Inject the prior art into the (ever increasing) context window, let in-context-learning to its thing and go?