review

name	review
description	知识复习测验 — 从知识库题库中选题，生成交互式 HTML 测验页面。支持智能选题（新题/薄弱/错题/巩固/随机）、session 记录、状态回流。Use when user says "复习", "出题", "测验", "quiz", "test me", "考考我", "random quiz", "复习 FDE", "考一下 transformer", or any request to review/test knowledge.

/review — Knowledge Review Quiz

Purpose

Read questions from questions/, select based on state from state/, render interactive HTML quiz with session recording. After quiz, session JSON flows back through sessions/ → state/ update.

Trigger

• "复习"、"出题"、"测验"、"quiz"、"test me"、"考考我"
• "随机出题"、"random quiz"、"来几道题"
• "复习 FDE"、"考一下 transformer"、"出几道 LLM 的题"
• "错题复习"、"只做错题"、"review mistakes"
• "巩固 FDE"、"新题"
• "记录结果"、"更新状态" (post-quiz state回流)

Execution

Step 0: Ensure question bank exists

Check questions/INDEX.md exists and contains at least one topic with ≥1 question. If not, prompt user to run /ingest to generate questions first.

Step 1: Parse parameters

Extract from user input:

• Topic filter: FDE / LLM / Agent / etc. No filter = all topics
• Question count: Default 10
• Mode:
  • default — 新题40% + 薄弱30% + 随机30%
  • new — 只选 attempts=0 的题
  • wrong — 只选 last_result="wrong" 的题
  • consolidate — 只选 correct/attempts < 0.8 且 attempts ≥ 1 的题
  • random — 纯随机（和旧行为一致）
• Difficulty: Default mixed, supports easy/medium/hard/mixed

Step 2: Load question bank and states

Read wiki/purpose.md for review priority context (which topics matter most). Read questions/bank.json to load all questions in a single read (no per-file I/O). Read questions/INDEX.md to enumerate available topics.

Read all state/*.json files to build in-memory state map:

{
  "fde-kpi-1": {"attempts": 3, "correct": 1, "wrong": 2, "last_result": "wrong", "last_attempt": "2026-05-19"}
}

Questions without a state file are treated as new (attempts=0, correct=0, wrong=0, last_result="", last_attempt="").

Step 3: Filter and select

1. Filter by topic if specified (from INDEX.md)
2. Apply mode rules to score/rank questions:
   • new: keep only attempts=0
   • wrong: keep only last_result="wrong"
   • consolidate: keep only 0 < correct/attempts < 0.8
   • default: 40% new + 30% weakest (lowest correct/attempts) + 30% random from remaining
3. If a mode yields fewer questions than requested, fill with random to hit count (or use all available)
4. Shuffle option order (A/B/C/D labels stay, content is randomized)

Step 4: Render HTML

Read quiz-template.html from this skill directory, replace placeholders:

Placeholder	Replacement
{{QUIZ_TITLE}}	Quiz title (e.g. "FDE 知识复习 · 10 题")
{{QUIZ_SLUG}}	Unique slug for localStorage key (progress recovery)
{{SESSION_ID}}	{date}-{time} e.g. 2026-05-19-153000
{{SESSION_DATE}}	ISO timestamp
{{MODE}}	Selection mode string (for session metadata)
{{TOTAL}}	Total question count
{{QUESTIONS_JSON}}	JSON array of all questions
{{TOPICS}}	Topic filter tag HTML

QUESTIONS_JSON format:

[
  {
    "id": "fde-kpi-1",
    "topic": "FDE",
    "question": "FDE 战略的核心 KPI 是什么？",
    "options": ["降低定制成本", "保持合同规模增长", "提高用户活跃度", "降低边际成本"],
    "answer": 1,
    "explanation": "Bob McGrew 明确指出：FDE 战略与 SaaS PMF 的度量逻辑相反...",
    "source": "fde-kpi-is-contract-growth-not-cost-reduction"
  }
]

Step 5: Save and present

Save to output/quiz-[topic]-[date].html. Open in browser.

When the user clicks "查看结果", the session JSON is automatically copied to clipboard. The user simply pastes (Ctrl+V) into Claude, which recognizes the session JSON and triggers state回流.

Step 6: State回流 (triggered by user after quiz)

When user says "记录结果" / "更新状态" and provides the session file or clipboard content:

1. Read session JSON from sessions/{session_id}.json (or parse clipboard content and save to sessions/)
2. For each answer in the session:
   • Read current state/{id}.json (or init defaults if missing)
   • Update: attempts += 1, correct += 1 (if result="correct"), wrong += 1 (if result="wrong")
   • Set last_result, last_attempt
3. Write all state files concurrently (one per question)
4. Verify: re-read all written state files, confirm counts match expected
5. Report: "Updated N question states. X correct, Y wrong this session. Overall accuracy: Z%."
6. Append to wiki/log.md:

   ## [YYYY-MM-DD HH:MM] review | {topic} · {total} questions | {score}/{total}
   

7. Update wiki/overview.md mastery data for affected topics.

If any state file write fails, report which ones and suggest re-running.

Design Spec

Uses paper-ink design tokens (same :root variables as plan-writer):

Token	Value	Usage
--correct	#5C8D4A	Correct answer
--incorrect	#D4726A	Wrong answer flash
--correct-bg	rgba(92,141,74,0.06)	Correct card background

Interaction rules:

• Click option → immediate verdict, cannot change
• Correct: option turns green + explanation slides down (warm orange left border)
• Wrong: option briefly flashes red → correct option highlighted green + explanation slides out
• Progress dots clickable to jump to any question
• Finish panel: accuracy + per-topic breakdown + [New Set] [Download Session] [Copy to Clipboard]
• Session JSON contains all answers for state回流

Reference

• quiz-template.html — Quiz HTML template (paper-ink design), in this skill directory
• questions/INDEX.md — Static question index (topic enumeration)
• state/*.json — Per-question attempt state
• sessions/*.json — Raw answer WAL
• prompts/generate-questions.md — Question generation prompt (/ingest Stage 5.5)