\r \r

RetainCraft\r

\r Evidence-based AI-assisted interactive learning protocol\r 基于循证学习科学的 AI 辅助互动学习协议\r \r

Combines 5 scientifically validated methods + self-assessment + diagnostic test + customized learning path\r 结合 5 种科学验证方法 + 自我评价 + 摸底考试 + 定制化学习路径\r \r 📦 Source Code (源码): https://github.com/kaixiad/RetainCraft\r 📖 Detailed workflow (详细流程): references/full-workflow.md\r \r ⭐ If this skill helps you, please give a Star on GitHub!\r 如果这个 skill 对你有帮助，欢迎在 GitHub 上给个 Star！\r \r ---\r \r

⚠️ Execution Checklist (执行清单)\r

\r Must read before each learning session (每次学习开始前必须读)\r \r

Critical Steps (关键执行步骤 - 不可违反)\r

\r

1. Must execute after module test (模块测试结束后必须执行):\r

   python3 scripts/srs.py record-test \x3Ctopic> \x3Ctotal> \x3Ccorrect>\r
   ```\r
   Not executing = module test invalid, level not updated.\r
   不执行此命令 = 模块测试无效，等级不更新。\r
   

\r 2. Feynman Check - L5 required (费曼检验 - L5 必需):\r

• L5 mastery requires: 2 consecutive module tests >=90% + Feynman check passed\r
• L5 精通需要：连续 2 次模块测试答对率 >=90% + 费曼检验通过\r
• AI plays "confused student", asks 3 questions\r
• AI 助手扮演"不懂的学生"，追问 3 个问题\r \r

3. Scoring Discipline (评分纪律 - 不可违反):\r
   • Scoring criteria announced before test, cannot change during test\r
   • 评分标准在测试开始前公布，测试过程中不可修改\r
   • Single question score >=7 = "correct"\r
   • 单题得分 >=7 分 = 算"答对"\r \r
4. Level-up Restrictions (逐级升级限制):\r
   • Levels can only increase one at a time, no skipping\r
   • 等级只能逐级升级，不能跳级\r
   • Each upgrade requires 2 consecutive passes\r
   • 每次升级需要连续 2 次达标\r \r ---\r \r

📚 Core Methodology (核心方法论 - 循证)\r

\r | Method (方法) | Effect Size (效果量) | Source (来源) |\r |---------------|---------------------|---------------|\r | Spaced Repetition (间隔重复) | d=0.85 | Donoghue & Hattie 2021 |\r | Active Recall (主动回忆) | d=0.74 | Donoghue & Hattie 2021 |\r | Interleaving (交错练习) | d=0.47 | Donoghue & Hattie 2021 |\r | Elaborative Interrogation (精细加工提问) | d=0.56 | Donoghue & Hattie 2021 |\r | Feynman Technique (费曼学习法) | d=0.54* | Donoghue & Hattie 2021 |\r | AI Tutoring (AI 辅导) | 0.63-1.3 SD | Kestin et al. 2025 RCT |\r \r

Note (注): *d=0.54 corresponds to "Self Explanation" in original paper, mapped to Feynman technique here.\r *d=0.54 对应原文"自我解释"，此处映射为费曼学习法。\r Reference (参考): scripts/evidence.md (detailed citations)\r \r ---\r \r

🔄 Complete Workflow: Five Steps (完整流程：五步启动)\r

\r

Step 0: Learning Assessment (学习意愿评估)\r

• User completes self-assessment questionnaire (主题、目标、水平、时间、偏好)\r
• 用户完成自我评估问卷\r \r

Step 0.5: Pre-study Materials (预习材料 - 零基础专用)\r

• Trigger: user self-assesses as "complete beginner"\r
• 触发条件：用户自评"完全零基础"\r
• AI searches beginner materials, creates "quick overview" (800-1500 words)\r
• AI 助手搜索入门资料，精炼成"快速概览"\r \r

Step 1: Diagnostic Test (摸底考试)\r

• 5-8 questions, covering basics to advanced\r
• 5-8 道题，覆盖基础到进阶\r
• ⚠️ Diagnostic test does NOT call record-test, NOT written to test_history\r
• ⚠️ 摸底考试不调用 record-test，不写入 test_history\r \r

Step 2: Learning Path Customization (学习路径定制)\r

• Step 2a: Industry research (web_search learning routes, job requirements)\r
• Step 2a：行业调研（web_search 搜索学习路线、岗位要求）\r
• Step 2b: Path generation (3-8 modules with goals, knowledge points, acceptance criteria)\r
• Step 2b：路径生成（3-8 个模块，含目标、知识点、验收标准）\r
