Extending the recognition-primed decision model to support human-agent collaboration