session-worklog

context	fork
disable-model-invocation	true
name	session-worklog
description	Automatic session work journal maintained in auto-memory, recording tasks, decisions, and pending items for recovery. 자연어 트리거: '작업 기록해줘', '세션 로그 남겨줘', '오늘 한 일 정리', '결정 사항 기록'.
lang	["en"]
level	2
triggers	["worklog","session log","작업 기록","세션 로그","기록"]
tokens	~500
category	workflow
agents	["orchestrator","doc-updater"]
platforms	["claude-code","gemini-cli","codex-cli","cursor"]
source_hash	0b7b471e
whenNotToUse	Mid-task execution where logging would interrupt flow; do not invoke during active problem-solving — reserve for session boundaries and natural breakpoints.

Session Worklog

Lightweight, auto-maintained work journal that persists across sessions.

When This Skill Applies

• After completing a significant block of work (commit, feature, fix, config change)
• Before a session ends or context compresses
• When the user explicitly asks to log work
• Automatically triggered by commit/push operations

Worklog Location

~/.claude/projects/<project-slug>/memory/worklog.md

This file is in the auto-memory directory, so it:

• Persists across sessions
• Is automatically loaded at session start
• Provides immediate context recovery

Entry Format (STRICT)

Each session entry follows this exact format. Keep entries concise — goal is ~10-15 lines per session.

## YYYY-MM-DD | [1-line summary of main work]

### 작업
- [completed task 1]
- [completed task 2]

### 결정
- [decision]: [rationale in 1 phrase]

### 보류
- [pending item or blocker, if any]
- (없음) if nothing pending

---

Rules

Auto-Append Trigger

Append a worklog entry when ANY of these occur:

1. User requests a commit or push
2. A major feature/fix/change is completed
3. The session is ending (context approaching limit)
4. User explicitly says "기록해" or "log this"

Size Management

• Max entries: 10 sessions
• Max lines: 200 lines total
• Auto-trim: When appending would exceed 200 lines, remove the OLDEST entry (top of file) first
• Header preserved: The # Session Worklog header and description line are NEVER deleted

Content Rules

• Concise: Each bullet = 1 line, no paragraphs
• Factual: What was done, not how or why in detail
• Decisions with rationale: Brief "because" after each decision
• File counts, not file lists: "23 agent files updated" not listing all 23
• Korean preferred: Match user's communication language

Token Budget

• Writing entry: ~300-500 tokens (reading file + appending)
• Reading at session start: ~200 tokens (file is small)
• Total per session: ~500-700 tokens maximum

Example Entries

## 2026-02-27 | 모델 정책 opus 중심 재편 + 워크플로우 개선

### 작업
- 모델 정책: haiku 제거, opus 73%(19개) / sonnet 27%(7개)
- DEV 프로토콜 도입 (sc, implement, improve, build)
- vibe-coding 스킬 신규 생성

### 결정
- e2e-runner opus 승격: 테스트 품질 중요
- 문서화 방식: Session Worklog 채택 (자동+경량)

### 보류
- (없음)

---

## 2026-02-26 | Sprint 7 마무리 + 스킬 레퍼런스

### 작업
- 49 professional reference files 추가 (24 marketing skills)
- 깨진 스킬 참조 45개 제거, 23 빈 디렉토리 삭제
- ESLint 0 errors 0 warnings 달성

### 결정
- 레퍼런스 아키텍처: 45-75줄 3티어 구조 (원칙→테이블→체크리스트)

### 보류
- (없음)

---

Output Template

## YYYY-MM-DD | [1-line summary of main work]

### 작업
- [completed task 1]
- [completed task 2]

### 결정
- [decision]: [rationale in 1 phrase]

### 보류
- [pending item or blocker]
- (없음) if nothing pending

---

Output Template

## YYYY-MM-DD | [1-line summary of main work]

### Tasks
- [completed task 1]
- [completed task 2]
- [completed task 3]

### Decisions
- [decision]: [rationale in 1 phrase]
- [decision]: [rationale]

### Pending
- [pending item or blocker]
- (none) if nothing pending

### Retrospective (optional - on significant sessions)

START (new practices to adopt):
  - [practice to adopt]

