GitHub bot as a tester

Bot can use AST to collect all available public API from the repository and randomly call any of it's part. If the something totally wrong happens the bot should create small reproducible unit test.
Bot should ask is this behavior is expected or not.

Or maybe be bot should just be able to create tests to make tests coverage complete if it is not complete yet.