zenstory-ai/novel-to-game

What it solves

NovelToGame prevents the common issue where AI-generated games based on books are generic reskins or simple clickable plot summaries. It provides a structured, source-grounded workflow that ensures the resulting game is a faithful adaptation of the novel's rules, characters, and world, while maintaining a traceable design process.

How it works

The project is a toolkit of "Agent Skills" designed for AI agents (like Claude Code, Codex, and Kimi Code). It breaks the adaptation process into a staged pipeline:

1. Source Analysis: Extracts cited rules, character agency, and visual anchors from the novel.
2. Concept Selection: Compares and selects a game direction.
3. World Design: Defines the core loop, player verbs, and failure states.
4. Whitebox Testing: Runs a narrow check to test the hardest systemic or spatial questions before full production.
5. Art Direction: Defines the visual grammar, lighting, and sound.
6. Production Build: Implements the game for a specific target runtime/engine.
7. QA: Verifies the game's actual play paths and rendering in the chosen runtime.

Who it’s for

Developers and creators who want to use AI agents to transform novels into playable games or interactive stories without losing the narrative depth or systemic complexity of the source material.

Highlights

• Source-Grounded: Uses citations to ensure game mechanics are derived from the novel.
• Target-Runtime Delivery: Builds specifically for the chosen platform/engine to avoid silent redesigns.
• Agent-Native: Integrates directly as skills for Claude Code, Codex, and Kimi Code.
• Evidence-Based QA: Verifies actual startup and gameplay loops rather than relying on subjective reports.
• Flexible Modality: Supports both systems-heavy games and narrative-led interactive stories.