Most of us use Claude Code or Codex as an autocomplete on steroids — ask, accept, ship. The AI does the thinking, you do the copy-paste. A new skill called learning-opportunities flips that dynamic. It's a small drop-in skill for Claude Code and Codex that detects when you're about to learn something and pauses to make sure you actually do.
The idea comes from Dr. Cat Hicks, a researcher who studies how developers learn. Instead of just handing you the answer, the skill nudges the model to surface the why — what concept you just bumped into, what the trade-offs are, what you might want to try yourself. It's deliberate practice baked into your editor, not a separate Duolingo tab you'll forget about.
What I like is how lightweight it is. No new platform, no subscription, no LMS. Just a skill file that changes the agent's behavior so it stops being a vending machine. You can fork it, tune the prompts to your stack, or strip it back if it gets chatty during a deadline crunch.
The broader point: AI tools default to maximum hand-holding because that demos well. If you actually want to grow as an engineer, you have to fight that default on purpose. Install something like this — or write your own — before you wake up in two years unable to debug without a copilot.
