Dream Novel Writer

Problem

Long-form writing often fails because of context drift: the model forgets character goals, unresolved foreshadowing, and earlier constraints after enough chapters are written.

How It Works

This skill turns novel creation into a reproducible pipeline:
- Project setup: each book gets one directory, default ~/novels/; on resume it reads summary.json and chapter_*.md to determine the next chapter number.
- Architecture generation: uses ./resources/architecture.md to derive the core seed, character dynamics, worldbuilding, and plot architecture.
- Batched outlining: for long works, it generates chapter blueprints in batches, appends them to temp_outline.txt, and avoids prose until the full outline is complete.
- Serialized writing and memory updates: each chapter is drafted from an assembled prompt, reviewed against outline requirements, and finalized as chapter_X.md.
- State persistence: after each chapter, it extracts summaries, character changes, foreshadowing, and other structured JSON, then updates the project state for the next chapter.

Caveats

It is most useful for multi-chapter projects with complex characters and recurring plot threads; the workflow is heavier than needed for a single short story. Its quality depends on the model producing clean JSON for the state-update scripts.