The most advanced writing compiler ever open-sourced. Now with a fiction intelligence engine.
58 files. 68,770 words. A 7-pass compilation pipeline that produces prose with authorial identity, argument force, epistemic discipline, and zero AI residue. Plus a complete fiction architecture system — chapter construction, 12 character role archetypes, dialogue warfare, power dynamics through staging, tension mechanics, thriller scene architecture, and transmedia character deepening.
Free. For everyone. Forever.
Created by Antonio T. Smith Jr. — Founder & CEO of Density6 LLC
v1.0 was the most comprehensive writing compiler available anywhere. 52 files. 14 genres. 6-pass pipeline. It compiled prose that no detector could flag and no editor would reject.
v2.0 adds a complete Fiction Intelligence Engine — the system that was missing from every AI writing tool on the planet. Not "creative writing tips." A structural engineering system for fiction at every level: sentence, scene, chapter, arc, and storyworld.
| What's New | What It Does |
|---|---|
| 7th compilation pass | Scene & Chapter Audit — structural analysis of narrative prose |
| Chapter Construction Doctrine | 11-section framework for building masterpiece-grade chapters |
| 12 Character Role Archetypes | Structural roles with deployment rules, stacking limits, diagnostics |
| Dialogue v2.0 | 7 laws + 9 tension elements + 15 advanced techniques + 100-point scoring |
| Power Dynamics | Props, space, gesture, consumption, environment as narrative weapons |
| Tension Mechanics | 8-phase slow-burn compression model + silence taxonomy + psychological warfare |
| Thriller Scene Architecture | Confined-space doctrine, 5 structural beats, WWGW framework |
| Transmedia Character Deepening | Diaries, audio dramas, lore bibles, interactive serials, AR experiences |
| 7 scoring systems | Independent 100-point rubrics for prose, chapters, dialogue, power, tension, thriller, transmedia |
| 5 new rewrite operators | scene_audit, role_audit, dialogue_stress_test, power_map, plant_audit |
| 6 new adversarial tests | Dialogue distinction, tension compression, power staging, foreshadowing, roles, full battery |
| 3 new gold outputs | Dialogue scene (94), thriller chapter (92), transmedia diary (90) |
Every AI writing tool on the market does the same thing: removes AI tells. They strip phrases, flag structures, and clean surfaces. The result is prose that doesn't sound like AI — but doesn't sound like anyone, either.
Clean generic prose is still generic prose.
For fiction, the problem is worse. AI-generated stories fail at the architectural level — flat tension curves, interchangeable characters, settings that are backdrops instead of forces, dialogue where everyone sounds the same, violence without cost, and foreshadowing that doesn't exist. No amount of sentence-level cleanup fixes structural failure.
Writing Intelligence doesn't scrub. It compiles.
Every piece of writing passes through a 7-pass pipeline that enforces voice, argument structure, evidence discipline, paragraph architecture, and sentence-level variance. For fiction, it also enforces scene construction, character deployment, power dynamics, tension mechanics, and foreshadowing integrity.
| Pass | Name | What It Does |
|---|---|---|
| 0 | Mission Lock | Declare intent, audience, voice, genre, scale |
| 1 | Diagnostic Scan | Identify AI residue, argument gaps, cadence repetition |
| 2 | Structural Rewrite | Fix section order, paragraph sequence, thesis placement |
| 3 | Sentence Surgery | Anti-slop rules with hard/soft/earned-exception tiers |
| 4 | Voice Restoration | Reinject authorial identity so cleanup doesn't flatten |
| 5 | Scene & Chapter Audit | Audit setting, props, roles, power, pacing, foreshadowing (v2.0) |
| 6 | Stress Test | Interrogate for weakness, genericity, cuttable content |
| 7 | Output Modes | Return clean/annotated/redline/scorecard/scene-audit |
The difference between removal and compilation is the difference between a spell-checker and a compiler.
150+ banned phrases. 9 structural anti-patterns. 59 cadence signatures. Fake depth detection. Fake authority detection. Genre-specific cliché libraries for 8 domains.
