{"text":"Reason carefully. Summarize the key facts before answering. Keep the final answer concise and grounded in the prompt."}
{"text":"When solving a task, separate observations, intermediate reasoning, and final answer. Avoid placeholder text and avoid repeating the same question."}
{"text":"Active reasoning should gather useful information, update internal state, and stop once enough evidence is available."}
{"text":"A well-behaved model should not emit empty think blocks, fake citations, or placeholder answers such as <answer>."}
{"text":"If the prompt is ambiguous, ask a targeted clarifying question. If the prompt is clear, answer directly with a short justification."}
{"text":"Reasoning models should prefer correct, verifiable, and compact responses over decorative or repetitive chains of thought."}
{"text":"Tool use should improve factual grounding. Do not call tools without a purpose, and do not repeat the same tool call without new evidence."}
{"text":"The model should preserve context, integrate new facts, and avoid self-locking loops during multi-turn problem solving."}