Think first, then compare notes with the AI
It's easy to skim an AI answer and nod along. This pattern asks for your take first, then offers a short reflection on where your thinking overlapped with the AI's and where it diverged. A small nudge, no big claims.
Inspired by writing on "cognitive surrender" — the tendency to stop thinking when an AI is happy to do it for you.
Automatically detects opinion, analysis, strategy, and decision questions — the types where your own reasoning has the most value.
After you think and the AI answers, get specific feedback: where you were strong, what you missed, and where your thinking diverged.
Say "just answer" or "skip" any time. The skill is a training partner, not a gatekeeper.
Four small steps
Claude identifies the question type and asks you to think first.
Articulate your reasoning, instinct, or even "I have no idea."
Full answer plus feedback on your thinking.
If you raised something interesting, explore it further.
Different questions get different coaching prompts
"Before I weigh in — what's your current read on this?"
"Walk me through what you've observed so far."
"What options are you already considering?"
"What are the trade-offs as you see them?"
Copy a prompt and paste it into Claude Code or any LLM
Should I use a monorepo or separate repos for my microservices?
What's the best way to handle authentication in a Next.js app?
Why is my deployment pipeline so slow?
Which database should I choose — Postgres or MongoDB?
How should I architect this event-driven system?
Other patterns
Think first with Cognitive Flex, then reach for Animal Magic when you need to generate or analyse, or Dinner Party when you want pressure from many angles.