A Claude Code and Codex Skill for Deliberate Skill Development

AI wants to turn your coding into homework, and the comments are already grading it

TLDR: This new plugin tries to turn AI-assisted coding into a learning session by offering short lessons while you work. Commenters are split: some love the idea of AI as a tutor, while others say it looks overhyped and want proof it actually helps people learn.

A new tool on GitHub is pitching a very grown-up idea: don’t just let AI build your app, let it teach you while you build. The plugin works with Claude Code and Codex and can pop up short 10–15 minute exercises after big project changes or even after a commit, trying to stop people from sleepwalking through AI-generated code. In plain English, it’s basically saying: sure, let the robot help—but you still need to use your brain.

And wow, the community had thoughts. Some readers were instantly intrigued, with one saying they wanted to use it to learn Java Spring and get quizzed by AI like a digital tutor with homework energy. Another begged for a full example because the idea sounds promising, but the actual experience is still a little mysterious from the outside. That curiosity quickly turned into skepticism: one commenter flat-out asked where the benchmarks were and whether this is actually better than just making a simple skill the usual way. Translation: cool concept, but show the receipts.

The funniest mini-scandal came from a commenter who looked under the hood and declared the whole thing suspiciously fancy for what seemed like “just a prompt in a bash script (yikes).” Ouch. Meanwhile, another reader zeroed in on the line about people accepting AI code without really understanding it and replied, essentially, “Holy truth.” So the vibe is split between believers, doubters, and exhausted coders realizing their helpful AI assistant may also be enabling some truly elite self-delusion.

Key Points

  • The article introduces Learning Opportunities as a Claude Code and Codex skill that adds optional 10–15 minute learning exercises during AI-assisted coding work.
  • The exercises are described as evidence-based and use learning techniques such as prediction, generation, retrieval practice, and spaced repetition.
  • The repository can be installed as a plugin marketplace in Codex and Claude Code, with plugins including learning-opportunities, learning-opportunities-auto, and orient.
  • An optional automatic prompting feature can prompt users after git commits, with direct support for Linux and macOS and additional setup for Windows.
  • The article argues that AI coding tools can create learning risks related to the generation effect, fluency illusion, spacing, metacognition, and reduced testing and retrieval opportunities.

Hottest takes

"just the following prompt in a bash script (yikes)" — neuralkoi
"No benchmarks and evals present" — zihotki
"Holy truth." — aledevv
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.