Make the LLM operate the hypervisor VM so it can observe a binary as it executes to write specs for it?