note-review-coach

Note Review Coach

Design Principles

This skill is only responsible for making the review loop more reliable, not for forcing question types onto the model. By default, maintain a high degree of freedom: first read the notes and the user's intent, then naturally decide what questions to ask.

Hard constraints are only used to prevent obvious failures:

• Generating questions without reading the material.
• Mislabeling "note recall" as ability training such as reading, translation, or writing.
• Self-test questions revealing answers in advance.
• The test paper being internally inconsistent, for example, question rules, material, options, and answer spaces conflicting with one another.
• Grading not returning to the note evidence.

Do not sacrifice natural judgment just to satisfy this document. A good output can look like a real mock exam paper, or like one-on-one coaching question by question, depending on the user's goal and the note content.

Working Approach

1. Read the material

First determine the review scope, then read the material.

• The current note or a note referenced by the user takes priority.
• If the scope is unclear, there are too many candidates, or the user needs to choose, use ask_user。
• Small documents can be directly read_note; for large documents, first get_note_outline, then read the relevant sections.

When reading, distinguish two types of information:

• Practiceable content: original text, paragraphs, concepts, examples, vocabulary lists, question requirements, scoring criteria, processes, cases, wrong answers, conclusions.
• Design clues: exam point values, question type names, unit numbers, passage numbers, material sources, table of contents, page numbers, teacher hints.

Both types are important, but their uses differ. Design clues can help construct the test paper; whether to directly test them depends on the review contract for this round.

2. Determine the review contract

First judge in your mind what the user actually wants; only ask if it is unclear and would significantly affect the output.

Common contracts:

• Note recall: the goal is to remember the notes themselves, the exam scope, point values, question sources, and key facts.
• Concept mastery: the goal is to understand the concepts, relationships, boundaries, and examples in the notes.
• Ability training: the goal is to practice a certain real task, requiring actionable material, situations, or output requirements for the user.
• Exam/task simulation: the goal is to stay close to real question types and form a set of answerable mock tests.
• Mixed diagnosis: a small amount of scope recall + mainly comprehension/ability questions, used to identify weak points.

If the notes contain both a scope list and practiceable material, and the user only says "review/quiz me," you can use ask_user：

A. 笔记回忆：考范围、结构、关键事实和易忘点
B. 能力训练：按笔记材料练阅读、翻译、写作、推理、应用或输出
C. 混合诊断：少量范围回忆 + 主要理解/能力题，答完批改

If the user has already clearly asked for a "mock exam paper," "follow the real exam," "practice translation/reading/writing," etc., do not ask further; directly generate according to that contract.

3. Generate questions

Question types are naturally determined by the contract and the material; do not apply a template first.

• Note recall: you can test the table of contents, point values, scope, question sources, and key facts, but label it as scope/structure recall.
• Concept mastery: use more explanation, comparison, discrimination, examples, counterexamples, and short answers.
• Ability training: must provide an actionable task carrier and clear output requirements.
• Exam/task simulation: reproduce real input, question-type instructions, answering constraints, and scoring focus as much as possible; when materials are insufficient, simulated materials may be generated, but they must be labeled "simulated/rewritten/generated based on notes".
• Mixed diagnosis: a small amount of scope recall is allowed, but the main body should test comprehension, reasoning, output, and transfer.

A full exam preview may present multiple-choice options in plain text. For per-question interactive answering, multiple-choice questions should preferably use ask_user to collect answers; if a full preview is currently being generated, do not let ask_user interrupt the exam structure.

Question-type references should be read only when self-checking is needed or the user specifies a complex question type:references/question-types.md. Do not mechanically apply them after reading the reference.

4. Pre-output self-check

Before output, perform a quick quality gate:

• Contract consistency: question labels and stems are consistent. Questions labeled as reading/translation/writing/application must not merely ask for source, score, or unit name.
• Clear sources: which parts come from the original notes and which are simulated materials generated based on the notes must be stated clearly.
• Exam paper self-consistency: the number of questions, materials, options, answering rules, and answer space are consistent.
• No answer leaks in self-testing: when the user asks to "test me/self-test", do not provide answers, explanations, or scoring points first; provide them only when the user explicitly asks for an answer version or an exam preview with answers.
• Sufficient materials: ability training must have operable materials, contexts, or output constraints. If materials are insufficient, request supplements, or explicitly downgrade to limited practice based on notes.

5. Grading and summary

After the user answers, grade by question number. Reread the notes when necessary; do not rely only on hidden answers.

Grading for each question should include:

• Judgment: correct / partially correct / incorrect / unanswered.
• Basis: which note, chapter, block, or question material it comes from.
• Gaps: what concepts, evidence, conditions, steps, or expression requirements were missed.
• Corrected answer: provide a concise, learnable version.

In the summary, output mastery status, error-cause map, suggested rereading locations, and directions for the next round of practice.

Write to review record

By default, do not modify the original notes. Only after the user explicitly requests saving, or confirms through ask_user , create a new note.

Recommended title:

复习记录 - <原笔记标题或主题> - <日期>

After saving, 复习 tags may be added; when the user requests synchronization, call sync_note。