Most developers using AI coding assistants have noticed the pattern: Claudee Code](/tools/claude-code/) or Codex writes the code, you accept it, ship it, and three weeks later you can't explain how any of it works. A new open-source plugin directly addresses this.
Learning Opportunities is a skill for Claude Code and Codex that inserts optional 10-15 minute exercises after you complete architectural work - creating new files, changing schemas, doing refactors. The exercises draw on actual learning science: retrieval practice (recalling information from memory to strengthen retention), spaced repetition, and the "generation effect" (writing code yourself builds deeper understanding than accepting generated output).
The project's README is unusually candid about the problem. It identifies five specific ways AI coding tools work against skill retention: the fluency illusion (clean generated code feels understood even when it isn't), reduced generation (you produce less code yourself), cramming velocity (AI speed eliminates natural reflection pauses), weakened metacognition (no space to notice gaps in your own knowledge), and fewer opportunities for self-testing.
Installation is straightforward for both tools. For Claude Code, add it as a plugin marketplace with /plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git. An optional learning-opportunities-auto variant triggers exercise prompts automatically after each git commit. There's also an orient skill that generates structured orientation lessons when you're navigating an unfamiliar codebase - drawing on research into how expert developers sample codebases rather than reading them exhaustively.
This won't appeal to every developer. Shipping fast on familiar ground means pausing for retrieval exercises adds friction you probably don't want. But if you're using Claude Code to build in languages or patterns you're still learning, this is a practical answer to a real problem that most AI coding tools don't acknowledge exists.