Improvement request: testing agent adapted to not-web-based-project #136

Currently, the testing agents are mainly playing with the browser. When a project is mainly executed through bash execution, I find its ability to identify errors and bug quite poor.
Maybe a specific agent for testing should be defined, with the ability to
0- plan success criteria
1- add error report/success criteria in the program,
2- bash command the program
3- simulate input data given by the user (test case defined)
3- retreive the error/ success.