work-log

Preamble — run this FIRST before any skill logic:

# Find moonkit root via symlink resolution
_MK_DIR=$(cd "$(dirname "$(readlink ~/.claude/skills/work-log 2>/dev/null || readlink .claude/skills/work-log 2>/dev/null || echo .)")/.." 2>/dev/null && pwd) || true
[ -n "$_MK_DIR" ] && _UPD=$("$_MK_DIR/bin/moonkit-update-check" 2>/dev/null || true) && [ -n "$_UPD" ] && echo "$_UPD"

If the output shows UPGRADE_AVAILABLE <old> <new>:

1. Tell the user: "moonkit v{new} 버전이 있습니다. (현재 v{old}) 업데이트할까요?"
2. If yes → run $_MK_DIR/bin/moonkit-upgrade
3. If no → continue. To snooze 24h: echo $(( $(date +%s) + 86400 )) > ~/.moonkit/update-snooze

If preamble produces no output, proceed normally.

/work-log — Work Journal, Impact Report & Achievement Organizer

Three modes:

• Journal mode: Chronological activity log with insights
• Report mode: Internal performance report — 6 business-value categories + impact tagging + TOP 5 + Executive Summary
• Resume mode: Engineer competency framework — 13 perspectives, 1-10 scoring, rewriting suggestions

User-invocable

When the user types /work-log, run this skill.

Arguments

• /work-log — interactive mode & period & source selection
• /work-log last — previous full week journal
• /work-log 2026-01-01 2026-03-22 — custom date range journal
• /work-log Q1 — current year Q1 journal
• /work-log H1 — current year first half (Jan~Jun)
• /work-log report — this week's performance report
• /work-log report Q1 — Q1 performance report
• /work-log report 2025-07-01 2025-12-31 — custom range report
• /work-log resume — this week's resume preparation
• /work-log resume H1 — H1 resume summary
• /work-log setup — re-run first-time setup wizard

HTML output flag

Append --html to any command to also generate a styled HTML version:

• /work-log report Q1 --html
• /work-log resume H1 --html
• /work-log --html (interactive mode — asks as usual, HTML output included)

Mode Detection

Parse $ARGUMENTS to determine:

1. Mode: Parse in this order:
   • report (anywhere in args) → report mode
   • resume (anywhere in args) → resume mode
   • Period token only (e.g., Q1, last, dates) → journal mode
   • Empty arguments (interactive mode) → AskUserQuestion with 3 modes (see Phase 0 Step 2.5). Do NOT silently default.
2. Execution: If user typed in conversation → interactive. If scheduled task → automated.
3. Period: Parse date range, last, Q1-Q4, H1/H2 from arguments. If no period argument is provided, ask interactively (see Phase 0 Step 3). Do NOT silently default to current week.
4. HTML output: If --html is present in args, set html_output = true. Also check html_output in config. The flag overrides config (flag present = true). In interactive mode, ask via Question 4 in Step 2.5.

---

Phase 0: Config & Date Range

Step 1: Read config

cat ~/.work-log/config.yaml 2>/dev/null || cat ~/.weekly-log/config.yaml 2>/dev/null || echo "CONFIG_NOT_FOUND"

Supports both ~/.work-log/ and legacy ~/.weekly-log/ paths. If CONFIG_NOT_FOUND or arguments contain setup, run the First-Run Wizard.

Step 2: First-Run Wizard (interactive only)

Use AskUserQuestion calls:

First call (up to 3 questions):

1. Language: "언어를 선택하세요" — options: "한국어 (ko)", "English (en)"
2. Sources: "어떤 소스를 활성화할까요?" (multiSelect) — "GitHub", "Slack", "Jira", "Google Calendar", "Notion"
3. Output directory: "저장 경로" — "/work-log (Recommended)", "/Documents/work-log", "Custom path"

Second call (scheduling): 4. Organize schedule: "organize 모드를 자동으로 실행할까요?" — options: "매주 자동 실행 (토요일 10시) (Recommended)", "수동으로만 실행", "커스텀 스케줄"

• If "매주 자동 실행": set organize.schedule.enabled: true, cron: "0 10 * * 6"
• If "커스텀 스케줄": ask follow-up for cron expression and submode (report/resume)
• If "수동으로만 실행": set organize.schedule.enabled: false

Generate ~/.work-log/config.yaml using the skill's config.example.yaml as template.

Step 2.5: Interactive mode & period selection (when arguments are empty)

Only runs when $ARGUMENTS is empty AND interactive mode.

