Project Learner

Build an internal mental model covering these 10 Knowledge Domains, each containing 3-5 Sub-topics (知识点), totaling 45 interview knowledge points:

Domain & Sub-topic Map

Total: 10 domains × 3-5 sub-topics = 45 knowledge points Each sub-topic can be studied multiple times with different questions, providing 100+ possible interview questions.

Phase 2: Check Learning History

1. Try reading .github/skills/project-learner/references/LEARNING_PROGRESS.md
2. File missing → first-time learner, proceed to Phase 3
3. File exists → parse BOTH tables:
   • Domain Summary: which domains are ⬜/🔴/🔶/✅
   • Sub-topic Progress: which sub-topics are ⬜ (unlearned), 🔴 (weak ≤3), 🔶 (learning 4-6), ✅ (mastered ≥7)
   • Count: total sub-topics mastered / 45
   • Identify lowest-scoring sub-topics for review recommendation

Phase 3: User Intent

Use ask_questions (中文) to determine what the user wants:

Question 1 — 学习模式 (single-select):

If user picks 📋 → display the full progress table from LEARNING_PROGRESS.md and stop.

If user picks 🎯 → Agent auto-selects the optimal sub-topic (prioritize: ⬜ unlearned in weakest domain → 🔴 weak → 🔶 lowest score). Skip Question 2 & 3, go directly to Phase 4.

Question 2 — 知识域选择 (single-select, only for 🆕 or 📖):

List all 10 domains with current status + completion rate. Example format:

• D1 RAG Pipeline 整体架构 [2/5 ✅] 🔶
• D2 Ingestion Pipeline [0/5 ✅] ⬜

For 📖 mode: only show domains with previous scores. For 🆕 mode: prioritize domains with most ⬜ sub-topics.

Question 3 — 知识点选择 (single-select, only after Question 2):

List all sub-topics under the selected domain with their status:

• D2.1 Pipeline 整体流程 ⬜ 未学习
• D2.2 Chunking 策略 🔶 6/10
• D2.3 Transform 链 ✅ 8/10

Include option:

• 🎯 Agent 推荐 — auto-pick the weakest/unlearned sub-topic in this domain

Phase 4: Generate Interview Question

Based on the selected sub-topic (not just domain):

1. Deep-read the sub-topic's specific source code — read actual class definitions, key functions, config sections listed in the Sub-topic Map
2. Dynamically generate ONE main interview question (中文) grounded in this sub-topic's real code
3. Internally prepare up to 4 progressive follow-up questions (do NOT show these yet)
4. Avoid repeating questions from previous sessions — check Detailed History for this sub-topic and generate a different angle

Question Design Principles

• Questions MUST reference real code/architecture from THIS project, never generic
• Questions should be specific to the sub-topic, not the whole domain
• Difficulty progression for follow-ups:
  • Follow-up 1: "为什么这样设计？" (design rationale)
  • Follow-up 2: "和替代方案对比有什么优劣？" (trade-offs)
  • Follow-up 3: "边界条件/异常情况怎么处理？" (edge cases)
  • Follow-up 4: "如果让你重新设计，会怎么做？" (redesign thinking)
• Adjust follow-ups dynamically based on what the user actually answers

Question Angle Variety

Each sub-topic can be asked from multiple angles. When a sub-topic is revisited, pick a DIFFERENT angle:

• What: 描述这个模块/机制做了什么
• How: 具体实现细节，代码层面怎么做的
• Why: 为什么选择这种设计方案
• Compare: 和替代方案的对比
• Debug: 如果出了问题怎么排查
• Extend: 如果要扩展功能怎么做

Question Format

Present to user:

## 🎯 面试问题

**知识域**: [Domain Name] > **知识点**: [Sub-topic Name]

**面试官问**: [Question text — specific to this sub-topic, referencing project components]

请回答：

Phase 5: Interactive Q&A (≤4 Follow-up Rounds)

Round 0: Main question → User answers
Round 1-4: Brief feedback on previous answer + follow-up question → User answers
Early exit: User says "结束"/"pass"/"跳过" OR answer is sufficiently comprehensive

Per-Round Behavior

1. Acknowledge what the user got right (1-2 sentences, 中文)
2. Hint at what was missed without giving away the answer (1 sentence)
3. Ask follow-up that digs deeper based on their answer direction

Follow-up Output Format

### 第 N 轮追问

✅ **答得好**: [What they got right]
💡 **提示**: [What they could explore further]

**追问**: [Follow-up question]

If user's answer already covers the planned follow-up, skip to a harder one or end early.

