feat: yes allowed apps (whitelist) please

There are many, patently non-destructive apps or app command formulations that could already be safely always allowed now to save time. nl is a good example of a tool that is called a lot for the agent to orient itself in the code and is reliably non-dangerous.

It would be great to implement feature as a list of RegEx such that safe sub-commands can be allowed without allowing other sub-commands that could cause problems if not human-in-the-looped.