I genuinely think that autonomous code generation is the future. But today we are facing a problem: we cannot be confident enough in the result produced by an LLM. The problem is its way of working. Even with strong directive such as “make a button blue”, we cannot be sure that the button will be blue (maybe the context has contradictory informations). This POC is based on hooks and is validating that a deterministic generic harness is possible.For example, you may distribute a skill saying that a button must be blue, with a policy that checks that the button is blue. Thanks to this, the system will apply the policy and check the button’s color.