New Book Starter Kit (Modular Edition)

This is the standard starting point for all new book projects. It features a modular writing system that allows you to configure the AI writer (Claude) for specific genres (Fiction/Romantasy or Biography/Non-Fiction) while enforcing universal high-quality "human" writing standards.

Prerequisites

Before using this kit, ensure you have the following installed:

• Bun: JavaScript runtime and package manager.
• Vale: A syntax-aware linter for prose.
• Gemini CLI: The interactive CLI agent. Ensure it is available globally as gemini.

How to Use This Kit

1. Run init_book.sh from inside your new (empty) project directory:

   mkdir my_new_book && cd my_new_book
   bash /path/to/_New_Book_Starter_Kit/init_book.sh
   

   This creates the directory structure, symlinks shared modules, and copies template files.
2. Configure CLAUDE.md:
   • Open CLAUDE.md.
   • Select Track: The system dynamically loads Fiction or Nonfiction modules at session start. Delete the track you do not need to simplify the file.
   • Fill in the "Project Specifics" section at the bottom.
3. Initialize GEMINI.md:
   • Open GEMINI.md.
   • Fill in Project Identity, Thesis, and Scope.
4. Start Planning:
   • Use manuscript/00_master_outline.md to begin structuring your book.
5. To update from the kit later: Run bash update_book.sh from inside your project.

The Module System

The writing rules are stored in modules/:

• Universal:

  • _STYLE_AUTHORITY.md: Banned words, forbidden transitions, structural rules, crutch word limits.
  • _HUMAN_PATTERNS.md: Sentence rhythm (burstiness), paragraph variation.

• Genre-Specific:

  • _NARRATIVE_VOICE.md: POV rules (1st/3rd), all tenses, dialogue mechanics.
  • _ROMANCE_HEAT.md: Heat levels, fantasy integration.
  • _BIOGRAPHY.md: New Journalism style, factual rigor, avoiding hagiography.

Dual-AI Automation

This kit uses two AI agents with distinct roles:

| | Claude (Writer/Editor) | Gemini (Architect/QA) | |---|---|---| | Role | Writes and revises prose | Reviews, researches, verifies | | Skills | /scene-brief, /draft, /de-ai-audit, /revision-guide, /chapter-done | de-ai-audit, adversarial-review, kdp-format, continuity-audit | | Automation | 3 hooks (vocabulary scan, checklist reminder, context preservation) + 1 auto-loading rule | Plan Mode subagents, model routing |

Typical Workflow

1. /scene-brief — Claude gets a Research Brief from Gemini, generates a Scene Brief, user approves
2. /draft — Claude writes prose with automatic QC checks
3. /de-ai-audit — Claude scans for AI patterns (also runs automatically via hook)
4. /revision-guide — Claude requests 3 Gemini audits, builds a prioritized fix list, user approves
5. /chapter-done — Claude runs completion checklist, Gemini issues verification, user signs off

NotebookLM Integration

Run bash .claude/scripts/notebooklm-prep.sh to bundle your manuscript for Google NotebookLM with pre-written audit prompts for continuity, voice consistency, timeline, character arcs, and plot threads.

Directory Structure

• manuscript/: Your actual book content.
• research/: Structured research notes (facts, not prose).
• context/: Project bible details (characters, timeline, voice).
• modules/: The brain of the writing system.
• output/: Compiled files (PDF, EPUB, DOCX).
• .claude/: Claude Code automation (hooks, rules, skills, scripts).