Use a single AskUserQuestion call with up to 4 questions to collect mode, period, sources, and organize submode at once:

Question 1 — Mode:

• Question: "어떤 용도로 생성할까요?"
• Options:
  • "업무일지 (Journal)" — 시간순 활동 기록 + 인사이트 (Recommended)
  • "성과 보고 (Report)" — 경영진 시각의 6개 카테고리 + 임팩트 태깅 + TOP 5 + Executive Summary
  • "이력서 정리 (Resume)" — 엔지니어 역량 프레임워크 13개 관점 + 1-10 스코어링 + 리라이팅 제안

Question 2 — Period:

• Question: "기간을 선택하세요"
• Options (show up to 4, prioritize by relevance):
  • "이번 주 (월요일 ~ 오늘)"
  • "지난주 (월 ~ 일)"
  • "이번 달" / "지난달"
  • "Q1 (13월)" / "Q2 (46월)" / etc. — show only the current and previous quarter
  • "H1 (16월)" / "H2 (712월)" — show only the current and previous half
• AskUserQuestion allows max 4 options, so pick the most relevant 4 based on the current date. Always include "이번 주" and at least one long-range option (Q or H). The user can always select "Other" for custom dates (YYYY-MM-DD ~ YYYY-MM-DD).
• If user selects "Other": ask a follow-up for custom dates, including the target year if it differs from the current year (e.g., "2025-07-01 ~ 2025-12-31").

Question 3 — Sources:

• Question: "어떤 소스를 사용할까요?"
• multiSelect: true
• Options: Show only the sources that are enabled: true in config, pre-labeled with checkmarks. For example if all 5 are enabled:
  • "GitHub" — 커밋, PR, 코드 리뷰
  • "Slack" — 채널별 대화 토픽
  • "Jira" — 티켓 상태, 에픽
  • "Google Calendar" — 미팅, 주최 여부
  • "Notion" — 페이지 생성/편집 이력
• If user deselects a source, skip that source in Phase 1 data collection. This lets users exclude noisy or irrelevant sources per run without changing config.

Question 4 — HTML output:

• Question: "HTML 파일도 함께 생성할까요?"
• Options:
  • "Markdown만 (Recommended)" — .md 파일만 생성
  • "Markdown + HTML" — .md와 styled .html 파일 모두 생성
• If html_output: true in config, default to "Markdown + HTML".
• If --html flag was passed in arguments, skip this question and set html_output = true.

In automated mode (scheduled task), skip this step — default to journal mode + current week + all enabled sources + config's html_output setting.

Step 2.7: Partial answer recovery (when AskUserQuestion is rejected)

If AskUserQuestion is rejected but partial answers are visible in the rejection message:

1. Accept all clearly answered questions — do not re-ask them.
2. Only re-ask unanswered questions — either via a single follow-up AskUserQuestion or by inferring from context (e.g., if all other answers are clear, ask only the missing one).
3. Never ask "어떤 부분을 clarify 하고 싶으신가요?" — the user didn't ask to clarify; the system reported a rejection. Proceed with what you have.
4. If the user's text response after rejection provides the missing answer (e.g., "html 만들어줘"), accept it and proceed immediately.

Step 3: Calculate date range

If a period argument was provided, parse it directly:

• Two dates (YYYY-MM-DD YYYY-MM-DD) → use directly
• last → previous Monday to Sunday
• Q1~Q4 → quarter boundaries of current year
• H1 → Jan 1 ~ Jun 30, H2 → Jul 1 ~ Dec 31

If NO period argument was provided (interactive mode only), use AskUserQuestion to ask:

• Question: "기간을 선택하세요"
• Options:
  • "이번 주 (월요일 ~ 오늘)" — calculate dynamically
  • "지난주 (월 ~ 일)"
  • "이번 달"
  • "지난달"
  • "Q1 (13월)" / "Q2 (46월)" / etc. — show only the current and previous quarter
  • "H1 (16월)" / "H2 (712월)" — show only the current half
  • "직접 입력 (YYYY-MM-DD ~ YYYY-MM-DD)"
• If user selects "직접 입력": ask a follow-up AskUserQuestion for start and end dates.
• If user selects "이번 달": start = first day of current month, end = today.
• If user selects "지난달": start = first day of previous month, end = last day of previous month.

