Writer Memory

Writer Memory - Agentic Memory System for Writers

Persistent memory system designed for creative writers, with first-class support for Korean storytelling workflows.

Overview

Writer Memory maintains context across Antigravity sessions for fiction writers. It tracks:

• Characters (캐릭터): Emotional arcs (감정궤도), attitudes (태도), dialogue tone (대사톤), speech levels
• World (세계관): Settings, rules, atmosphere, constraints
• Relationships (관계): Character dynamics and evolution over time
• Scenes (장면): Cut composition (컷구성), narration tone, emotional tags
• Themes (테마): Emotional themes (정서테마), authorial intent

All data persists in .writer-memory/memory.json for git-friendly collaboration.

Commands

Memory Types

캐릭터 메모리 (Character Memory)

Tracks individual character attributes essential for consistent portrayal:

Example:

/writer-memory char add 새랑
/writer-memory char update 새랑 arc "체념 -> 욕망자각 -> 선택"
/writer-memory char update 새랑 tone "담백, 현재충실, 감정억제"
/writer-memory char update 새랑 speechLevel "해체"
/writer-memory char update 새랑 keywords "그냥, 뭐, 괜찮아"
/writer-memory char update 새랑 taboo "사랑해, 보고싶어"

세계관 메모리 (World Memory)

Establishes the universe your story inhabits:

관계 메모리 (Relationship Memory)

Captures the dynamic between characters over time:

Example:

/writer-memory rel add 새랑 해랑 romantic
/writer-memory rel update 새랑 해랑 "첫 키스 - 새랑 회피"
/writer-memory rel update 새랑 해랑 "해랑 고백 거절당함"
/writer-memory rel update 새랑 해랑 "새랑 먼저 손 잡음"

장면 메모리 (Scene Memory)

Tracks individual scenes and their emotional architecture:

테마 메모리 (Theme Memory)

Captures the deeper meaning woven through your story:

Synopsis Generation (시놉시스)

The /synopsis command generates an emotion-focused summary using 5 essential elements:

5 Essential Elements (시놉시스 5요소)

1. 주인공 태도 요약 (Protagonist Attitude Summary)

   • How the protagonist approaches life/love/conflict
   • Their core emotional stance
   • Example: "새랑은 상실을 예방하기 위해 먼저 포기하는 사람"

2. 관계 핵심 구도 (Core Relationship Structure)

   • The central dynamic driving the story
   • Power imbalances and tensions
   • Example: "사랑받는 자와 사랑하는 자의 불균형"

3. 정서적 테마 (Emotional Theme)

   • The feeling the story evokes
   • Not plot, but emotional truth
   • Example: "손에 쥔 행복을 믿지 못하는 불안"

4. 장르 vs 실제감정 대비 (Genre vs Real Emotion Contrast)

   • Surface genre expectations vs. actual emotional content
   • Example: "로맨스지만 본질은 자기수용 서사"

5. 엔딩 정서 잔상 (Ending Emotional Aftertaste)

   • The lingering feeling after the story ends
   • Example: "씁쓸한 안도, 불완전한 해피엔딩의 여운"

Character Validation (캐릭터 검증)

The /validate command checks if dialogue matches a character's established voice.

What Gets Checked

Validation Results

• PASS: Dialogue is consistent with character
• WARN: Minor inconsistencies, may be intentional
• FAIL: Significant deviation from established voice

Example:

/writer-memory validate 새랑 "사랑해, 해랑아. 너무 보고싶었어."

Output:

[FAIL] 새랑 validation failed:
- TABOO: "사랑해" - character avoids direct declarations
- TABOO: "보고싶었어" - character suppresses longing expressions
- TONE: Too emotionally direct for 새랑's 담백 style

Suggested alternatives:
- "...왔네." (minimal acknowledgment)
- "늦었다." (deflection to external fact)
- "밥 먹었어?" (care expressed through practical concern)

Context Query (맥락 질의)

Natural language queries against memory, with full Korean support.

