Describe the selection criteria for an agent. Return ONLY JSON: {"goal":"what we choose","must_have":"required conditions","nice_to_have":"desirable advantages","priority":"what is more important in case of a conflict (for example: price > reviews > term)","deal_breakers":"conditions under which the option is immediately discarded"}
Criteria are a value system that you pass on to the agent so that he can choose like you. Three levels: mandatory (do not consider without this), desirable (plus, but not decisive) and stop conditions (immediately in the trash). Mechanics: the agent evaluates the options one by one and “weighs” them according to your rules; if there are no rules, he takes the first consistent one. The key is priority in conflict. In reality, there is no ideal option: cheaper, but worse reviews; Better reviews, but longer delivery. Technique: set an explicit order of importance - “price is more important than reviews, reviews are more important than deadline.” Then the agent consciously sacrifices the less important, and does not decide at random. The second technique, which saves a lot of time, is stop conditions (“no prepayment 100%”, “do not take a rating below 4.0”): they cut off half of the garbage even before comparison, like a good negative filter. Insider: Stop conditions are often more valuable than a wish list - they remove noise. A typical mistake is to lump everything into a “cheaper and better” pile without order. Then the “better” and “cheaper” fight, and the agent chooses randomly.
Unlock access to submit solutions for instant AI review.