Use when a model call or a chain of them has to be judged from what the model was actually sent and what it returned: a system prompt with its tools and data, a structured answer, an agentic walk of tool calls, a re-judging. Produces one prompt document in aiview, the request as sent, the reply as returned, and the checks the prompt's own rules imply with what each hit. Any agent, any stack. Not for designing a screen (design-prototype), not for recording or mocking replies, and not for reviewing the code that builds the prompt.
