A 2x2 matrix is as good as the axes you choose. I built a skill that forces Claude to reject the obvious pair, generate alternatives, and produce 2 different matrices. Then I tested it against my own answer to the same exercise.

Persona-based prompts make LLMs sound smart, but that’s not enough for high-quality co-thinking. Here’s how I designed a reasoning protocol for Claude that generates perspectives my mind couldn’t see.