Webnovel Writer is a Claude Code plugin that tackles one of the hardest problems in AI-assisted creative writing: maintaining consistency across extremely long narratives. With support for serialized works spanning over 2 million characters, this open-source system uses RAG (Retrieval-Augmented Generation) with reranking to prevent the "forgetting" and "hallucination" problems that plague AI writing tools when working on long-form content. The project has gained 999 stars with 292 new stars in a single day, signaling strong interest from the AI writing community. ## The Problem It Solves AI writing assistants excel at short-form content but struggle with novels and serialized fiction. After a few chapters, the AI loses track of character details, plot threads, world-building rules, and narrative promises made to readers. Webnovel Writer addresses this directly with a structured memory and retrieval system. ## Key Features ### RAG-Enhanced Context Management The core innovation is a retrieval-augmented generation pipeline with reranking that maintains a knowledge base of the entire work. When writing a new chapter, the system retrieves relevant context from previous chapters, character profiles, and world-building documents, ensuring consistency even hundreds of chapters into a series. The RAG subsystem uses external embedding and reranking APIs (ModelScope, Jina) to build and query a vector store of the entire narrative history. ### Multi-Agent Architecture Webnovel Writer employs specialized agents for different phases of the writing process: | Agent | Role | |-------|------| | Planner | Outlines plot structure, chapter arcs, and pacing | | Writer | Generates narrative prose based on plans and context | | Reviewer | Checks for consistency, quality, and adherence to established rules | Each agent can inherit the active Claude session model or be configured independently with Sonnet, Opus, or Haiku. ### Entity Tracking System A dedicated "pursuit system" monitors narrative elements across the entire work: - **Story hooks**: Promises and mysteries set up for readers - **Cooldown points**: Tracking pacing to avoid reader fatigue - **Minor resolutions**: Ensuring subplots get addressed - **Debt tracking**: Flagging unresolved narrative threads ### Visual Dashboard A read-only monitoring panel provides real-time visibility into: - Project status and progress metrics - Entity knowledge graphs showing character and plot relationships - Chapter and outline browsing - Reader engagement metrics ## Workflow The system operates through four primary commands within Claude Code: ```bash /webnovel-init # Initialize a new novel project /webnovel-plan # Plan chapters and story arcs /webnovel-write # Generate narrative content /webnovel-review # Review and validate consistency /webnovel-dashboard # Launch monitoring dashboard ``` Each novel project gets its own directory with centralized configuration, making it straightforward to manage multiple works simultaneously. ## Technical Requirements - Python 3.10+ - Claude Code (as the host environment) - External RAG APIs for embedding and reranking - Optional: Gemini CLI and Codex for extended capabilities ## Installation The plugin installs directly through the Claude Code marketplace: ```bash claude plugin marketplace add lingfengQAQ/webnovel-writer --scope user python -m pip install -r requirements.txt ``` ## Why This Matters The web fiction market is enormous, particularly in East Asia where serialized novels spanning thousands of chapters are common. Webnovel Writer represents one of the first serious attempts to build AI tooling specifically for this format. By combining RAG for long-term memory, multi-agent workflows for quality control, and entity tracking for narrative consistency, it offers a structured approach to a problem that simple prompt engineering cannot solve. ## Conclusion Webnovel Writer demonstrates the potential of Claude Code's plugin ecosystem for domain-specific creative applications. Its RAG-backed memory system and multi-agent architecture address the fundamental limitations of AI in long-form writing. For fiction authors working on serialized content, it provides a structured framework that keeps AI assistance useful far beyond the context window limitations of any single model.