Fiction Writing Skills System - Product Documentation

Document Index

Document	Description	Status
01_PRODUCT_ROADMAP.md	Three-phase roadmap with research foundation, features, success criteria, and feedback touchpoints	Complete
02_PHASE1_PRDs.md	Detailed Product Requirements Documents for all Phase 1 MVP features	Complete
03_IMPLEMENTATION_GUIDE.md	Technical implementation guide for Claude Code, including file structures, command definitions, and templates	Complete
05_AI_COAUTHOR_PLAN.md	Opinionated AI co-author architecture for SF-thriller / SF-philosophical writing with style modeling and diagnostics	Complete
06_COAUTHOR_EXECUTION_RUNBOOK.md	Operational runbook for canon setup and deterministic validation	Complete
07_COAUTHOR_SYSTEM_DESIGN.md	Detailed system design with architecture and end-to-end flow diagram	Complete

---

Executive Summary

This product creates a hybrid skill/agent system for Claude Code that helps fiction writers produce consistent, craft-informed scenes while maintaining creative control.

Core Problem Solved

AI-assisted scene drafting frequently loses:

• Character voice consistency
• Continuity details (knowledge, objects, timeline)
• World rule compliance (magic systems, geography)
• Authentic prose quality (avoids AI "tells")

Solution Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Story Bible (Foundation)                  │
│  story-bible.md | world-rules.md | scene-tracker.md         │
│  characters/*.md                                             │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                    Phase 1: MVP Pipeline                     │
│                                                              │
│  outline.md → [Scene Architect] → blueprint.md               │
│                      ↓ pause for review                      │
│  blueprint.md → [Scene Draft] → draft.md                     │
│                      ↓ pause for review                      │
│  draft.md → [Character Truth] → audit.md                     │
│                      ↓                                       │
│                 [Human Polish]                               │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                    Phase 2: Agents (Future)                  │
│                                                              │
│  Scene Director (diagnoses, orchestrates)                    │
│       ├── Prose Surgeon (line-level craft)                   │
│       └── Continuity Keeper (cross-references)               │
│                                                              │
│  Additional Skills:                                          │
│  - Dialogue Subtext Pass                                     │
│  - Tension Curve Analysis                                    │
│  - AI Humanization Pass                                      │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                    Phase 3: Scale (Future)                   │
│                                                              │
│  - Multi-scene arc tracking                                  │
│  - Voice fingerprinting                                      │
│  - Genre-specific modules                                    │
│  - Collaboration features                                    │
└─────────────────────────────────────────────────────────────┘

---

Research Foundation

The system is grounded in proven craft methodologies:

Story Grid (Shawn Coyne)

• Five Commandments: Every scene has Inciting Incident, Progressive Complication, Crisis, Climax, Resolution
• Value Shift: Every scene must change something
• Scene Purpose: Reveal character, setting, context, or advance plot

Stephen King's "On Writing"

• Write first draft fast, revise ruthlessly
• Show don't tell (avoid adverbs, use active voice)
• Natural dialogue (your vocabulary, not thesaurus words)
• Kill your darlings (cut 10%+)

Character Voice Craft

• Voice = how character interprets the world (worldview filter)
• Consistency builds trust
• Subtext matters (what's NOT said)
• Differentiation: syntax, rhythm, vocabulary, tics

Anti-AI Prose Patterns

System actively counters:

• Over-consistency (uniform sentences)
• Regression to mean (generic descriptions)
• Overused vocabulary ("delve," "multifaceted," etc.)
• Lack of sensory variety
• Missing concrete details

---

Phase 1 MVP Features

1.1 Story Bible Schema

Structured templates for tracking story facts:

• story-bible.md - premise, theme, structure
• characters/*.md - profiles with voice guides
• world-rules.md - constraints and hard limits
• scene-tracker.md - continuity tracking

1.2 Scene Architect Skill

Transform rough idea → structured blueprint following Five Commandments

1.3 Scene Draft Skill

Transform blueprint → full prose with voice enforcement and world rule checking

1.4 Character Truth Pass

Audit draft for character consistency (voice, knowledge, behavior)

1.5 Pipeline Runner

Orchestrate skills with pause-for-review between stages

---

Success Criteria (MVP)

1. Writer produces complete scene from outline in under 2 hours
2. Character voice remains consistent across scenes
3. No world rule violations in generated content
4. Prose passes "gut check" for human quality
5. Writer maintains creative control at each stage

---

Customer Feedback Integration

Post-Phase 1 Interview (After MVP)

• What's working/not working?
• Are templates capturing what you need?
• Voice preservation accuracy?
• What's missing?

Post-Phase 2 Interview (After Agents)

• Scene Director diagnosis accuracy?
• Prose Surgeon helpfulness?
• Continuity error catch rate?
• What's the new bottleneck?

---

Getting Started

Quick Start: Stage 0 (Story Development)

1. Validate setup: python scripts/validate_coauthor_setup.py --root .
2. Run /project:skills:story-concept → produces canon/story-concept.md (L1)
3. Run /project:skills:story-arc-builder → produces canon/story-arc.md (L2)
4. Run /project:skills:act-outline (per act) → produces canon/acts/act-{N}-outline.md (L3)
5. Pipeline state advances automatically through L1→L2→L3 in .pipeline-state.yaml
6. Continue to Stage 1 (chapter planning) with /project:pipeline-run

Mob Review (Mode C)

At any pipeline step, run a mob review where specialist agents critique the current artifact:

/project:skills:mob-session

The Lead Editor orchestrates specialist agents (Plot Analyst, Character Specialist, Depth Partner, Continuity Agent, Prose Crafter) in a structured 4-phase protocol:

1. Structure Offer — organizes your input
2. Comment Queue — each agent gives one focused comment with canon citations
3. Resolution Check — summarizes accepted changes, checks governance rules
4. Commit — writes approved changes to canon files

You maintain full control: accept, reject, revise, or park each suggestion. See docs/mob_protocol.md for details.

Background Reading

1. Read 01_PRODUCT_ROADMAP.md for full context
2. Review 02_PHASE1_PRDs.md for detailed requirements
3. Use 03_IMPLEMENTATION_GUIDE.md to build the system
4. Review 05_AI_COAUTHOR_PLAN.md for the opinionated co-author architecture
5. Run 06_COAUTHOR_EXECUTION_RUNBOOK.md and validate setup with python scripts/validate_coauthor_setup.py --root .

---

Version

• Version: 0.1.0
• Date: January 8, 2026
• Status: Phase 1 documented, ready for implementation