Example Queries

/writer-memory query "새랑은 이 상황에서 뭐라고 할까?"
/writer-memory query "규리의 현재 감정 상태는?"
/writer-memory query "해랑과 새랑의 관계는 어디까지 왔나?"
/writer-memory query "이 장면의 정서적 분위기는?"
/writer-memory query "새랑이 먼저 연락하는 게 맞아?"
/writer-memory query "해랑이 화났을 때 말투는?"

The system synthesizes answers from all relevant memory types.

Behavior

1. On Init: Creates .writer-memory/memory.json with project metadata and empty collections
2. Auto-Backup: Changes are backed up before modification to .writer-memory/backups/
3. Korean-First: Emotion vocabulary uses Korean terms throughout
4. Session Loading: Memory is loaded on session start for immediate context
5. Git-Friendly: JSON formatted for clean diffs and collaboration

Integration

With OMC Notepad System

Writer Memory integrates with .omc/notepad.md:

• Scene ideas can be captured as notes
• Character insights from analysis sessions are preserved
• Cross-reference between notepad and memory

With Architect Agent

For complex character analysis:

Task(subagent_type="oh-my-claudecode:architect",
     model="opus",
     prompt="Analyze 새랑's arc across all scenes...")

Character Validation Pipeline

Validation pulls context from:

• Character memory (tone, keywords, taboo)
• Relationship memory (dynamics with dialogue partner)
• Scene memory (current emotional context)
• Theme memory (authorial intent)

Synopsis Builder

Synopsis generation aggregates:

• All character arcs
• Key relationship events
• Scene emotional tags
• Theme expressions

Examples

Full Workflow

# Initialize project
/writer-memory init 봄의 끝자락

# Add characters
/writer-memory char add 새랑
/writer-memory char update 새랑 arc "체념 -> 욕망자각 -> 선택"
/writer-memory char update 새랑 tone "담백, 현재충실"
/writer-memory char update 새랑 speechLevel "해체"

/writer-memory char add 해랑
/writer-memory char update 해랑 arc "확신 -> 동요 -> 기다림"
/writer-memory char update 해랑 tone "직진, 솔직"
/writer-memory char update 해랑 speechLevel "반말"

# Establish relationship
/writer-memory rel add 새랑 해랑 romantic
/writer-memory rel update 새랑 해랑 "첫 만남 - 해랑 일방적 호감"
/writer-memory rel update 새랑 해랑 "새랑 거절"
/writer-memory rel update 새랑 해랑 "재회 - 새랑 내적 동요"

# Set world
/writer-memory world set setting "서울, 현대, 20대 후반 직장인"
/writer-memory world set atmosphere "도시의 건조함 속 미묘한 온기"

# Add themes
/writer-memory theme add "포기하지 않는 사랑"
/writer-memory theme add "자기 보호의 벽"

# Add scene
/writer-memory scene add "옥상 재회"

# Query for writing
/writer-memory query "새랑은 이별 장면에서 어떤 톤으로 말할까?"

# Validate dialogue
/writer-memory validate 새랑 "해랑아, 그만하자."

# Generate synopsis
/writer-memory synopsis

# Export for reference
/writer-memory export

Quick Character Check

/writer-memory char 새랑

Output:

## 새랑

**Arc (감정궤도):** 체념 -> 욕망자각 -> 선택
**Attitude (태도):** 방어적, 현실주의
**Tone (대사톤):** 담백, 현재충실
**Speech Level (말투):** 해체
**Keywords (핵심어):** 그냥, 뭐, 괜찮아
**Taboo (금기어):** 사랑해, 보고싶어

