What is knowledge-teacher?

Knowledge Teacher is a Skill for OpenCode, meaning it is a defined capability an AI coding agent can use, that acts as a teaching assistant. When you ask it to explain a concept, principle, or piece of technology, it adjusts how deep its explanation goes based on your actual knowledge level, and produces a complete, textbook-style teaching document rather than a short answer. The first time you use it, it asks about your background: whether you have zero prior knowledge, some self-taught familiarity, or a formal academic background in the subject. It remembers this so future explanations skip the question and go straight to teaching at the right depth. It also asks how you want the output delivered: as a chat reply, a Markdown file, or a self-contained HTML page with rendered math formulas. Its teaching method follows a set structure: start from what you already know, build an intuitive picture before giving a formal definition, walk through the reasoning step by step without skipping any, ground the idea in a concrete example, and close with a summary. Before writing, it searches for authoritative sources such as Wikipedia, academic papers, and official documentation, then compiles them into a proper reference list and cross-references them in the text. For complex topics that involve many sub-concepts, such as Transformers, it can split the material into a table-of-contents page plus several linked documents instead of one long file. After generating a document, the Skill runs a self-check pass on it, looking for logical gaps, incomplete reasoning, inconsistent terminology, and incorrect formulas before handing over the final result. Example outputs bundled with the repository cover topics like reinforcement learning alignment algorithms, RAG retrieval and reranking, Docker, Linux, and classical mechanics. Installing it means copying the knowledge-teacher folder into an OpenCode project's skills directory, or cloning the repository directly into that location. It requires OpenCode version 1.0.0 or newer and Node.js 18 or newer, with an internet connection needed only if you want it to search for references. The project is released under the MIT license.