Update

Decide which category the extracted knowledge belongs to.

• If existing category matches → append to that file
• If no match → create new category file + update index.md
• If ambiguous → ask the user

3. Read Existing KB

From ~/.claude/my-knowledge-base/:

1. Read index.md (category listing)
2. Read the relevant category file
3. Check if the same topic already exists

4. Self-Assessment per Topic

NON-NEGOTIABLE: You MUST call the AskUserQuestion tool for this step. Do NOT print options as plain text. Do NOT ask the user to type a number. The AskUserQuestion tool provides interactive selection UI — use it.

For each extracted topic, call AskUserQuestion exactly like this:

{
  "questions": [{
    "question": "How well do you understand <Topic Name>?",
    "header": "Level",
    "multiSelect": false,
    "options": [
      { "label": "Know it well", "description": "Can explain it to someone else confidently" },
      { "label": "Somewhat uncertain", "description": "Got the gist but not confident on details" },
      { "label": "Don't really get it", "description": "Need to revisit this topic" },
      { "label": "Assess me", "description": "Run a quiz to verify my understanding" }
    ]
  }]
}

Mapping:

• "Know it well" → ✅ Know: Record as self-assessed
• "Somewhat uncertain" → ❓ Uncertain: Record as self-assessed
• "Don't really get it" → ❌ Don't know: Record as self-assessed
• "Assess me" → Run Assessment: See Step 5
• "Other" (auto-provided): User can type a custom note — interpret and map accordingly

When there are multiple topics, ask one question per topic sequentially. Do NOT try to batch all topics into a single text prompt.

5. Assessment (only when requested)

Triggered ONLY when:

• User picks option 4 ("Assess me") for a topic
• User explicitly asks for assessment at any point

Assessment flow:

1. Generate 4 questions about the topic
2. NON-NEGOTIABLE: Present ALL 4 questions in a SINGLE AskUserQuestion call. Do NOT print quiz questions as plain text. Do NOT ask the user to type answers. Do NOT call AskUserQuestion multiple times — put all 4 questions in one call.

{
  "questions": [
    {
      "question": "Q1: Webhook과 Polling의 핵심 차이는?",
      "header": "Q1",
      "multiSelect": false,
      "options": [
        { "label": "Webhook이 Polling보다 느리다", "description": "" },
        { "label": "Webhook은 이벤트 발생 시 push, Polling은 반복 조회", "description": "" },
        { "label": "잘 모르겠음", "description": "" }
      ]
    },
    {
      "question": "Q2: Webhook에서 받는 쪽이 해야 하는 일은?",
      "header": "Q2",
      "multiSelect": false,
      "options": [
        { "label": "상대 서버에 반복 요청 보내기", "description": "" },
        { "label": "HTTP POST를 받을 수 있는 URL을 열어놓기", "description": "" },
        { "label": "잘 모르겠음", "description": "" }
      ]
    }
  ]
}

CRITICAL — AskUserQuestion formatting rules (DO NOT VIOLATE):

• All 4 questions in ONE call (AskUserQuestion supports up to 4 questions)
• label = the full answer text that the user sees and selects. NEVER use "A", "B", "C", "D" as labels.
• description = always empty string "". NEVER put answer content in description.
• Last option of each question = "don't know" in the user's language (e.g., "잘 모르겠음" for Korean).
• "Don't know" is always counted as incorrect.
• Use the user's conversation language for both question and label text.
• Each question should have 3 options: 1 correct + 1 wrong + "don't know"

3. Evaluate answers
4. Map score to level:
   • 0-1 correct → ❌ Don't know
   • 2-3 correct → ❓ Uncertain
   • 4 correct → ✅ Know
5. Synthesize a knowledge analysis (NOT raw O/X listing):
   • What the user demonstrably understands (concepts behind correct answers)
   • Where the gaps are (WHY they got it wrong, what concept is missing)
   • This analysis is what gets recorded — not the raw quiz results
6. Record with score + analysis: e.g., ❓ Uncertain (2/4) — Assessed. <analysis>

6. Update KB

Format — self-assessed (no score)

### <Topic Name>
- **Level**: ❓ Uncertain
- 2026-03-22: ❓ Uncertain — Self-assessed. Understood the basics but not confident on edge cases

Format — after assessment (with synthesized analysis)

DO NOT list raw O/X results. Synthesize what the user knows and where the gaps are.

### <Topic Name>
- **Level**: ❓ Uncertain (2/4)
- 2026-03-22: ❓ Uncertain (2/4) — Assessed. Understands the push/pull cycle and conflict resolution flow. Gaps: doesn't yet grasp the upstream tracking concept (why -u is needed on first push) and the relationship between commit and push (that push only sends committed snapshots, not working directory changes).

❌ BAD (raw O/X listing — NEVER do this):

- 2026-03-22: ❓ Uncertain (2/4) — Assessed: commit 단위 전송(✅), 충돌 시 reject(✅), upstream 설정(❌), commit 없이 push(❌)

✅ GOOD (synthesized analysis):

- 2026-03-22: ❓ Uncertain (2/4) — Assessed. Understands that git transfers commits (not files) and knows the conflict→pull→push flow. Gaps: the concept of upstream branch tracking and the prerequisite of committing before pushing.

Existing topic — append timeline entry

NEVER delete existing entries. Append only:

### <Topic Name>
- **Level**: ✅ Know  ← update to latest level
- [initial]: ❌ Don't know — Had never encountered this
- 2026-03-20: ❓ Uncertain — Self-assessed after first reading
- 2026-03-22: ✅ Know (4/4) — Assessed. Solid grasp of all core concepts including upstream tracking and commit-push relationship.

Levels can change in any direction:

• ✅ → ❓ (thought I knew, but was wrong or forgot)
• ❓ → ✅ (confirmed understanding)
• ✅ → ❌ (completely wrong assumption exposed)

If two level changes happen on the same date, add HH:MM to disambiguate.

7. Update index.md

# My Knowledge Base

| Category | File | Entries | Last Updated |
|----------|------|---------|-------------|
| Claude Code | claude-code.md | 12 | 2026-03-22 |

8. Report

Summarize to the user:

• New topics added
• Topics with level changes (before → after)
• Assessment results (if any)
• Total entry count

Rules

• Existing timeline entries are NEVER deleted (append-only)
• Only Level (current) is updated to reflect latest state
• Always use absolute dates (YYYY-MM-DD), add HH:MM only when same-day conflict
• Memos should include "what was learned and in what context"
• Do NOT add content the user didn't explicitly discuss or confirm
• Score (N/5) is ONLY recorded when assessment was performed. Self-assessed entries have no score.
• Assessment is NEVER forced. Only runs on user's explicit choice (option 4) or request.