don't include something for debug-only if you are not willing to keep it there for ever. And as this can clearly be abused, I rather not have it in the NoAI framework

On the other side, such 'backdoor' can be very dangerous. The current code executed by the AI have to be aborted to execute your requested code, after which he should continue with his original program. I think everyone understand this is very tricky business. The only way would be via an Event, which you need to add support for in your AI, which makes the whole framework only more complex. So also from the point of view, I think it is better not to have it
Nevertheless, I am glad you bring it up

As I walked with the idea myself for quiet some time.
And to help you a bit with your problem: to debug, I suggest adding tons of print() statements. They help you trace your problem easier, than any other method
