my-session-wrap

my-session-wrap

Step 1: Check changes

Automatically identify files changed in this session.

• git status와 git diff --statCheck the list of changed files with
• If there are no changed files, output "No files were changed today" and exit
• If there are changed files, organize the file names and change summaries and pass them to Phase 1

---

Step 2: Phase 1 — Parallel analysis (run 4 agents simultaneously)

Based on the changes identified in Step 1, run the following 4 agents simultaneously . Each agent is independent of the others, so run them in parallel to save time.

Agent 1: doc-updater

• Role: Look at the changed files and find documents that need updating
• Targets to check: CLAUDE.md, README.md, context files in the agent/ folder
• Update current task checkboxes (required):
  1. CLAUDE.md의 ## 3. 현재 진행 상태 > 당면 과제 Read the section
  2. Compare the work completed in this session with the checkbox items
  3. Change completed items to [ ] → [x]change to
  4. If there are newly identified tasks in this session, add them as [ ]add as
  5. 마지막 업데이트 Update the date to today's date
• Output: "List of documents that need updating + suggested update content + checkbox change details"

Agent 2: automation-scout

• Role: Find recurring patterns in today's work and suggest automation
• Targets to check: repeatedly executed commands, tasks done manually, similar code patterns
• Output: "List of tasks that can be automated + skill/script suggestions"

Agent 3: learning-extractor

• Role: Organize what was learned in this session, problems solved, and new discoveries
• Targets to check: error resolution process, newly created features, applied patterns
• Output: "Today's learning summary (3–5 bullets)"

Agent 4: followup-suggester

• Role: Suggest what to continue in the next session
• Targets to check: unfinished work, areas that can be improved, TODO comments
• Output: "Prioritized list of next tasks"

---

Step 3: Phase 2 — Verification (sequential execution)

Receive the results of the 4 agents from Phase 1 and have duplicate-checker verify them.

Agent 5: duplicate-checker

• Role: Find and organize duplicates or contradictions in the Phase 1 results
• Complete duplicates: If the same content appears more than once, keep only one
• Partial duplicates: Merge similar content into one
• Output: "Final analysis results with duplicates removed"

---

Step 4: Integrate results + user selection

Organize the Phase 1 + Phase 2 results into a single summary and show it.

Output format

📋 세션 정리 결과 (2026-XX-XX)

📁 변경된 파일: X개
[파일 목록]

📝 문서 업데이트 제안 (doc-updater)
- [제안 목록]

⚡ 자동화 기회 (automation-scout)
- [제안 목록]

💡 오늘의 학습 (learning-extractor)
- [학습 요약]

➡️ 다음 할 일 (followup-suggester)
- [우선순위 목록]

Then use AskUserQuestion to have the user select the task to execute:

AskUserQuestion({
  "questions": [{
    "question": "어떤 작업을 실행할까요?",
    "header": "실행 선택",
    "options": [
      {"label": "커밋만", "description": "git add + commit만 실행"},
      {"label": "문서 업데이트 + 커밋", "description": "doc-updater 제안 적용 후 커밋"},
      {"label": "전체 실행", "description": "문서 업데이트 + 커밋 + 자동화 생성"},
      {"label": "건너뛰기", "description": "오늘은 정리 없이 종료"}
    ],
    "multiSelect": false
  }]
})

Step 5: Execute the selected task

Execute only the task corresponding to the option selected by the user.

• Commit only: git add -A && git commit -m "세션 정리: [날짜]"
• Document update + commit: apply doc-updater suggestions (including CLAUDE.md pending-task checkboxes) → commit
• Run all: document update → commit → generate automation script/skill draft
• Skip: output only the summary and exit

When "document update + commit" or "Run all" is selected, always execute:

1. CLAUDE.md update pending-task checkboxes according to doc-updater analysis results
2. 마지막 업데이트 update the date to today's date
3. include the changed CLAUDE.md in the commit

Step 6: Rocky sync (automatic)

After running all options (except Skip), automatically sync the session learnings to Rocky.

1. Extract the actual content from the learning-extractor and followup-suggester results:

   • SYNC_DECISION — Key decisions from this session (1–2 sentences)
   • SYNC_WHY — Why that method/order was chosen
   • SYNC_LEARNING — What Rocky should learn (using learning-extractor results)
   • SYNC_NEXT — Follow-up tasks (using followup-suggester results)

   If extraction is not possible (short session, no learning points, etc.): skip all of Step 6 — do not run the commands below

2. On successful extraction, sync_to_rocky.sh call (Slack notification in the order webhook → bot token):

   export SYNC_DECISION="<핵심 판단>"
   export SYNC_WHY="<이유>"
   export SYNC_LEARNING="<Rocky가 배울 점>"
   export SYNC_NEXT="<다음 할 일>"
   bash scripts/sync_to_rocky.sh "세션 요약 한 줄"
   

   • If exit 2 is returned: content incomplete → output an error message and skip (session wrap continues)
   • Other failures such as SSH errors: output only the error message, wrap completes normally

Rocky sync failure/skip does not prevent session wrap-up.

---

Quick Reference

When it is good to use

• When wrapping up after a long work session (1 hour+)
• Right before switching projects (organizing the current state)
• Organizing before a weekly/monthly retrospective

When it is okay to skip

• When only short questions or lookups were made
• When only reading code without changing it
• When commits have already been completed cleanly

---

Execution Flow

┌─────────────────────────────────────────────┐
│  Step 1. Git 상태 확인 (오늘 뭘 했나?)         │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 2. Phase 1: 병렬 분석 (동시 실행)        │
│  ┌────────────┬────────────┐                 │
│  │doc-updater │automation- │                 │
│  │            │scout       │                 │
│  ├────────────┼────────────┤                 │
│  │learning-  │followup-   │                 │
│  │extractor  │suggester   │                 │
│  └────────────┴────────────┘                 │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 3. Phase 2: 검증 (순차 실행)             │
│  duplicate-checker → 중복 제거                │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 4. 결과 통합 + 사용자 선택               │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 5. 선택된 작업 실행                      │
└──────────────────────┬──────────────────────┘
                       │
┌──────────────────────▼──────────────────────┐
│  Step 6. Rocky 동기화 (SSH → macbook_sync.md) │
│  + Slack #project-hailmary 알림              │
└─────────────────────────────────────────────┘