I really like the core of this idea (haven't tested the tool yet) - solve context-breaking not by better prompting but with a validation step that lives after the result but before human review. Solves the "CAPLOCK PLEA" approach completely, if not token-efficiently. But nowadays an ounce of determinism is worth its weight in plutonium.
I've enjoyed his work too. Particularly the "Dying Earth" stories. I also read "Planet of Adventure".
My goodness did that man have an enormous vocab. I needed a dictionary with me to read his work. Great stories though.