webnovel-writer is a consistency system for long-form serialized fiction that turns writing constraints into automated steps before and after each chapter, ensuring settings, plot threads, and rules are maintained across millions of words. It addresses the problem of AI-generated web novels forgetting established settings and breaking consistency as stories reach hundreds of chapters.
Project overview
The project tackles a well-documented failure mode of AI-generated long fiction—setting drift and broken plot continuity—by introducing a strict Story System pipeline with automated fact extraction, state projection, and multi-dimensional quality review for each chapter.
Project type
AI Agent · RAG · Prompt Engineering
Use cases
Knowledge Q&A
Deployment
Refer to project documentation
License
GPL-3.0
Best for
Creators who write long-form serialized fiction and need automated consistency management across hundreds of chapters
Writers who work within the Claude Code environment and are comfortable configuring Python dependencies and optional RAG APIs
Key capabilities
Defines contracts and commits for chapters, serving as the single source of truth for facts before writing and after committing.
Prepares context, drafts text, runs multi-dimensional reviews, polishes, extracts facts, and drives state/index/memory projections for a completed chapter.
Automatically synchronizes newly committed facts into queryable state, indices, chapter summaries, long-term memory, and vectors.
Reviews written chapters across consistency, pacing, out-of-character detection, and engagement metrics.
A pre-packaged dashboard to browse project status, entity graphs, chapter contents, and engagement data without running npm build.
Manages vector indices and retrieval status, falling back to BM25 keyword search if embedding API keys are not provided.
Includes 37 built-in Chinese web novel genre templates, with support for combining multiple genres.
Checks directories, files, databases, RAG, dependencies, and dashboard artifacts to provide stage-aware status and repair suggestions.
Limitations and risks
The project relies on the Claude Code environment to run skill commands and execute its automated pipeline; it cannot operate independently of this execution environment.
Achieving optimal semantic recall requires external paid API keys for embeddings and reranking; without them the system falls back to BM25 keyword search.
Getting started
Install the plugin via the Claude Code Marketplace, install Python dependencies, initialize a book project via /webnovel-init, and configure RAG by copying .env.example to .env. Setup difficulty is rated medium due to the Claude Code environment requirement, Python dependency installation, and optional external RAG API configuration for optimal semantic recall.