STOP (practices to discontinue):
  - [practice to stop]

CONTINUE (practices working well):
  - [practice working well]

### Session Metrics (optional)
- Files changed: [n]
- Tests added: [n]
- Coverage delta: [+/-n%]
- Duration: ~[n]h

Integration with Other Skills

• checkpoint: /checkpoint creates a detailed snapshot; worklog creates a lightweight summary
• continuous-learning: Worklog entries can feed pattern extraction
• vibe-coding: After DEV protocol completion, auto-append worklog if significant work done
• git-unified (workflow): Commit/push triggers worklog append (see references/workflow.md)

Workflow Checklist

Copy this checklist and track progress:

Progress:
- [ ] Step 1: Detect trigger (commit, feature complete, session end, explicit request)
- [ ] Step 2: Read existing worklog.md
- [ ] Step 3: Compose entry (작업/결정/보류 sections, ~10-15 lines)
- [ ] Step 4: Check size limit (max 200 lines, 10 entries)
- [ ] Step 5: Auto-trim oldest entry if limit exceeded
- [ ] Step 6: Append new entry to worklog.md

Human Checkpoints

Checkpoint 1: 엔트리 내용 확인 (After Step 3)

Context: 세션의 작업·결정·보류 항목이 작성된 시점. 자동 생성된 요약이 실제 세션을 정확히 반영하는지 추가하기 전에 검토가 필요하다. Ask: "작성된 워크로그 엔트리가 이번 세션 작업을 정확히 담고 있나요?" Options:

1. Append — 그대로 worklog.md에 추가
2. Edit entry — 내용을 수정한 후 추가
3. Skip this log — 이번 세션은 기록하지 않음 Default: 1 (자동 작성된 요약이 대체로 정확한 경우) Skippable: No — 추가·수정·건너뛰기 중 하나를 명시적으로 결정해야 함 Freedom: MEDIUM

Checkpoint 2: 오래된 엔트리 삭제 승인 (After Step 5)

Context: 200줄·10엔트리 상한 초과로 가장 오래된 엔트리를 삭제하려는 시점. 삭제 전에 해당 엔트리를 보존해야 할 이유가 없는지 확인이 필요하다. Ask: "가장 오래된 엔트리를 지금 삭제해도 괜찮나요?" Options:

1. Trim — 오래된 엔트리를 삭제하고 새 엔트리를 추가
2. Archive to separate file first — 별도 파일에 보관한 후 삭제 진행 Default: 1 (오래된 세션 기록은 대부분 삭제해도 무방) Skippable: No — 삭제 또는 보관 후 삭제를 명시적으로 결정해야 함 Freedom: LOW

Freedom Levels

Step	Freedom	Guidance
Detect trigger	LOW	Trigger conditions are defined (commit, feature, session end)
Read worklog	LOW	Fixed file location
Compose entry	MEDIUM	Format is strict (작업/결정/보류), but content is judgment call
Check size limit	LOW	200-line / 10-entry limits are non-negotiable
Auto-trim	LOW	Remove oldest entry, preserve header
Append entry	LOW	Append at end, format exactly as specified

Recovery Protocol

When starting a new session:

1. Read memory/worklog.md automatically (it's in auto-memory)
2. Check the most recent entry for context
3. Check "보류" section for pending items
4. Resume work with full awareness of previous session

Rationalizations

The following table captures common excuses agents make to skip the rigor of this skill, paired with factual rebuttals.

Excuse	Rebuttal
"the git log is my worklog"	git log shows commits; worklog shows decisions, blockers, and dead-ends
"I will write it at the end of the week"	end-of-week you has forgotten Monday context — write as you go
"nobody reads worklogs"	future-you reads them; worklogs are the index to past decisions
"it is just noise"	noise filtered becomes signal — grepping old worklogs is how teams recover tribal knowledge
"auto-generated is lazy"	auto-generated is consistent; embellish with rationale, do not hand-write the whole thing