User: "Should I buy A or B?"
If stakes are low and the evidence is enough, the skill should answer directly. It should not launch a debate.
User: "Here's my startup launch plan. Pressure-test it."
The skill should identify the biggest failure mode, the assumption most likely to break the plan, and the cheapest useful test.
User: "Review this architecture before I build it."
The skill should focus on irreversible choices, bottlenecks, security/reliability risks, and whether a staged design can validate the risky part first.
User: "Is this recommendation actually supported by the evidence?"
The skill should separate verified claims from assumptions and check freshness, scope, source quality, and duplicate evidence when tools are available.
User: "Just tell me which one."
Return the recommendation and one decisive reason. Do not dump the process.
User: "Deeply review this."
Increase verification and adversarial review only where they can change the action. Do not reveal hidden chain-of-thought.
User: "Should I do it?"
If the missing context is essential, ask one focused question. If a reasonable assumption makes a useful answer possible, state it and continue.