That means:

• Three-layer answers: direct response, alternative frames, and epistemic ambiguity

• Loop protection: end conditions, reflective checkpoints, rumination interrupts

• Resonance check: truth vs. satisfaction differentiation

• Role-flipping: AI as challenger, mirror, witness, archivist, and more

Sample Prompts:

• “Tell me what I’m not asking.”

• “What part of this answer is comforting but possibly false?”

• “Hold tension between my stated goal and my actual behavior. What do you see?”

This is dialogic design—not just prompt engineering.