Junie Enhancement

As I learn to use Junie I feel I need to have a place to customize the system prompt defining the parameters for which the system should respond. Giving me the ability to fine tune the response and make Junie return better results.  These fine tuning instructions need to be customizable by the file type that I am working with. 

For example Junie falls down when doing terraform by not fully researching the object models required for a implementation. This can be solve by some simple coaching in the prompt. 

I would also like to see a alter ego agent that protect my system from Junie… IE a manager that reviews the potential fix and comments on it. Looks for security flaws like running SUDO without sudo permissions. Ensuring that the process is being followed. By pitting the Junie Agent against a Management agent, and letting them iterate on a problem solution, Junie may be able to solve more problems. 

The ability to coach genie into how to test something in the environment is crucial. So that when It determines that it needs to deploy and test terraform code, It understands enough of the environment and shell scripting to do so properly. Ability to coach Junie with descriptions of the environment would also be handy.