Exam Review Agent Skill

exam-review — mistake and confusion review

Purpose

Clear recorded mistakes/confusions before the exam. Replay only existing records; teach no new chapter and invent no question.

Activation

Use at final review or when the student explicitly asks to replay mistakes/find gaps.

Inputs

• study_state.json's mistake_archive and confusion_log when state exists; otherwise generated 「❌ 错题档案」 and confusion compatibility rows.
• references/quiz_bank.json, used to fetch each recorded mistake by exact ID.

Workflow

• Replay recorded mistakes. Reload state (or update_progress.py show), fetch each exact bank item, and ask it again. Never add an unrecorded/bank-external question.

For requires_assets=true or maybe_requires_assets=true, before asking, explaining, hinting, or solving, render every question-side question_context / figure / diagram / table asset, labelled 题面图 or Question-side asset. Only later may solution/review show 答案图 / Answer-side asset. A path is not an image. Preserve but never display student_attempt; its physical path is tainted across the complete quiz, teaching, and content-unit layers, including duplicate official-looking declarations. Missing/unreadable or UI-unrenderable prompt assets cause a fail-closed skip; stub / page_reference also require the original prompt page first. Use scripts/show_question_assets.py for every replay and treat a nonzero result as a skip; do not render a raw bank path directly. See exam-quiz and docs/file-format.md §4.

• Update mistakes. Correct replay → 已订正; still wrong → explain from the stored explanation and retain it.

• Replay confusions. Reload confusion_log; ask the student to restate what/why/how. Correct restatement → 已回顾; vague → explain once and retain 待回顾.

• Persist the open list first. Compile unresolved mistakes plus 待回顾 confusions for the final sprint/exam-cheatsheet. Pipe each conclusion/list to python "${CLAUDE_SKILL_DIR}/scripts/notebook.py" --workspace <ws> add-entry --chapter <ch> --type review --id <slug> --title <gist>; same IDs replace and rebuild the index. Then send a digest and language-pack notebook link. If writing fails, say so and give the full list in chat; file-less clients use chat/text breakpoints.

• Persist row status. If study_state.json is absent and Python works, run update_progress.py --workspace <ws> init first; only when Python truly cannot run may Markdown rows be changed in place. Write results via update_progress.py set-mistake-status / set-confusion-status, and genuinely new rows via add-mistake / add-confusion. A nonzero Python command is a fail-loud error, not permission to hand-edit. Never leave a mastered row stale or overwrite another skill's writes.

Output Contract

• Produce one 「还没拿下的清单」 with current 已订正 / 已回顾 / 待回顾 statuses, notebook link, and refreshed panel.
• Persist conclusions before the digest and persist every status via update_progress.py set-mistake-status / set-confusion-status.
• Student prose is English by default, Simplified Chinese for a Chinese opening, or explicit bilingual blocks.

Language packs

• 中文 → ../../locales/zh/skills/exam-review.md
• English → ../../locales/en/skills/exam-review.md
• 双语 → compose both blockwise, zh then > EN:, under docs/language-policy.md

Display aliases are normalized to zh, en, or bilingual.

Boundaries

• study_state.json is the source of truth; mutate rows only via update_progress.py. Mark replay results via update_progress.py set-mistake-status / set-confusion-status; set-check alone is not a substitute.
• Default question scope is mixed. A restricted scope excludes/counts missing source_type; announce any one-turn override first: 「⚠️ 临时覆盖你的 <scope> 范围偏好」 / ⚠️ Temporarily overriding your <scope> scope preference. Use scripts/select_questions.py.
• Share confusion rows with confusion-tracker; replay only recorded bank items and never erase concurrent writes.