A Skill That Turns Bug-Fixing Into Actual Learning
May 14, 2026·1 min read
Someone built a Claude Code and Codex skill called [learning-opportunities](https://github.com/DrCatHicks/learning-opportunities) that flips how you use AI coding assistants. Instead of just shipping the fix, it nudges the model to pause an
Someone built a Claude Code and Codex skill called [learning-opportunities](https://github.com/DrCatHicks/learning-opportunities) that flips how you use AI coding assistants. Instead of just shipping the fix, it nudges the model to pause and check if the moment is actually a chance for you to learn something — a new pattern, a tricky bug class, an unfamiliar API.
The pitch is simple but sharp. AI assistants are very good at making you faster and very bad at making you better. If the model fixes every issue silently, your skill ceiling stops moving. This skill adds a deliberate layer: spot the teachable moment, surface it, then decide together whether to explain, pair, or just ship.
What I like is the framing. It treats your editor as a learning environment, not just a code-generation pipe. That is the right mental model for anyone worried about atrophy from over-relying on Copilot-style flows. You still get speed when you want speed, but you also get reps on the hard parts.
My take — install it, or at least steal the idea into your own system prompt. The agents are not going away, so the only real lever left is whether you grow alongside them or quietly stop growing.