story-sandbox 0.1.0

Autonomous story generation sandbox for Claude Code. Characters develop on their own, world self-heals, output to Obsidian vault.

story-sandbox

English | 中文

Autonomous story generation sandbox for Claude. Set up a world and characters, then watch them develop on their own. Stories are unpredictable — even you don't know what happens next.

Output goes directly to your Obsidian vault with full visualization: relationship graphs, timelines, Dataview dashboards, and Canvas story maps.

---

What it does

You set:   World rules + Characters + Target length
                    ↓
Sandbox:   Characters act autonomously (limited perspective = emergence)
           World self-heals (plot holes get explained, not deleted)
           New locations/characters appear naturally
                    ↓
Output:    Obsidian vault with scenes, chapters, relationship graphs

Install

pip install story-sandbox

Then add to your Claude config:

{
  "mcpServers": {
    "story-sandbox": {
      "command": "story-sandbox"
    }
  }
}

Or use the Claude Code CLI:

claude mcp add story-sandbox -- story-sandbox

Quick start

Open Claude Code and say:

Start a story sandbox

Claude will guide you through:

1. Setting up the world (4 questions: era, rule, conflict, mood)
2. Creating characters (name, gender, personality, background, motive, secret)
3. Choosing target length (short / medium / long)
4. Choosing autonomy mode (full / semi / constrained)
5. Running the simulation (characters act on their own)

Tools

Tool	Description
sandbox_init	Initialize vault with directory structure and world docs
sandbox_add_character	Add a character with personality, background, secret
sandbox_get_state	Read current sandbox state (characters, foreshadowing, scenes)
sandbox_write_scene	Write a scene file and update character states
sandbox_compile_chapter	Compile scenes into a novel chapter
sandbox_update_graph	Update Obsidian Canvas with character-location relationships
sandbox_check_consistency	Scan for naming conflicts, timeline issues, relationship drift
sandbox_export	Export materials for story-long-write integration

How it works

Emergence through limited perspective

Each character only knows what they should know — not other characters' inner thoughts, secrets, or unseen events. This creates natural misunderstandings, misjudgments, and unexpected conflicts.

Self-healing world

When a character encounters an undefined location or concept, the system automatically creates it. When inconsistencies are detected, they're explained rather than deleted ("the memory was tampered with").

Convergence curve

Phase	Progress	Behavior
Free growth	0-30%	High foreshadowing density. New characters/locations frequent.
Recovery	30-70%	Foreshadowing recovery priority rises. No new S-level foreshadowing.
Focus	70-90%	Foreshadowing only decreases. Characters focus on core conflict.
Endgame	90-100%	All S-level foreshadowing must resolve. Final confrontations.

Three foreshadowing levels

Level	Lifetime	Purpose
S (story)	Entire book	Core mysteries that span the whole story
A (arc)	10-20 rounds	Mid-level mysteries resolved per arc
B (scene)	3-5 rounds	Small hooks for immediate engagement

Obsidian output

your-vault/
├── 00-世界观/          World docs (auto-expanded)
├── 01-角色/            Character sheets (updated each round)
├── 02-场景/
│   ├── 01-第一章/      Scenes organized by chapter
│   └── ...
├── 03-时间线/          Timeline index
├── 04-关系图/          Relationship cards (one per character)
├── 05-状态面板/        Dataview dashboard
├── 06-画布/            Canvas story map
├── 07-素材导出/
│   └── 小说正文/       Novel chapters (01-xxx.md, 02-xxx.md)
└── sandbox-state.json

Graph View colors

Color	Category
Green	Characters
Orange	Locations
Purple	World rules

Scenes are excluded from Graph View to prevent clutter.

Example

A story about a near-future city where everyone has abilities but using them shortens lifespan:

World:  Near-future city, abilities shorten lifespan,
        Bureau vs Liberation Organization, dark mood
Characters:
  - Lin Mo (male): Ex-agent, bookstore owner, looking for missing sister
  - Su Qing (female): Investigative journalist, ability: read memories by touch
  - Chen Wei (male): Entrepreneur, secretly behind the sister's disappearance

After 50 rounds of autonomous development:

• 50 scenes across 10 chapters
• 7 characters (3 original + 4 emerged)
• 13 locations across 2 cities
• ~40,000 Chinese characters of novel text
• Complete relationship graph and timeline

Requirements

• Python 3.10+
• Claude Code or Claude Desktop
• Obsidian (with Dataview plugin for dashboard)

License

MIT