Nine paragraph blueprints with physics auditing. Twelve opening/closing types. Six evidence integration techniques. Eight scene grounding methods. Power dynamics through objects, space, gesture, and environment (v2.0). Tension mechanics with 8-phase compression, silence taxonomy, and psychological warfare (v2.0).
Chapter Construction Doctrine — Setting as active antagonist. Props as narrative weapons (5-layer system). Drink orders as character DNA. Migrating power objects. Psychological warfare through dialogue (4-layer interrogation, casual menace pattern). 8-phase slow-burn compression model. Silence as weapon. Planted detail doctrine. Identity/deception architecture. 15-point construction checklist.
12 Character Role Archetypes — Covert Operative, Double Agent, Interrogator, Enforcer, Celebrant, Observer, Overlooked Ally, Innocent, Ghost, Shadow, Fall Guy, Decoy. Profile requirements, inner conflict, tension contribution, tells to plant, fatal vulnerability. Role combination rules and deployment recommendations by chapter type.
Thriller Scene Architecture — Confined-space doctrine (9 archetypes). Cold open mechanics. Five structural beats. The turn. The button (5 types). Violence architecture. WWGW diagnostic framework.
Transmedia Character Deepening — Character diaries. Interactive web serials. Lore bibles. Podcast series. Audio dramas. Guided meditations. Animated shorts. Virtual tours. Scavenger hunts. AR experiences. Cross-format consistency rules. Canon hierarchy. Production prioritization matrix.
Seven pre-built voices — sovereign commander, literary recursive, sermon (Black church tradition), investor precision, founder manifesto, academic rigorous, casual sharp — plus a custom voiceprint builder.
Strategy, fiction (v2.0), sales, academic, speech, sermon, email, pitch deck, legal positioning, cinematic narration, dialogue (v2.0), government brief, medical writing, patent claims, thriller scene architecture (v2.0), transmedia character (v2.0).
100-point scoring across 10 axes. 8 auto-fail conditions. 19 rewrite operators (+5 in v2.0). 25 failure modes. Auto-diagnostic report with violation heatmaps.
Section architecture. Continuity. Style transfer. Auto-mode. Memory system. Cross-document audit. Chapter construction (v2.0). Character role archetypes (v2.0).
Rubric compiler. Claim-evidence binding. Citation intelligence. Anti-hallucination rules. 10 discipline reasoning libraries.
16 adversarial stress tests (+6 in v2.0). 10 gold outputs (+3 in v2.0). Detector benchmark methodology for 7 tools.
| System | Points | What It Measures |
|---|---|---|
| Prose Quality | 100 | Clarity, specificity, rhythm, voice, argument, evidence, density, audience, memorability, structure |
| Chapter Construction | 100 | Setting, props, tension, power dynamics, pacing, foreshadowing, roles, identity, fatal detail, aftermath |
| Dialogue | 100 | Voice distinction, subtext, tension elements, rhythm, conflict, attribution, compression, exposition, silence, memorability |
| Power Dynamics | 100 | Power object, dual-purpose objects, consumption, spatial coding, gesture warfare, environment |
| Tension Mechanics | 100 | Compression model, false relief, fatal detail, silence, psychological warfare, atmosphere |
| Thriller Scene | 100 | Confined space, cold open, five beats, the turn, the button, violence, WWGW progression |
| Transmedia | 100 | Voice consistency, canon integrity, standalone quality, cross-platform integration |
| Capability | stop-slop | deslop | Writing Intelligence v2.0 |
|---|---|---|---|
| Files | 4 | ~6 | 58 |
| Words | ~2,500 | ~4,000 | 68,770 |
| Phrase bans | ~40 | ~30 categories | 150+, tiered |
| Cadence signatures | ~6 | ~6 | 59 |
| Voiceprints | 0 | 0 | 7 + custom builder |
| Genre packs | 0 | 0 | 16 |
| Paragraph architecture | No | No | 9 blueprints + physics |
| Fiction intelligence | No | No | Full engine (4 files, 1,600+ lines) |
| Chapter construction | No | No | 11-section doctrine |
| Character archetypes | No | No | 12 roles with deployment rules |
| Dialogue system | No | No | 7 laws + 9 elements + 15 techniques |
| Power dynamics | No | No | Objects, space, gesture, environment |
| Tension mechanics | No | No | 8-phase compression model |
| Thriller architecture | No | No | Confined-space doctrine + WWGW |
| Transmedia | No | No | 10 formats with production matrix |
| Scoring systems | 50 pts / 5 axes | 50 pts / 5 axes | 7 systems × 100 pts each |
| Approach | Subtractive | Subtractive | 7-pass compiler |
Claude.ai: Settings → Skills → Install from file → select writing-intelligence.skill from Releases
Claude Code:
git clone https://github.com/antonio0720/writing-intelligence .claude/skills/writing-intelligence
Claude Projects: Upload SKILL.md and the references/ folder to project knowledge.
