A pipeline needs to: (1) scrape raw data, (2) clean the data, (3) analyse the cleaned data. How should sub-agents be structured?