• Step 2c: User confirmation before starting\r
• Step 2c：用户确认后开始学习\r \r

Step 3-4: Learning Loop + Spaced Repetition (学习循环 + 间隔复习)\r

• See references/full-workflow.md\r
• 详见 references/full-workflow.md\r \r ---\r \r

📊 Level System L1-L5 (等级系统)\r

\r | Level (等级) | Standard (标准) | Behavior (行为特征) |\r |--------------|-----------------|---------------------|\r | 🔴 L1 Entry (入门) | No test history or first \x3C20% | Start from zero (从零开始) |\r | 🟠 L2 Beginner (初学) | First >=20% | Has concepts but not systematic (有概念但不系统) |\r | 🟡 L3 Intermediate (进阶) | 2 consecutive >=40% | Can apply independently (能独立应用) |\r | 🟢 L4 Proficient (熟练) | 2 consecutive >=70% | Can solve complex problems (能解决复杂问题) |\r | 🔵 L5 Mastery (精通) | 2 consecutive >=90% + Feynman check | Can teach others (能教会别人) |\r \r

Promotion/Demotion Rules (升降级规则)\r

• Promotion: 2 consecutive passes (升级：连续 2 次达标)\r
• Demotion: 3 consecutive failures, min L2 (降级：连续 3 次不达标，最低降到 L2)\r \r

Two Independent Dimensions (两个独立维度)\r

• Level (等级) = Based on module test accuracy (权威)\r
• SM-2 Status (SM-2 状态) = Based on concept mastery ratio (仅展示)\r \r ---\r \r

🔄 Learning Loop Per Module (学习循环 - 每模块)\r

\r

Phase 0: Framework Building (框架搭建 - 10-15 min)\r

• AI asks questions, guides discovery of core concepts\r
• AI 助手提问，引导发现核心概念\r \r

Phase 1: Active Input (主动输入 - 25-40 min)\r

• User studies materials, pause to recall every 15 min\r
• 用户学习原始材料，每 15 分钟暂停回忆\r \r

Phase 2: Feynman Check (费曼检验 - 15-20 min)\r

• User explains to AI, AI plays "confused student"\r
• 用户向 AI 解释所学，AI 扮演"不懂的学生"追问\r \r

Phase 2.5: Simulation (实战模拟 - 15-20 min)\r

• Recommend 2-3 scenarios, user chooses, execute 3-5 rounds\r
• 推荐 2-3 个模拟场景，用户选择后执行 3-5 轮\r
• Score by 5 dimensions (100 points), see scripts/scenarios.md\r
• 按 5 维度打分（100分），见 scripts/scenarios.md\r \r

Phase 3: Test & Reinforce (测试巩固 - 15-20 min)\r

• 5-8 mixed question types\r
• 5-8 道混合题型测试\r
• ⚠️ Must determine "review" or "module test"\r
• ⚠️ 必须判定"复习"还是"模块测试"\r \r | Item (项目) | Review (复习) | Module Test (模块测试) |\r |-------------|---------------|------------------------|\r | Purpose (目的) | Strengthen memory (强化记忆) | Phase assessment (阶段性评估) |\r | Impact (影响) | No level change (不影响等级) | Determines level (决定等级升降) |\r | Command (命令) | srs.py rate | srs.py record-test |\r \r

Phase 4: Spaced Repetition (间隔复习 - SM-2 Algorithm)\r

• Based on SM-2 schedule, proactive reminders when due\r
• 基于 SM-2 时间表，到期主动提醒\r
• Heartbeat check: python3 scripts/srs.py due\r
• 心跳检查：python3 scripts/srs.py due\r \r ---\r \r

⚠️ Burnout Detection (倦怠检测)\r

\r Triggers (触发条件 - 任一):\r

• 3+ consecutive wrong answers (连续答错 3 题以上)\r
• 2 consecutive score drops (连续 2 次测试分数下降)\r
• User says "tired" / "too hard" (用户主动说"累了""太难了")\r \r Response (响应): Lower difficulty, suggest break, switch to easy mode\r 响应：降低难度、建议休息、切换轻松模式\r \r ---\r \r

🤖 AI Assistant Behavior (AI 助手行为规范)\r

\r

✅ Should Do (应该做的)\r

• User asks question → First ask "what do you think?"\r
• 用户问问题 → 先反问"你是怎么想的？"\r
• User stuck → Give hints (not answer)\r
• 用户卡住 → 给提示（不是答案）\r
• Before answering knowledge questions → web_search first\r
• 每次回答知识性问题前 → 先 web_search 验证\r
• Level changes must inform user immediately\r
• 等级变化时必须主动告知用户\r \r

