quiz

name

quiz

description

Use when the user wants to test themselves / be quizzed / do active recall on a topic or note in their vault, e.g. "quiz me on Markov state models", "test me on this topic", "quiz me on stuff I haven't reviewed lately". Generates active-recall questions FROM their own notes, asks them interactively (interleaved, with a confidence check), grades answers against the note, classifies each gap, and flags misconceptions. Stamps a review-date for spacing but never writes note prose.

Quiz (active recall)

Test the user on their own notes via active recall, the point is they retrieve the answer, which is what strengthens memory. You generate questions from their notes, ask them interactively (one at a time), and grade their answers against what the note says plus correct domain knowledge.

Core rule: never write note content. The only thing this skill writes is the review-date frontmatter stamp (the reviewed-role key from the profile, a metadata edit, like /vault-mind:connect's links or /vault-mind:fetch's local key), never note prose. Don't offer to "save a summary" or fix notes; if a note is wrong/thin, tell the user so they revise it.

Read the vault profile (.vault-mind/profile.md) first if you haven't this session, it defines this vault's paths and frontmatter schema. If it's missing, run /vault-mind:init. Below, "notes dir", "review-date key", etc. refer to the profile's values.

Procedure

1. Scope the quiz

• Topic/note given (quiz me on X): gather the relevant notes (match names, tags, wikilinks) and any readings the user has marked read on it. These are the source of truth for grading.
• No topic (quiz me on what I haven't reviewed): pick ~5–8 notes by spacing , prioritize the oldest review-date (when they last quizzed it), falling back to the created-date for notes never reviewed. Favor dense/foundational notes. Spread across areas.
• Interleave: when the session spans more than one note/topic, deliberately mix them rather than clustering all questions on one note (e.g. topic A → topic B → A → C…). Interleaving forces discrimination and improves transfer; blocked practice feels easier but retains worse. (Within a single-note quiz this doesn't apply.)
• Tell the user what the quiz covers and roughly how many questions, then start.

Useful (run from the vault root; substitute the profile's dir/key names):

rg -l 'topic' <notes_dir> <resources_dirs>             # files on the topic
rg --files-without-match '^<reviewed_key>:' <notes_dir>/*.md   # notes never reviewed (prime targets)
rg -N '^(<date_key>|<reviewed_key>):' <notes_dir>/*.md         # compare note age vs last review

2. Ask, one question at a time

• Ask a single question, then stop and wait for the answer. Never show the answer with the question, never dump a list of questions.
• Mix cognitive levels so it's real recall, not recognition:
  • Recall: state what a concept claims.
  • Application/transfer: use the idea on a new case.
  • Connection: "How does X relate to [[Y]]?" (leverages the Zettelkasten).
  • Why/derivation: justify or derive, not just state.
• Prefer open-ended over multiple-choice (open-ended forces generation).
• Confidence check: after they answer but before you grade, have them rate how sure they are (low / med / high). This calibration is the point of the next step; don't skip it.
• Adapt difficulty to how they're doing.

3. Grade against the note + reality

After each answer:

• Compare to the note and to correct domain knowledge.
• Be honest and specific: what they nailed, what they missed, what was vague or wrong.
• Use the confidence rating. The dangerous quadrant is high-confidence + wrong: flag those loudly as misconceptions, not slips; they're what active recall exists to catch. Low-confidence + right means it's shaky and worth re-quizzing sooner.
• Classify the gap so they know the action, not just the miss:
  • faded recall: they clearly knew it, couldn't retrieve → re-quiz sooner (spacing).
  • misconception: confidently wrong / wrong model → /vault-mind:feynman it, then revise the note.
  • never-learned: the note doesn't cover it / they never had it → learn it from scratch, then write it.
  • imprecise wording: right idea, sloppy expression → tighten the note's prose.
• If the note itself is wrong, outdated, or thin, flag it: "your note says X but that's imprecise because…", so they can fix it. Do not fix it for them.
• Keep it tight; then move to the next question.

4. Stamp the review (spacing)

For each note actually quizzed, set its review-date key (per the profile) to today's date so future "what haven't I reviewed" sessions space correctly. This is the only write the skill makes, a metadata key, never note body. Preserve frontmatter key order; add the key if absent. (If they're quizzing to find weak spots rather than do a real review, ask before stamping.) If the profile defines no review-date key, say so and offer to add one via /vault-mind:init.

5. Wrap up

End with a short verbal summary (spoken, not written to the vault): strong areas, weak spots grouped by gap type (above), and which notes to revisit or sharpen. Offer to go again on the weak ones, interleaving the misconception/faded items first.

If a weak spot is never-learned (a real knowledge gap, not faded memory), suggest learning it from scratch first (an external from-scratch tutor skill, if installed), then writing/revising the note, then re-quizzing. (This skill only tests what's already in the vault.)