A new Claude Code plugin called Learning-Opportunities, created by psychological scientist Dr. Cat Hicks, is asking the question most developers would rather not sit with: are you actually learning anything, or just shipping faster?
The plugin triggers after significant architectural work — new files, schema changes, refactors — and offers optional 10-15 minute exercises built directly from the developer's own codebase. Dr. Hicks calls it an 'adaptive dynamic textbook.' The techniques come from cognitive science: retrieval practice, spaced repetition, the generation effect. In practice that means predictions, teach-it-back prompts, debug scenarios, step-by-step execution traces — things that force active engagement rather than passive acceptance.
The design is a deliberate pushback against five failure modes Dr. Hicks identifies in agentic coding environments. Clean, readable AI output produces fluency illusions — you read it, it makes sense, you merge it, but you couldn't have written it yourself. Passive acceptance skips the active processing that actually builds understanding. Fast loops with no reflection violate spacing principles, essentially cramming at machine velocity. Complete answers from agentic models cut out retrieval practice entirely, the self-testing that makes knowledge durable. And the whole workflow tends to degrade metacognition — your ability to accurately judge what you know versus what you've only seen.
The plugin comes with two companions. Learning-Goal walks developers through MCII — Mental Contrasting with Implementation Intentions — a structured technique for setting learning targets and pre-planning responses to anticipated obstacles. Orient applies empirical research on expert program comprehension to generate orientation lessons for unfamiliar repositories, favouring strategic sampling over exhaustive reading and optionally integrating with Simon Willison's showboat tool to produce executable demo documents. A related but independent project, blendtutor by Dr. Michael Mullarkey, applies a complementary approach to the R ecosystem, building interactive console-based lessons with LLM-powered feedback via Fireworks AI.
What's interesting isn't just the tools themselves but what their existence points to: the Claude Code plugin marketplace is developing a category of software whose purpose isn't to make you faster, but to ensure you're still thinking. That this infrastructure is distributable via namespaced skill invocation and community-shareable manifests suggests the platform is becoming something more than a coding accelerator. Whether developers will voluntarily slow down to use it — in an industry where velocity is the only scorecard that gets counted — remains the open question.