The agent asks the user only questions — never declares facts, never gives answers, never produces explanations. Each question is targeted to advance the user one step toward an insight they will construct themselves. Use when the user explicitly wants to reason something out without being told (debugging an unfamiliar piece of code, working through a design choice, untangling a confusion), or anytime they say "ask me questions" or "don't tell me, ask me".
Performs a formative assessment of the user's current understanding of a topic or piece of code. Output is ALWAYS a discrete Bloom or SOLO level + a concrete gap list + a suggested next skill — NEVER a 1-100 score (false precision, false closure, borrowed confidence). Use when the user wants to check their current level, after one or more learning sessions, at the end of a study sprint, or anytime they say "test me", "where am I at", "how well do I know X", "assess this".
Builds durable understanding by forcing the user to connect a new concept to things they already know — analogies to other languages, prior patterns, similar abstractions in different domains. The agent never volunteers the analogy; the user produces it, the agent stress-tests it. Use when the user is learning something new and wants to anchor it to existing knowledge, when reading code that uses an unfamiliar pattern, when a concept "isn't sticking", or anytime they say "what does this remind me of?" or "connect this to what I know".
Verifies the user's understanding of a piece of code or a topic by making them explain it first, then stress-testing the explanation against the real artefact (code, tests, official documentation). Use when the user wants to deepen comprehension after an LLM has written code, to consolidate understanding of a freshly-studied topic, when they say "make sure I really understand this" or "test my knowledge", or after merging an LLM-written PR.
Produces ASCII flashcards (question on the front, answer on the back, separated by a fold marker) for retrieval practice on a topic or piece of code. Each card targets one atomic fact, anchored to a citation. The user generates the question phrasing; the agent verifies the answer against the real artefact. Use when the user wants reusable retrieval material after a learning session, when capturing facts from a /explain-and-check or /quiz-me session, or anytime they say "give me flashcards" or "I want to drill this".
Generates a file of practical exercises (with solutions hidden behind a clearly marked fold) that target the user's understanding of a topic or piece of code. The user solves the exercises themselves; the agent verifies their solutions against the real artefact. Use when the user wants to consolidate by practising, after a learning session that left only conceptual understanding, when "I get it in theory but I want to try", or anytime they say "give me exercises", "let me try", "learn by doing", "practise this".
Produces atomic, linked markdown notes that capture one idea each, written in the user's own words, and cross-linked by topic. Each note carries a citation to the source artefact (code, docs, prior notes). Use when the user wants to turn a learning session into durable knowledge, when capturing insights from a /explain-and-check or /connect-to-what-you-know session, when "I want notes I can come back to", or anytime they say "save this as notes" or "make me notes".
Forces the user to retrieve knowledge from memory before consulting any source, then verifies what they produced against the real artefact (code, tests, official documentation). Use when the user wants to test what they have actually learned about a topic or a piece of code, to consolidate knowledge after a learning session, after a few days have passed since they last touched a subject, or anytime they say "let me see if I remember", "test my recall", or "let's check what I retained".