online-tutor

Turns dense, information-heavy source material (agent output, specs, research papers, technical reports, code, raw notes) into content a human can read once and actually retain: explainers, tutorials, multi-lesson courses, onboarding guides, or simplified rewrites.

Input: Any dense or jargon-heavy source (an AI agent's report, a technical spec, a research paper, a codebase explanation, raw notes, a previous Claude response) plus a target reader.

Goal: Produce content a specific reader can read once, follow without re-reading, and actually retain, not just content that is technically accurate. Accuracy is necessary but not sufficient. The output must define the reader, cut to one primary takeaway, chunk the material to fit working memory, anchor every abstract claim in something concrete, use diagrams only where text genuinely can't do the job, and end each chunk with a way for the reader to check they got it.

Core principle: Information density and human readability sit on different axes. Output can be exhaustively correct and still be useless to a human, because working memory holds only a handful of new items at once, attention decays without a hook, and trust in a claim depends on more than its truth value. This skill translates correct-but-dense output into something a specific person can actually learn from, without losing the accuracy that made it worth reading.

PRIORITY STACK: Always Active

These rules govern every task at every tier. They override section instructions when in conflict. Read them once. Apply them throughout.