bestseller-writer（畅销书作家）

Bestseller Writer 📚

One idea → Full manuscript → Amazon KDP listing. Fully autonomous.

Turn a shower thought into a publishable book. 5-stage multi-agent pipeline handles everything: story structure, characters, all chapters (written in parallel), editorial pass, and a complete Amazon KDP publishing package.

No writing experience needed. No blank page. Just your idea.

What You Get

Quick Start (CLI)

Install dependency

cd skills/bestseller-writer/scripts
npm install

Generate a book

# Thriller
node generate.js --idea "A detective discovers evidence of her own murder" --genre thriller

# Self-help  
node generate.js --idea "How to build a $10k/month business in 90 days using AI" --genre self-help

# Romance
node generate.js --idea "Two rival food truck owners forced to share a festival spot" --genre romance

# Business/non-fiction
node generate.js --idea "Why most startups fail at hiring and how to fix it" --genre business

Options

--idea, -i     Your book concept (required)
--genre, -g    Genre: thriller, romance, self-help, business, memoir, fantasy, nonfiction
--output, -o   Output directory (default: ./book-output/[slug])
--chapters     Number of chapters (default: 25)
--batch        Parallel chapter batch size (default: 4)
--planner      Model for Planner + Editor (default: claude-opus-4-5)
--writer       Model for writing agents (default: claude-sonnet-4-5)

Environment

export ANTHROPIC_API_KEY=sk-ant-...

Agent-Orchestrated Mode

If you're running inside an OpenClaw agent, the agent follows this pipeline directly using sessions_spawn. Trigger with:

"Write me a book about [idea]" or "Generate a bestseller about [idea]"

The agent will:

1. Create output directory at ~/Desktop/books/[slug]/
2. Run all 5 stages sequentially (chapters in parallel batches)
3. Deliver kdp_package.md as the final deliverable
4. Report word count and publish checklist

Pipeline Architecture

💡 Your Idea
     │
     ▼
┌─────────────────────────────────────┐
│  Stage 1: PLANNER (Opus)            │
│  Market positioning, title options, │
│  full chapter-by-chapter outline,   │
│  comparable titles, pitch           │
└──────────────────┬──────────────────┘
                   │
                   ▼
┌─────────────────────────────────────┐
│  Stage 2: CHARACTER DESIGNER        │
│  (Sonnet)                           │
│  Deep profiles, arcs, voice guide   │
│  (fiction) or authority/reader      │
│  avatar framework (non-fiction)     │
└──────────────────┬──────────────────┘
                   │
         ┌─────────┴─────────┐
         ▼                   ▼
  ┌─────────────┐     ┌─────────────┐
  │ Chapters    │ ... │ Chapters    │  ← Parallel batches
  │ 1-4 (Sonnet)│     │ 21-25       │    (4 at a time)
  └──────┬──────┘     └──────┬──────┘
         └─────────┬─────────┘
                   ▼
┌─────────────────────────────────────┐
│  Stage 4: EDITOR (Opus)             │
│  Consistency, pacing, voice,        │
│  opening/ending assessment,         │
│  marketability score                │
└──────────────────┬──────────────────┘
                   │
                   ▼
┌─────────────────────────────────────┐
│  Stage 5: KDP PACKAGER (Sonnet)     │
│  Title, description, keywords,      │
│  categories, pricing, cover brief,  │
│  Midjourney prompt, launch checklist│
└─────────────────────────────────────┘
                   │
                   ▼
        📦 MANUSCRIPT.md + kdp_package.md

Genre Support

Revenue Model

Amazon KDP passive income per book:

• Average: $50-500/month
• Stack 10 books: $500-5,000/month
• Stack 30 books: $1,500-15,000/month

Launch strategy (built into kdp_package.md):

1. Publish at $0.99 → enroll KDP Select
2. Day 1-4: Paid launch, collect reviews
3. Day 5-9: Free promo (spikes ranking)
4. Day 10+: Raise to $2.99-$4.99 (70% royalty)
5. Repeat with next book

Niche targeting (highest ROI):

• Cozy mysteries with recurring characters (series = repeat buyers)
• Self-help for specific professions (accountants, nurses, teachers)
• Business books for specific industries
• Local history/interest books (low competition)

Cost Estimate

One book costs ~$1.20 to generate and can earn $50-500/month. ROI is infinite.

Agent Instructions (for OpenClaw)

When the user asks to write a book or generate a manuscript:

Step 1 — Collect Info

Ask for (or infer from context):

• The idea (required)
• Genre (infer if obvious, otherwise ask)
• Fiction or non-fiction

Step 2 — Run Pipeline

Create output dir: ~/Desktop/books/[slug]/

Spawn agents in sequence using sessions_spawn with runtime="subagent":

Agent 1 — Planner: Task: Full planning prompt (see scripts/generate.js buildPlannerPrompt) Save output to: plan.md

Agent 2 — Character Designer: Task: Character/authority prompt with plan as context Save output to: characters.md

Agents 3-N — Chapter Writers (parallel, max 4 at once): Task: Chapter writing prompt with plan + characters + prev chapter ending Save output to: chapter_NN.md

Agent N+1 — Editor: Task: Editorial pass prompt with plan + sample chapters Save output to: editorial_memo.md

Agent N+2 — KDP Packager: Task: KDP package prompt with plan + editorial Save output to: kdp_package.md

Step 3 — Assemble

Concatenate all chapter_NN.md files into MANUSCRIPT.md

Step 4 — Deliver

Send user:

• Word count
• Path to MANUSCRIPT.md
• Key items from kdp_package.md (chosen title, pricing, first keyword string)
• Next steps for publishing

Troubleshooting

Chapters are too short? The writer agents are prompted for 2,000-2,500 words. If output is shorter, re-run that specific chapter with: "This chapter is too short. Expand to at least 2,000 words, adding more scene depth, dialogue, and sensory detail."

Voice is inconsistent? The editor stage catches this. After the editorial memo, re-run any flagged chapters with the specific feedback.

KDP keywords not relevant? Edit kdp_package.md keywords manually using Google Keyword Planner or Publisher Rocket to verify search volume before uploading.

Want a series? After Book 1 is done, pass plan.md + characters.md into a new run with --idea "Book 2: [continuation]". Characters and world are already built.