my-history-insight

my-history-insight

Find patterns in past Claude Code session records and extract insights. It plays a role like "taking out a diary and looking back on this month".

Execution Flow

┌─────────────────────────────────────────────┐
│  Step 1. 세션 파일 탐색                        │
│  ~/.claude/projects/ 에서 .jsonl 파일 목록     │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 2. 사용자 메시지만 추출                  │
│  (전체의 ~6% — 나머지 94%는 부속 자료)         │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 3. 병렬 분석 (3개 에이전트 동시)         │
│  ┌──────────────┬──────────────┐             │
│  │session-      │pattern-      │             │
│  │summarizer   │finder        │             │
│  ├──────────────┴──────────────┤             │
│  │growth-tracker               │             │
│  └─────────────────────────────┘             │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 4. 인사이트 리포트 출력                  │
└─────────────────────────────────────────────┘

Step 1: Explore session files

Check the list of session files in the current project.

Location: ~/.claude/projects/ subfolder
File format: UUID-style .jsonl files
Exclude files inside the subagents/ folder (subagent records)
Determine the analysis scope by identifying the number of files and total size

If there are too many files (20 or more), analyze only the most recent 10.

Step 2: Extract user messages

각 .jsonl Extract only the actual conversation content from the files.

Each line of a jsonl file is a JSON object
"role":"user" Extract only the content of items that are
Exclude all tool_use, tool_result, and system messages
Identify the topic of the session from the extracted text

Key point: ~94% of session files are auxiliary material such as file change history and tool execution logs. The actual user conversation is only ~6%, so extract only this and analyze quickly.

Step 3: Parallel analysis (run 3 agents simultaneously)

Three agents analyze the extracted conversation content simultaneously.

Agent 1: session-summarizer

Role: Summarize in one line what kind of work each session was
Output: A list of summaries per session (date + topic)

Agent 2: pattern-finder

Role: Detect recurring patterns across all sessions
Items to check:
- The most frequently performed type of work (automation, documentation writing, API integration, etc.)
- Recurring errors or retry patterns
- Frequently used tools or technologies
Output: Top 3 patterns + evidence

Agent 3: growth-tracker

Role: Track changes and growth over time
Items to check:
- Changes in task complexity between early sessions vs recent sessions
- Newly added technologies or tools
- Traces of recurring tasks having been automated
Output: 2~3 growth points

Step 4: Output insight report

Integrate the results of the 3 agents and output in the format below.

📊 히스토리 인사이트 리포트
분석 세션: XX개 | 기간: YYYY-MM-DD ~ YYYY-MM-DD

━━━━━━━━━━━━━━━━━━━━━━━━

📅 세션별 요약 (최근 순)
- [날짜] 세션명: 한 줄 요약
- ...

━━━━━━━━━━━━━━━━━━━━━━━━

🔍 반복 패턴 Top 3
1. [패턴명]: 설명
2. [패턴명]: 설명
3. [패턴명]: 설명

━━━━━━━━━━━━━━━━━━━━━━━━

📈 성장 트래킹
- [성장 포인트 1]
- [성장 포인트 2]
- [성장 포인트 3]

━━━━━━━━━━━━━━━━━━━━━━━━

💡 핵심 인사이트
[전체를 관통하는 핵심 한 문장]

Analysis scope options

When running the skill, have the scope selected via AskUserQuestion:

AskUserQuestion({
  "questions": [{
    "question": "어떤 범위로 분석할까요?",
    "header": "분석 범위",
    "options": [
      {"label": "현재 프로젝트만", "description": "지금 작업 중인 프로젝트 세션만 분석"},
      {"label": "최근 10개 세션", "description": "가장 최근 10개 세션만 빠르게 분석"},
      {"label": "전체 세션", "description": "모든 Claude Code 세션 분석 (시간 더 소요)"}
    ],
    "multiSelect": false
  }]
})

Quick Reference

When: Monthly retrospective, when switching projects, when wondering "what have I been doing?"
Time required: ~2 minutes (based on the current project)
Output file: sync/history-insight-YYYY-MM-DD.md (optional)