AI长篇小说创作系统 - 基于文件级长期记忆的Smart State模式，支持百万字级长篇小说持续创作

This project addresses the challenges of long-form AI novel creation, specifically issues like plot drift, character inconsistency, and an unnatural "AI flavor" over millions of characters. It provides a comprehensive system for continuous novel writing, targeting engineers, researchers, and power users seeking robust AI-assisted content generation. The primary benefit is the guarantee of consistency and quality across extensive narratives, transforming AI from a drafting tool into a collaborative writing partner.

How It Works

The system employs a sophisticated five-layer collaborative mechanism to ensure long-form consistency: quality gates for chapter closure, RAG for plot retrieval, a knowledge graph for entity management, outline anchors for progress constraints, and cross-agent review for validation. Its "anti-AI flavor" solution uses a two-pass "humanizer" process to identify and correct common AI writing patterns by replacing abstract language with concrete details and actions. Architectural choices include "Iron Law" constraints, a reverse brake mechanism to prevent premature conflict resolution, an event matrix with cooldowns for pacing, and a multi-step pipeline for chapter generation.

Quick Start & Requirements

Installation involves running scripts/install-portable-skill.sh with the desired AI tool (e.g., claude-code, codex). Key commands include /一键开书 to initialize a project with a title, genre, and plot seed, /继续写 to continue chapter generation, and /修复本章 for recovering from gate failures. LLM configuration is managed via .novel_writer_config.yaml, supporting providers like OpenAI, Anthropic, Kimi, and GLM. Official user guides and command playbooks are available in the references/ directory.

Highlighted Details

• 3 Million Character Consistency: Achieved through a robust five-layer collaborative mechanism, ensuring narrative coherence over vast lengths.
• "Anti-AI Flavor" Solution: A detailed two-pass "humanizer" process targets 7 categories of AI writing patterns, focusing on replacing abstract phrases with concrete details and actions.
• Editorial Team Collaboration: Simulates a multi-person editorial workflow (Planning Editor, Writing Agent, Anti-AI Editor, etc.) to fundamentally eliminate AI hallucinations and inconsistencies.
• Iron Law Constraints: A set of strict, non-negotiable rules govern the writing process, preventing common AI pitfalls.
• Reverse Brake Mechanism & Event Matrix: Prevents premature resolution of core conflicts, ensures narrative hooks, and manages plot pacing by preventing repetition of event types.

Maintenance & Community

The provided README does not contain specific details regarding maintainers, community channels (like Discord or Slack), or sponsorship information.

Licensing & Compatibility

The project is released under the MIT License, which is highly permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

The system's complexity, with multiple interacting components and a detailed gate-checking process, may require a steep learning curve for debugging. Reliance on external AI tool integrations means setup is dependent on the successful installation and configuration of those tools. The "low context strategy" might limit the immediate narrative depth considered during writing, relying heavily on the effectiveness of the memory and retrieval systems.