Very cool work! Regarding your finding "the tool ran successfully and returned data" and "the tool ran successfully but found nothing." Couldn’t this be solved by designing better tool responses instead of adding another layer in between? Just curious and probing my understanding.