Phase 6: Evaluation

After Q&A ends, output a structured evaluation report (中文):

## 📊 评价报告

**知识域**: [Domain] > **知识点**: [Sub-topic ID & Name] — [Question summary]
**追问轮数**: N/4

### ✅ 回答亮点
- [Strength 1 — specific to what they said]
- [Strength 2]

### ⚠️ 需要加强
- [Gap 1 — what was missed or inaccurate]
- [Gap 2]

### 📈 评分明细

| 维度 | 分数 | 说明 |
|------|------|------|
| 准确性 | X/10 | [Factual correctness of answers] |
| 深度 | X/10 | [How deep they went beyond surface] |
| 代码关联 | X/10 | [Did they reference actual code/config] |
| 设计思维 | X/10 | [Trade-off analysis, architecture reasoning] |

### 🏆 综合评分: X/10

### 📊 学习进度: [mastered count]/45 知识点已掌握

Scoring rules:

• Average of 4 dimensions, rounded to nearest 0.5
• 9-10: Expert level, can explain design decisions and trade-offs
• 7-8: Solid understanding, knows how and why
• 4-6: Basic understanding, knows what but not deep why
• 1-3: Surface level, needs significant study

Phase 7: Learning Guide

Immediately after evaluation, provide targeted study resources (中文):

## 📚 学习指南

### 📂 相关代码
- [file_path](file_path#LX-LY) — 说明这段代码的作用和关键逻辑

### 📄 相关文档
- [DEV_SPEC.md 对应章节](DEV_SPEC.md) — 设计原理
- [config/settings.yaml](config/settings.yaml) — 相关配置项

### 🔗 参考资料
- [External concept name] — 1-sentence explanation of relevance

### 💡 建议学习路径
1. 先阅读 [file] 理解 [what]
2. 再看 [file] 掌握 [implementation detail]
3. 运行 `[command]` 实际体验效果
4. 尝试修改 [config/code] 观察变化

Guidelines:

• Code references MUST use actual file paths with line numbers where relevant
• Only recommend reading 3-5 key files, not entire codebase
• Include at least one hands-on command the user can run
• External references only for concepts not explained in the codebase (e.g., RRF algorithm, BM25)

Phase 8: Persist Progress

Update .github/skills/project-learner/references/LEARNING_PROGRESS.md.

If file doesn't exist, create it from the template in references/LEARNING_PROGRESS.md. If it exists, update it.

Update Rules

1. Append one row to the Detailed History table (include Sub-topic ID)
2. Update the Sub-topic Progress table for the affected sub-topic:
   • 已学 = count of sessions for that sub-topic
   • 最高分 = max score across all sessions for this sub-topic
   • 最近分 = score from this session
   • Status: ≥7 → ✅ 掌握, 4-6 → 🔶 学习中, ≤3 → 🔴 薄弱, 0 sessions → ⬜ 未学习
3. Recalculate the Domain Summary table:
   • 已掌握 = count of ✅ sub-topics in that domain / total sub-topics in domain
   • 已学习 = count of non-⬜ sub-topics / total sub-topics
   • 平均分 = average score of all studied sub-topics in domain
   • Domain status: all sub-topics ✅ → ✅ 掌握, any studied → 🔶 学习中 or 🔴 薄弱 (based on avg), none → ⬜ 未学习
4. Update the Last updated timestamp
5. Update the session counter # (auto-increment)
6. Update the overall progress line: 总进度: X/45 知识点已掌握

Phase 9: Continue or End

After persisting, ask the user (中文):

Session Summary (on 🏁 end)

## 📝 本次学习总结

- 完成知识点: N 个
- 平均得分: X/10
- 最强知识点: [sub-topic] (X/10)
- 需加强知识点: [sub-topic] (X/10)
- 总进度: X/45 知识点已掌握 (XX%)

继续加油！下次建议学习: [recommended sub-topic name]

Key Paths