An agent skill for learning programming topics quickly, with the LLM acting as a tutor rather than a code generator. It sets goals, finds a minimal set of resources, designs short timeboxed activities, guides you through them with questions instead of answers, and records what you learned.
It follows the self-directed learning framework from Kim et al. (2014), as applied to LLM-assisted learning by Lin (2023). The approach came out of this blog post, which walks through a real session end to end.
Anyone who wants to understand something rather than just get it done: a concept, a library's internals, an unfamiliar part of a codebase, a technique they'll need again. It works best on tasks just outside your current knowledge.
You: I want to understand how htmlwidgets send events back to Shiny.
Agent: What's your motivation for learning this rather than having me write it?
You: I'll need to do it again and want to know if the code is right.
Agent: Where are you starting from? Have you written an htmlwidget before?
...
Agent: Here's a plan: four activities, about an hour. Activity 1, 20 minutes:
reverse-engineer how DT handles row selection. Start when ready.
The agent asks one question at a time, keeps you to the timeboxes, and lets you do the work.
Results vary with the model. The skill relies on the agent knowing the topic well enough to design activities and check your reasoning, so it does better when the learning is grounded in a codebase the agent can read than when it's a concept the agent has to know from memory. In testing, small or budget models produced wrong code and explanations that a stronger model got right.
Claude Code
/plugin marketplace add thisisnic/rapid-learning
/plugin install rapid-learning@rapid-learning
Claude.ai
Download this repository as a zip, then upload it under Settings, Capabilities, Skills.
Other agents
Copy this folder into wherever your agent looks for skills. The skill is a single SKILL.md following the open Agent Skills format.
CC-BY-4.0. Use it, adapt it, share it, with credit.