In automated mode (scheduled task), default to current week without asking:

DOW=$(date +%u)
WEEK_START=$(date -v-$((DOW-1))d +%Y-%m-%d)  # macOS
WEEK_END=$(date +%Y-%m-%d)

Step 4: Ensure output directories

OUTPUT_DIR=$(grep 'output_dir:' <config_path> | awk '{print $2}' | sed "s|~|$HOME|")
mkdir -p "$OUTPUT_DIR/journals" "$OUTPUT_DIR/snapshots" "$OUTPUT_DIR/impact-reports" "$OUTPUT_DIR/pending-resume-updates"

---

Phase 1: Data Collection

IMPORTANT: Use parallel sub-agents for data collection. Launch one Agent per enabled source (up to 5 agents: GitHub, Slack, Jira, Calendar, Notion) running in the background simultaneously. Each agent collects its source data and returns a structured summary. Wait for all agents to complete, then merge results for Phase 2.

This dramatically reduces total collection time — especially for long periods where each source may take 30-60 seconds.

Collect from each enabled source independently. If a source fails, log and continue.

For long periods (>2 weeks): First check for existing snapshots and journals in the output directory. Use them as primary data to avoid redundant API calls. Only collect fresh data for uncovered date ranges.

ls "$OUTPUT_DIR/snapshots/"*.json 2>/dev/null | sort
ls "$OUTPUT_DIR/journals/"*.md 2>/dev/null | sort

If existing data covers the requested period, read those files instead of re-collecting. If gaps exist, collect only for the missing date ranges.

1a. GitHub (if enabled)

Use gh CLI. Auto-detect username via gh api user --jq '.login'.

For short periods (≤4 weeks): query the full range at once.

For long periods (>4 weeks): split into monthly sub-ranges to avoid GitHub Search API limits (100 commits, 50 PRs per query). For example, a 6-month range becomes 6 monthly queries. Aggregate and deduplicate results across all sub-ranges.

For commit discovery, prefer gh search commits (works with private repos):

# Short period:
gh search commits --author={username} --author-date=">={START}" \
  --json repository,sha,commit --limit 100
# Long period (repeat per month):
gh search commits --author={username} --author-date="{MONTH_START}..{MONTH_END}" \
  --json repository,sha,commit --limit 100

For PRs:

gh search prs --author={username} --created=">={START}" \
  --json repository,title,number,state,url,createdAt --limit 50
# Long period: use --created="{MONTH_START}..{MONTH_END}" per month

For reviews:

gh search prs --reviewed-by={username} --updated=">={START}" \
  --json repository,title,number,state,url --limit 50
# Long period: use --updated="{MONTH_START}..{MONTH_END}" per month

Apply repos_include / repos_exclude filters from config.

1b. Slack (if enabled)

Use slack_read_user_profile MCP (no params) → get user_id. Then slack_search_public_and_private:

query: "from:<@{user_id}> after:{WEEK_START} before:{WEEK_END}"
limit: 20, sort: "timestamp", include_context: false

Apply channels_include/channels_exclude filters. Group by channel, extract topics.

1c. Jira (if enabled)

Try Atlassian MCP first (getAccessibleAtlassianResources → searchJiraIssuesUsingJql). JQL: assignee = currentUser() AND updated >= "{WEEK_START}" Fall back to REST API if MCP unavailable.

1d. Google Calendar (if enabled)

Use gcal_list_events MCP. Filter: accepted, tentative, or organizer. Skip declined. Extract: title, time, duration, attendee count.

1e. Notion (if enabled)

Use notion-search MCP. Extract: title, URL, edit time.

API limitations to work around:

• query must be non-empty (use a space if no keyword).
• Returns max 25 results per query, ranked by semantic relevance — NOT by date.
• created_date_range filter applies to creation date, but the returned timestamp is last edited time. These can diverge significantly.
• For comprehensive coverage, run multiple searches with different keywords relevant to the user's work (e.g., project names, "review", "design", "onboarding", product names). Deduplicate by page ID.
• Post-filter results to keep only pages whose last_edited_time falls within the requested date range.

---

Phase 2: Output Generation

Branch based on mode:

• Journal mode → Phase 2A (Journal)
• Report mode → Phase 2B (Report)
• Resume mode → Phase 2C (Resume)

Journal mode runs Phase 2H (HTML, if enabled), Phase 3 (Brag Doc), Phase 4 (Resume Update), Phase 5 (Snapshot). Report and Resume modes skip Phase 3 and Phase 4, run Phase 2H (HTML, if enabled), Phase 5 (Snapshot) and Phase 6 (Completion).

