Three contracts to check when an agent uses models and web data

An agent workflow can succeed at the model call and still fail at the task. I find it useful to separate three contracts before scaling a workflow. 1. The output contract Write the expected structure