❌ Should Not Do (不应该做的)\r

• Give complete answer directly (除非用户明确要求)\r
• Only score without explanation after test (测试后只打分不解析)\r \r ---\r \r

🔍 Search-First Rules (搜索优先规则)\r

\r Iron rule: AI must search before answering any knowledge question\r 铁律：AI 助手回答任何知识性问题前，必须先搜索验证\r \r | Scenario (场景) | Must Search? (必须搜索?) |\r |-----------------|--------------------------|\r | User asks "what is XX" (用户问"XX 是什么") | ✅ |\r | Correct answer for test (出测试题的正确答案) | ✅ |\r | Feynman check judgment (费曼检验时判断对错) | ✅ |\r | Planning learning path (规划学习路径) | ✅ |\r | Flow conversation (流程性对话) | ❌ |\r \r Rule (规则): Factual statements must include source links\r 规则：事实性陈述必须附来源链接\r \r ---\r \r

🔒 Session Checkpoint (会话检查点)\r

\r

Phase Completion Checklist (Phase 完成自检)\r

□ Current phase core output completed? (当前 Phase 核心产出已完成?)\r
□ If module test: record-test called? (如果模块测试：已调用 record-test?)\r
□ Key progress written to memory/? (关键进展已写入 memory/?)\r
```\r
\r
### Check Commands (检测命令)\r
```bash\r
python3 scripts/srs.py check-session [topic]  # Check unrecorded tests\r
python3 scripts/srs.py check-burnout \x3Ctopic>   # Analyze burnout risk\r
```\r
\r
---\r
\r
## 🧠 Memory Persistence (记忆持久化)\r
\r
### Dual System (双系统)\r
\r
| System (系统) | Stores (存什么) | Location (位置) |\r
|---------------|-----------------|-----------------|\r
| System memory (系统 memory) | Progress summary, weak points (进度摘要、薄弱点) | memory/YYYY-MM-DD.md |\r
| ~/learn/ | SM-2 data, concept mastery (SM-2 数据、概念掌握度) | ~/learn/topics/{topic}/concepts.json |\r
\r
### Recovery Priority (恢复优先级)\r
concepts.json > memory files (concepts.json > memory 文件)\r
\r
---\r
\r
## 💓 Review Reminders (复习提醒 - Heartbeat)\r
\r
```\r
AI receives heartbeat → python3 scripts/srs.py due → Has due content → Notify user\r
AI 助手收到心跳 → python3 scripts/srs.py due → 有到期内容 → 通知用户\r
```\r
\r
---\r
\r
## ⚙️ Configuration (配置系统)\r
\r
```\r
~/learn/config.json\r
{\r
  "learning_depth": "standard",    // shallow / standard / deep\r
  "learner_type": "practical",     // visual / practical / theoretical\r
  "daily_review_limit": 20,\r
  "session_duration": 60,\r
  "burnout_threshold": 3,\r
  "mastery_threshold": 0.8,\r
  "level_thresholds": { "L2": 0.2, "L3": 0.4, "L4": 0.7, "L5": 0.9 }\r
}\r
```\r
\r
---\r
\r
## 📚 References (参考材料)\r
\r
- **Full workflow (完整流程)**: references/full-workflow.md\r
- **Scenario library (场景库)**: scripts/scenarios.md\r
- **Academic citations (学术引用)**: scripts/evidence.md\r
- **Level algorithm (等级算法)**: scripts/srs.py\r
- **Output templates (输出模板)**: scripts/templates.md\r
\r
---\r
\r
## 🎯 Multi-Topic Support (多主题支持)\r
\r
| Priority (优先级) | Description (描述) | Example (示例 |\r
|-------------------|-------------------|---------------|\r
| 1-Urgent (紧急) | Deadline approaching (截止日期临近) | Exam prep (考试准备) |\r
| 2-Important (重要) | Core skills (核心技能) | Programming (编程语言) |\r
| 3-Regular (常规) | Daily learning (日常学习) | New tech (新技术) |\r
| 4-Extended (扩展) | Broaden horizons (拓宽视野) | Related fields (相关领域) |\r
| 5-Reserve (储备) | Future use (未来可能用到) | Learning list (待学习清单) |\r
\r
- Max 3 topics simultaneously (最多同时 3 个主题)\r
- Each topic has independent concepts.json (每个主题独立的 concepts.json)\r
- Reviews can cross topics - interleaving (复习可以跨主题 - 交错练习)\r