---

Phase 2A: Journal Generation

Generate chronological activity journal. Read references/journal-template.md for the full template structure including the Insights section (Growth Metrics, Skill Radar, What to Improve, Strategic Direction, Owner's Scorecard).

Save to: {OUTPUT_DIR}/journals/{END}.md

---

Phase 2B: Report Generation

Read references/report-template.md for the full template structure, categorization rules, and auto-categorization signals.

Take all collected data from Phase 1 and categorize each activity into one of 6 business-value categories: 매출/딜리버리, 운영 효율화, 장애 & 보안 리스크 감소, 비용 절감, 채용 & 홍보, 팀 빌딩/리더십.

Also tag each activity with impact metrics where possible:

임팩트 태그	키워드/시그널
💰 매출	customer, partner, delivery, release, launch, 고객, 파트너, 납품
💰 비용	optimize, reduce, automate, remove dependency, 최적화, 비용, 절감
⏱ 시간	speed up, reduce time, pipeline, CI/CD, 시간 단축, 자동화
🛡 리스크	fix, security, bug, crash, validation, 장애, 보안, 수정
✅ 품질	test, refactor, clean, lint, 테스트, 리팩토링, 품질

Use the keyword signals and source-based defaults from the template to auto-categorize. For ambiguous items, pick the category with the strongest signal match. If impact is quantified in the commit message or PR (e.g., "reduced from 50s to 22s") → use that number directly.

Generate the report using the Korean or English template based on the config language setting.

Save to: {OUTPUT_DIR}/reports/{END}-report.md

---

Phase 2C: Resume Generation

Read references/resume-template.md for the full competency framework, 13 evaluation perspectives, scoring rubric, and rewriting rules.

Take all collected data from Phase 1 and classify each activity into the most relevant perspective among the 13 perspectives across 4 areas:

• 기술/코드/설계 (7 perspectives)
• 제품 (2 perspectives)
• 커뮤니케이션 (2 perspectives)
• 문화 (2 perspectives)

For each perspective with matching activities:

1. List activities with source links
2. Generate a narrative: problem recognized → how it was solved → result/impact
3. Score each item 1-10 using the scoring rubric
4. For items scoring < 7: generate a stronger rewriting suggestion

Interactive mode: If activity data alone is insufficient for a complete narrative, use AskUserQuestion to ask the user for context (batch 3-4 items per question).

Automated mode: Generate narratives from data only. Mark insufficient items with "[추가 컨텍스트 필요]" placeholder.

Generate the output using the Korean or English template based on the config language setting.

Save to: {OUTPUT_DIR}/resume/{END}-resume.md

---

Phase 2H: HTML Generation (optional)

Skip this phase if html_output is false.

After generating the markdown output (Phase 2A, 2B, or 2C), convert it to a styled HTML file. Read references/html-template.md for the design system and structural mapping rules.

Process

1. Read the generated markdown file.
2. Generate a self-contained HTML file (inline CSS, no external dependencies except Google Fonts).
3. Save to the same directory as the markdown file, with .html extension:
   • Journal: {OUTPUT_DIR}/journals/{WEEK_END}.html
   • Report: {OUTPUT_DIR}/reports/{WEEK_END}-report.html
   • Resume: {OUTPUT_DIR}/resume/{WEEK_END}-resume.html
4. Open the HTML file in the browser: open {html_path}

Design principles

• Self-contained: All CSS inline. Only external dependency is Google Fonts (Noto Sans KR + Inter).
• Mode-aware styling: Each mode has a distinct color accent — Journal (blue), Report (gradient blue-purple), Resume (green).
• Responsive: Works on mobile and desktop. Print-optimized with @media print.
• Structural mapping: Map markdown structure to semantic HTML components:
  • # H1 → header block with gradient background
  • ## H2 → section header with numbered badge
  • ### H3 → card with left border accent
  • **Bold label**: value → styled key-value pair
  • Bullet lists → styled list with colored dot indicators
  • Tables → responsive styled tables
  • **임팩트** / **Impact** blocks → highlighted impact box
  • **증거** / **Evidence** → subtle reference line
• Stats bar: Extract key numbers (PRs, commits, meetings, etc.) from the content and render as a top-level stats card row.

---

Phase 3: Brag Document Update

Note: Phase 3 is skipped in organize mode.

