Study

Learn a topic in plain words from your own notes, then remember it with short quizzes.

When to use

• You say /study Porter Five Forces, "help me revise NPV" or "explain this to me".
• You say /study quiz or "quiz me". Cards that are due today are asked one at a time.
• You want flashcards from lecture notes or readings that are already in your vault.

Before you start

1. Read vault/80_me/USER.md for your name, what you are learning or working on, and your language.
2. Pick the mode from the first word of the argument:
   • quiz (or "quiz me"): follow workflows/quiz.md straight away. Do not run clarify: quiz mode needs no topic, course, depth or exam date (a topic after quiz only narrows which due cards to ask).
   • anything else: learn mode, workflows/learn.md.
3. Learn mode only: run the clarify step (/clarify, type study), and only for what is not already clear. If the topic is obvious from the request or the vault (and the course, when one applies), skip it and show the one-line summary. Otherwise ask one question at a time, and skip anything you can already see in the vault:
   • Topic: what exactly?
   • Course, only if one applies: which one? Look in vault/20_areas/courses/ first and offer the matches as choices. A topic from work or your own reading needs no course.
   • Depth: quick overview, solid understanding, or exam-ready? (default: solid understanding)
   • Exam date, optional. If there is one, it decides how many cards and how soon.
4. Get today's date from the system, never from memory: node system/scripts/date.mjs (local date).

Steps

Learn mode (/study <topic>)

Full detail is in workflows/learn.md. In short:

1. Clarify (above, only if the topic is unclear, or a course applies and is unclear). Show a one-line summary and wait for a yes.
2. Find the user's sources: course notes in vault/20_areas/courses/<course>/ (when a course applies), source notes in vault/40_sources/notes/, wiki pages in vault/30_wiki/. Search by topic name, synonyms and, when there is one, the course name.
3. Explain the topic Feynman style: a simple analogy, the idea in plain words, then one worked example. Use only what the sources say. Cite every claim: [Source: [[note]] | YYYY-MM-DD | confidence: high|medium|low].
4. Say clearly what is missing. Offer /ingest (add the slides or reading) or research (the researcher agent). Never fill gaps from memory without the label [Inference] or [Unverified].
5. Create 5 to 10 cards in vault/50_learning/cards/<topic-slug>/ (format in workflows/learn.md). Mix recall questions with application questions. Each card links to its source.
6. Schedule: each new card is box 1 and due tomorrow.
7. Add one task for the review day: node system/scripts/tasks.mjs add "Review N cards" --tag study --due <tomorrow> --priority medium --link "50_learning/cards/<topic-slug>" If a "Review N cards" task already exists for that date, tick it and add one with the new total (see workflows/learn.md, step 4).

Quiz mode (/study quiz [topic])

Full detail is in workflows/quiz.md. In short:

1. Collect cards with status: "active" and due on or before today. Limit to the topic if one was given.
2. Ask one card at a time. Show only the question. Wait for the answer.
3. Grade kindly. Show the correct answer and the source link. Count it right if the idea is right, even if the wording differs.
4. Update box and due (schedule below).
5. Summarise: score, strongest and weakest topics, next due date.
6. Update the review task for the next due date.

Leitner schedule (SPEC section 12)

Box the card lands in	Due again after
1	1 day
2	3 days
3	7 days
4	16 days
5	35 days

Right answer: box plus 1 (box 5 stays at 5). Wrong answer: back to box 1. The new due is today plus the days for the new box.

Model routing

• Card writing and explanations: sonnet / medium (this skill).
• Grading answers in quiz mode: the helper-triage agent (haiku / low) is fine. If you delegate grading, send only the question, the correct answer and the user's reply.
• A deep check of a whole card set is a review job: give it to the helper-review agent (sonnet / high, read-only). Read its findings yourself before you tell the user the set is fine.

Outputs

• Cards: vault/50_learning/cards/<topic-slug>/<Card title>.md.
• A task tagged #ab/study: "Review N cards", due on the next review day.
• A short explanation in chat, with citations. Nothing else is saved.
• In quiz mode: updated box and due on each graded card, and an updated review task.

Safety

• Ground the explanation only in the user's sources. Label everything else [Inference] or [Unverified].
• Never invent a citation, a page number or a quote.
• Never delete a card. To retire one, set status: "retired" (ask first).
• This skill helps the user learn. It does not write answers for graded coursework.
• No personal data about other people in cards.
• Cards are written only inside vault/50_learning/cards/.

Extend this

See system/blueprints/study-extras.md: session notes from a syllabus (a whole course is set up by /course), lecture transcript ingest, FSRS scheduling or Anki export, a mastery map and exam mode.