Aggregate trigger evaluation results across CLIs, compute trigger rates, and optimize skill descriptions using train/validation splits.