Read existing {OUTPUT_DIR}/brag-doc.md. Extract accomplishments from the generated journal or impact report:

• PRs merged (especially features and significant changes)
• Jira tickets completed
• Meetings led (user was organizer)
• New Notion pages created
• Notable Slack threads initiated

Transform into: action verb + what + result/impact format. Append under ### {WEEK_END} heading. Do NOT duplicate existing items.

---

Phase 4: Resume Update

Note: Phase 4 is skipped in organize mode. The organize resume submode generates its own resume-ready output but does NOT modify the resume file directly.

Read resume at resume_path from config. Detect format (.tex or .md).

Interactive mode: Use AskUserQuestion to let user select which brag items to add. Format-aware insertion (LaTeX: \SmallBulletItem + \Impact{}, Markdown: bullet points).

Automated mode: Write candidates to {OUTPUT_DIR}/pending-resume-updates/{WEEK_END}.md.

---

Phase 5: Snapshot

Save JSON to {OUTPUT_DIR}/snapshots/{WEEK_END}.json:

{
  "date": "{WEEK_END}",
  "range": { "start": "{WEEK_START}", "end": "{WEEK_END}" },
  "mode": "journal|impact|organize",
  "organize_submode": "report|resume|null",
  "sources": {
    "github": { "collected": true, "commits": 0, "prs": 0, "reviews": 0 },
    "slack": { "collected": true, "messages": 0 },
    "jira": { "collected": true, "tickets": 0 },
    "calendar": { "collected": true, "meetings": 0 },
    "notion": { "collected": true, "pages": 0 }
  },
  "failed_sources": [],
  "output_path": "{path to journal, impact report, or organize output}",
  "html_path": "{path to .html file, or null if html_output is false}",
  "brag_items_added": 0
}

---

Phase 6: Completion Report

## work-log 완료

**기간**: {START} ~ {END}  |  **모드**: {journal / impact / organize-report / organize-resume}

| 소스 | 상태 | 수치 |
|------|------|------|
| GitHub | ✅ | 커밋 {N}, PR {N}, 리뷰 {N} |
| Slack | ✅ | 메시지 {N} |
| Jira | ⬚ 비활성 | — |
| Calendar | ✅ | 미팅 {N} |
| Notion | ❌ 실패 | {reason} |

**저장된 파일:**
- {journal / impact report / organize output}: `{path}`
- HTML: `{html_path}` (html_output이 true일 때만 표시)
- Brag doc: {N}개 항목 추가 (organize 모드에서는 생략)
- 이력서: {업데이트됨 / pending / 변경 없음} (organize 모드에서는 생략)

For organize mode, the completion report additionally shows:

• 서브모드: report / resume
• 카테고리 수: {N}개 카테고리에 활동 분류됨
• 이력서 스코어 (resume submode only): 평균 {N}/10, 리라이팅 제안 {N}건

Resume mode: 영문/한글 복사용 포맷 + 프리뷰 (4개 섹션)

Resume 모드의 완료 보고 시, 위 테이블 아래에 아래 4개 섹션을 모두 대화에 출력한다:

1. ## 복사용 포맷 (English) — 영문 원본 코드블록 (LaTeX 또는 Markdown)
2. ## 복사용 포맷 (한국어) — 한글 원본 코드블록 (동일 매크로, 내용만 한국어)
3. ## 프리뷰 (English) — 영문 원본을 읽기 쉬운 Markdown으로 렌더링
4. ## 프리뷰 (한국어) — 한글 원본을 읽기 쉬운 Markdown으로 렌더링

Markdown 파일(.md)과 HTML 파일(.html)에도 동일하게 4개 섹션을 모두 포함한다. 4개 중 하나라도 빠지면 안 된다. 상세 규칙은 resume-template.md, html-template.md 참조.

---

Error Handling

• ALL sources fail → suggest checking config/auth
• Config syntax error → suggest /work-log setup
• Resume not found → skip Phase 4, mention in report
• Output dir not writable → alert immediately
• Never crash silently

Organize-specific errors

• 기간 내 수집 데이터 0건 → "해당 기간에 수집된 활동이 없습니다" + 소스별 실패 사유 표시
• 인식 불가 서브모드 토큰 → period로 파싱 시도, 실패 시 에러 메시지 "'{token}'은 유효한 서브모드나 기간이 아닙니다"
• resume 스코어 생성 실패 → 기본값 5/10 사용 + 로그
• automated 모드에서 서브모드 미지정 → 기본값 report 사용