GhostEditor

AI-powered developmental editing for self-published authors. Upload a manuscript — fiction or nonfiction — and get structured, chapter-by-chapter feedback on consistency, argument quality, pacing, and genre or format conventions.

How It Works

Fiction

1. Upload a manuscript (DOCX, PDF, or TXT) and select your genre
2. Story Bible is generated automatically from Chapter 1 — characters, timeline, settings, world rules, voice profile, plot threads
3. Chapter Analysis runs sequentially, updating the bible and flagging issues: continuity errors, pacing problems, genre convention violations
4. Feedback Dashboard shows results organized by chapter and severity (Critical / Warning / Note)

Nonfiction

1. Upload a manuscript and select a nonfiction format (Academic, Personal Essay, Journalism, Self-Help, or Business)
2. Argument Map is generated — central thesis, argument threads, evidence log, voice profile, structural markers
3. Section Analysis evaluates each section across five dimensions: argument structure, evidence quality, clarity, organization, and tone — with format-specific conventions applied
4. Document Synthesis produces an overall assessment with scores for thesis clarity, argument coherence, evidence density, and tone consistency, plus prioritized recommendations
5. Feedback Dashboard shows per-section results with dimension filtering and a document summary panel

Architecture

React (Vite)  →  FastAPI  →  Claude API (Sonnet + Haiku)
                    ↕
              PostgreSQL + Redis (arq) + S3/MinIO

• Story Bible / Argument Map generation: Claude — builds a structured JSON representation of the manuscript's core elements
• Chapter / Section analysis: Claude — flags issues against the bible or argument map with severity, location, and suggestions
• Document synthesis (nonfiction): Claude — produces a document-level assessment from structured analysis data
• Job queue: arq (Redis) — sequential processing with retry, stall recovery, and stuck-manuscript auto-recovery
• Storage: PostgreSQL (manuscripts, bibles, argument maps, analysis results), S3/MinIO (original files)

Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+
• Docker

Setup

# Start infrastructure (uses project-isolated ports)
docker-compose -f infra/docker-compose.yml up -d

# Backend
cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp ../.env.example ../.env  # then edit with your keys
PYTHONPATH=. alembic upgrade head
PYTHONPATH=. uvicorn app.main:app --reload

# Worker (separate terminal)
cd backend
PYTHONPATH=. .venv/bin/arq app.jobs.worker.WorkerSettings

# Frontend (separate terminal)
cd frontend
npm install
npm run dev

Port Assignments

Service	Port
PostgreSQL	5433
Redis	6380
MinIO API	9002
MinIO Console	9003
Backend API	8000
Frontend	5173

Project Structure

├── backend/
│   ├── app/
│   │   ├── analysis/        # Claude API, prompts, JSON repair, export
│   │   │   ├── prompts/     # Versioned prompt files (bible, analysis, splitting, nonfiction)
│   │   │   ├── story_bible.py / argument_map.py   # Fiction / nonfiction generation
│   │   │   ├── chapter_analyzer.py / nonfiction_analyzer.py
│   │   │   └── nonfiction_synthesis.py
│   │   ├── auth/            # JWT auth, registration, password reset
│   │   ├── manuscripts/     # Upload, extraction, chapter/section detection
│   │   ├── jobs/            # arq worker, stall recovery, GDPR purge cron
│   │   ├── email/           # Resend integration, drip sequences
│   │   ├── stripe/          # Checkout, webhooks, subscriptions
│   │   └── db/              # SQLAlchemy models, Alembic migrations
│   └── tests/
│       ├── eval/            # Prompt eval harness (5 fiction + 5 nonfiction samples)
│       └── unit/            # 138+ unit tests
├── frontend/src/
│   ├── pages/               # Dashboard, manuscript, feedback, bible, argument map views
│   ├── components/          # Shared UI components
│   ├── api/                 # Typed API client
│   └── __tests__/           # Vitest smoke tests
├── infra/
│   └── docker-compose.yml
└── docs/
    ├── blueprint.md         # Full MVP specification
    ├── decisions/           # Adversarial protocol decision records (DECISION-001 through 011)
    ├── review/              # Codebase audit report
    └── build_log.md         # Development progress

Development Process

All significant design decisions go through a three-agent adversarial review (Architect / Adversary / Judge) documented in docs/decisions/. See CLAUDE.md for the full protocol.

Tests

# Backend unit tests (138+)
cd backend
PYTHONPATH=. .venv/bin/pytest tests/unit/ -v

# Backend eval harness (requires ANTHROPIC_API_KEY — costs real API money)
PYTHONPATH=. .venv/bin/pytest tests/eval/ -v -m api

# Frontend smoke tests
cd frontend
npm test

License

Proprietary. All rights reserved.