Tutor
Tutor teaches any subject, technical or not. It finds your real goal and starting point, chooses explanations and examples that fit, and checks understanding instead of assuming it. It belongs to the Learning domain.
When to use it
Choose by what you want; when the session shows a better fit, Tutor offers to switch:
- Explanation: one self-contained answer, complete without exercises.
- Tutoring: you apply the idea or explain it in your own words, and Tutor responds to the misconceptions your answers show. Ordinary exercises belong here.
- Continuing learning: a series, a study plan, or progress kept across sessions. A whole course or examination's material comes either paced, written as you progress, or up front, all written before you start. Tutor takes this from your request or asks once, and records it in the study plan's goal.
When to use something else
- For practice papers that follow a specific written examination's real format and marking, use Examiner.
- Explanations that are only incidental to other work do not need Tutor.
How to start it
Start it yourself with /tutor, or the agent may start it when your request matches. For example (illustrative):
/tutor : I study history, and I am tired of people citing Gödel to prove things the theorem never said. Teach me the first incompleteness theorem from scratch, make me build the self-referring sentence myself, then show me popular claims about it and have me say which ones are wrong.
What happens
Tutor starts once the goal is clear enough and asks only questions that would change the teaching. When your request leaves the kind of help or the depth open, it lets you choose once, and it offers every choice through your agent's choice prompt when it has one, else as a short numbered list. It researches in proportion to how current, contested, or high-stakes the material is, and states uncertainty instead of inventing detail. It uses diagrams, simulations you can steer, or animations only when they explain better than words. It consults Reason before committing to a plan or teaching sequence, after a failure or change of direction, and before presenting an explanation or a diagnosis of your understanding, but not on other tutoring turns, where your next answer shows whether a choice worked.
When an idea is new to you, Tutor starts from what the idea is for and how it works, with a concrete case from your subject, and says where it stops applying. It uses an analogy only as a bridge, saying where the analogy breaks down. In tutoring, it moves from a worked example to a problem you finish to one you solve alone, gives less help as you need less, and ends with a question you answer from memory, with how sure you are, before it shows the answer. It counts something as understood only when more than one task, on a case different from the one it taught with, shows it, and it schedules review as answering from memory rather than re-reading.
A continuing session starts with any review that is due, before new material. When you follow a course and share its materials, Tutor uses their notation and terms. For an examination goal, each topic's name starts with the examination's syllabus area, so the study plan's two sections line up. Work that looks assessed, such as homework or a take-home examination, gets teaching toward it: Tutor teaches the method on a parallel example and guides your own attempt, never writes the submission, and says why once. Work you say is not assessed gets ordinary help.
What you get
An explanation or a tutoring session writes nothing unless you ask to keep something; when one is substantial or would gain from a diagram, simulation, or animation, Tutor offers to keep it as a lesson page, lets you choose whether to include a visual that explains it better than words, and stops offering pages once you decline one. Everything it saves is HTML, never Markdown, apart from runnable code files such as lab starter code. Continuing learning saves files under .learning/<subject>/ in the folder where you started Tutor:
lessons/<NN>-<topic>.html: one self-contained HTML record per lesson, numbered in learning order, holding the teaching material and what you demonstrated;study-plan.html: the one study plan for the subject, shared with Examiner;START_HERE.html: the page to open first. In plain words it says what each kind of file in the folder is, where to pick up, and what to ask to continue. Tutor and Examiner refresh it whenever they save.
Tutor creates each file from a skeleton, so names, numbering, and required sections are right, and after every save it checks that links between your files work and tells you about cited web pages it could not reach. These scripts need Python 3.11 or later.
These are ordinary files you may commit. Delivered teaching is never recorded as understanding; only your own answers and work show it.
When it stops
An explanation ends when it answers your question. A tutoring or continuing-learning session ends by separating what was taught from what you demonstrated, what remains unresolved, and the next step. Tutor promises no automatic reminders.
Why it's built this way
The failure it prevents. "Does that make sense?" is the least useful question in teaching, because the answer is almost always yes. An explanation that is delivered is easily assumed to have landed. Tutor helps a weaker agent teach like a skilled tutor: find the learner's real goal and starting point, choose the explanation that makes the idea land, and check understanding instead of assuming it.
What it refuses, and why. It never records taught material as understanding; only your own answers and work show what you understand, and how clear an explanation felt is not evidence. It uses no mandatory questionnaire, forces no exercises on a direct explanation, and follows no fixed teaching ritual. It never claims you have a learning style, a retention percentage, a brain-based explanation, or a percentage of understanding, because research does not support them. It uses visuals and motion only when they explain something words cannot; animations have playback controls and respect reduced-motion settings. It promises no automatic reminders.
Floor and ceiling. A weaker agent tells an explanation, tutoring, and continuing learning apart, and responds to the misconceptions your answers actually show. It chooses among worked examples, comparisons, analogies, counterexamples, visuals, and practice from memory with a research-grounded guide that says when each helps and when to stop using it. A stronger agent chooses its examples, analogies, demonstrations, and depth of research to fit the subject and the learner.
Related skills
Tutor works alone. When you continue a series, it reads the latest Examiner marking of each attempt and records each misunderstanding or recurring pattern Examiner diagnosed as still open, linked to that marking, in a lesson on every topic it affects, or the first lesson that teaches that topic, until you show it again to Tutor on that topic. It leaves out slips and open questions, and a gap a lesson on that topic already links to a marking of that attempt; a gap you resolved some other way is checked with you again once. It can suggest examination practice. See How skills hand work to each other.