Any LLM: Include SKILL.md in your system prompt. Reference files load on demand.
Then write. The skill triggers on any writing, editing, rewriting, scoring, fiction, dialogue, chapter construction, or style transfer task.
Clean up AI-sounding text:
Rewrite this to sound like a real person wrote it: [paste text]
Write a thriller scene from scratch:
Write a 1,200-word thriller scene set in a basement restaurant. Two characters — one is undercover, the other suspects it. Use confined-space doctrine. Deploy at least 4 character roles. Follow the 8-phase compression model. Score it on both prose quality and chapter construction when done.
Score your fiction chapter:
Run a full scene audit on this chapter. Score it on all 5 applicable systems: prose quality, chapter construction, dialogue, power dynamics, and tension mechanics. Show me what's strong and what needs work.
[paste chapter]
Build a character voice:
Build a custom voiceprint from these 3 writing samples. Then write dialogue for this character that is distinguishable from every other character in the scene.
[paste 3 samples]
Audit your dialogue:
Run the dialogue stress test on this scene. Check voice distinction (can I tell who's speaking with names removed?), subtext depth, tension elements active, and attribution technique. Score it on the dialogue v2.0 rubric.
[paste dialogue]
For 200+ more prompts organized by task, genre, voice, and skill level, see USER_GUIDE.md.
writing-intelligence/
├── SKILL.md # 7-pass compiler core (v2.0)
├── README.md # This file
├── USER_GUIDE.md # Complete how-to guide with 200+ prompts
├── CONTRIBUTING.md # Contribution guidelines
├── LICENSE # MIT
├── references/
│ ├── anti_patterns/ # 7 files — detection library
│ ├── positive_patterns/ # 6 files — construction system (+2 in v2.0)
│ ├── voiceprints/ # 8 files — voice profiles + builder
│ ├── genre_packs/ # 16 files — domain-specific rules (+2 in v2.0)
│ ├── diagnostics/ # 4 files — scoring + operators + failures
│ ├── compiler/ # 8 files — architecture + memory + fiction engine (+2 in v2.0)
│ └── academic/ # 2 files — college writing + disciplines
└── tests/ # 5 files — benchmarks + adversarial + gold
58 files. Every one earns its place.
Pull requests welcome. Read CONTRIBUTING.md for guidelines.
v2.0 contribution areas: new character role archetypes, thriller sub-genre packs, transmedia format templates, fiction genre extensions (mystery, horror, romance, literary), voiceprints for fictional character types, power dynamics case studies, tension mechanic variations, international dialogue patterns, additional adversarial test cases, and detector benchmark results.
MIT. Use it, fork it, extend it, teach with it, build on it.
Credit appreciated: Antonio T. Smith Jr. / Density6 LLC
Antonio T. Smith Jr. is the Founder and CEO of Density6 LLC and the inventor of Operational Machine Consciousness (OMC) and Existential Recursive Intelligence (ERI). He builds sovereign infrastructure across AI automation, institutional trading, financial literacy gaming, publishing, and revenue operating systems.
Writing Intelligence was built because the existing tools stopped at removal. v1.0 started where they stopped — it compiled prose with authorial identity, argument discipline, and structural integrity. v2.0 goes further. It treats fiction as an engineered system — with load-bearing components at every level from sentence to storyworld.
It's free because access to powerful language shouldn't depend on what you can afford. The student at 2 AM and the executive before the board meeting deserve the same tool. So does the novelist at 3 AM and the screenwriter before the pitch.
Writing should sound like a person wrote it. Fiction should feel like a person lived it. This makes sure both happen.