**Relationships:**
- 해랑: romantic (intimacy: 6/10, status: healing)

**Scenes Appeared:** 옥상 재회, 카페 대화, 마지막 선택

Storage Schema

{
  "version": "1.0",
  "project": {
    "name": "봄의 끝자락",
    "genre": "로맨스",
    "created": "2024-01-15T09:00:00Z",
    "lastModified": "2024-01-20T14:30:00Z"
  },
  "characters": {
    "새랑": {
      "arc": "체념 -> 욕망자각 -> 선택",
      "attitude": "방어적, 현실주의",
      "tone": "담백, 현재충실",
      "speechLevel": "해체",
      "keywords": ["그냥", "뭐", "괜찮아"],
      "taboo": ["사랑해", "보고싶어"],
      "emotional_baseline": "평온한 무관심",
      "triggers": ["과거 언급", "미래 약속"]
    }
  },
  "world": {
    "setting": "서울, 현대, 20대 후반 직장인",
    "rules": [],
    "atmosphere": "도시의 건조함 속 미묘한 온기",
    "constraints": [],
    "history": ""
  },
  "relationships": [
    {
      "id": "rel_001",
      "from": "새랑",
      "to": "해랑",
      "type": "romantic",
      "dynamic": "해랑 주도 → 균형",
      "speechLevel": "반말",
      "evolution": [
        { "timestamp": "...", "change": "첫 만남 - 해랑 일방적 호감", "catalyst": "우연한 만남" },
        { "timestamp": "...", "change": "새랑 거절", "catalyst": "과거 트라우마" },
        { "timestamp": "...", "change": "재회 - 새랑 내적 동요", "catalyst": "옥상에서 재회" }
      ],
      "notes": "새랑의 불신 vs 해랑의 기다림",
      "created": "..."
    }
  ],
  "scenes": [
    {
      "id": "scene-001",
      "title": "옥상 재회",
      "characters": ["새랑", "해랑"],
      "location": "회사 옥상",
      "cuts": ["해랑 먼저 발견", "새랑 굳은 표정", "침묵", "해랑 먼저 말 걸기"],
      "narration_tone": "건조체",
      "emotional_tag": "긴장+그리움",
      "purpose": "재회의 어색함과 남은 감정 암시",
      "before_after": "새랑: 무관심 -> 동요"
    }
  ],
  "themes": [
    {
      "name": "포기하지 않는 사랑",
      "expression": "해랑의 일관된 태도",
      "scenes": ["옥상 재회", "마지막 고백"],
      "character_links": ["해랑"],
      "author_intent": "집착이 아닌 믿음의 사랑"
    }
  ],
  "synopsis": {
    "protagonist_attitude": "새랑은 상실을 예방하기 위해 먼저 포기하는 사람",
    "relationship_structure": "기다리는 자와 도망치는 자의 줄다리기",
    "emotional_theme": "사랑받을 자격에 대한 의심",
    "genre_contrast": "로맨스지만 본질은 자기수용 서사",
    "ending_aftertaste": "불완전하지만 따뜻한 선택의 여운"
  }
}

File Structure

.writer-memory/
├── memory.json          # Main memory file
├── backups/             # Auto-backups before changes
│   ├── memory-2024-01-15-090000.json
│   └── memory-2024-01-20-143000.json
└── exports/             # Markdown exports
    └── export-2024-01-20.md

Tips for Writers

1. Start with Characters: Build character memories before scenes
2. Update Relationships After Key Scenes: Track evolution actively
3. Use Validation While Writing: Catch voice inconsistencies early
4. Query Before Difficult Scenes: Let the system remind you of context
5. Regular Synopsis: Generate periodically to check thematic coherence
6. Backup Before Major Changes: Use /backup before significant story pivots

Troubleshooting

Memory not loading?

• Check .writer-memory/memory.json exists
• Verify JSON syntax is valid
• Run /writer-memory status to diagnose

Validation too strict?

• Review taboo list for unintended entries
• Consider if character is growing (arc progression)
• Intentional breaks from pattern are valid for dramatic moments

Query not finding context?

• Ensure relevant data is in memory
• Try more specific queries
• Check character names match exactly