The new workflow paradigm expects prompts to be embedded into the integrations directory (e.g. integrations/prompts) where they are automatically picked up. In testing, when this approach is not employed and instead we add prompt information via .add_context in agent.py the workflow prompt injection fails. The agent then has no idea what workflow is active despite the user's selection of a workflow.
Additionally, prompts should remain a top-level directory in the repo as this makes them most obvious and aligns with industry standard.