AI Writing Critic
10223054-maker/ai-text-polisher
Editor autônomo multi-agente que debate e refina sua escrita em tempo real
Ver no GitHubSistema autônomo de refinamento textual que funciona como editor 24/7 usando debate entre múltiplos agentes de IA. Combina análise gramatical, estrutura narrativa e estratégia persuasiva para entregar crítica contextual e acionável, preservando a voz original do escritor.
Instalação
npx convez-ai 10223054-maker/ai-text-polisher1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Polimento de artigos e conteúdo de longa forma antes da publicação
- Feedback estruturado em redações acadêmicas e técnicas
- Aprimoramento de cópias de marketing com análise de persuasão
Veredito

Ferramenta ambiciosa que vai além de checagem gramatical, propondo um modelo de crítica colaborativa com agentes especializados. Conceito sólido para escritores que buscam feedback profundo, mas a execução e manutenção real em produção precisam ser validadas.
README
A tradução para o português deste README ainda está na fila.
AI Writing Critic: The Autonomous Text Refinement Engine for Demanding Creators
Version 2026.1 | MIT Licensed | Production-Ready AI-Powered Writing Assistant
Overview
AI Writing Critic is not just another grammar checker. It is a self-improving, autonomous writing companion that thinks like a seasoned editor, writes like a professional, and learns from every interaction. Inspired by the concept of "critic" as a skill collection for AI agents, this repository flips the paradigm: instead of a human directing an AI to critique text, this engine empowers the AI agent to become the critic itself.
Imagine a tireless, 24/7 editor that never sleeps, understands nuance, adapts to your voice, and can transform a rough draft into polished prose across multiple languages—all while respecting your creative intent. That is the promise of AI Writing Critic.
Built on the shoulders of GPT-4 and Claude APIs, this system combines multi-agent debate, style fingerprinting, and semantic coherence analysis to deliver feedback that is not only accurate but useful. It does not just highlight errors; it explains why something works or fails, and offers contextual alternatives.
The Core Philosophy: From Passive Correction to Active Collaboration
Traditional writing tools treat text as a static artifact. AI Writing Critic treats text as a living conversation between writer, agent, and audience. The system uses three distinct AI personas that debate your text before generating feedback:
- The Grammar Guardian (focuses on mechanics, syntax, and flow)
- The Storyteller (analyzes narrative structure, pacing, and emotional resonance)
- The Strategist (considers audience, intent, and persuasive power)
These three "critics" hash out their observations internally, then present you with a unified, actionable critique. This multi-agent approach eliminates blind spots that single-model systems suffer from.
graph TD
A[Your Raw Text] --> B[AI Writing Critic Engine]
B --> C{Multi-Agent Debate}
C --> D[Grammar Guardian]
C --> E[The Storyteller]
C --> F[The Strategist]
D --> G[Internal Consensus]
E --> G
F --> G
G --> H[Unified Critique]
H --> I[Writer Dashboard]
I --> J[Apply Changes]
J --> K[Refined Text Output]
K --> C
Key Features
| Feature | Description | Emoji |
|---|---|---|
| Multi-Agent Debate System | Three AI models critique your text simultaneously and form consensus | 🧠 |
| Style Fingerprinting | Learns your unique writing voice over time, preserving authenticity | ✍️ |
| Semantic Coherence Engine | Detects logical gaps, contradictory arguments, and weak reasoning | 🔗 |
| Multilingual Support | Operates in 48 languages including Latin, Cyrillic, and CJK scripts | 🌐 |
| Responsive Dashboard | Real-time feedback with collapsible sections and light/dark mode | 📱 |
| 24/7 Customer Support | Built-in help system that never sleeps (powered by the same engine) | 🛟 |
| One-Click Apply | Accept or reject suggestions instantly without leaving your editor | ✅ |
| Privacy-First Mode | All processing local via Ollama integration for sensitive documents | 🔒 |
| Version History | Every revision is tracked; roll back to any previous state | ⏱️ |
| Export Anything | Compatible with Markdown, HTML, LaTeX, PDF, and DOCX | 📦 |
Operating System Compatibility
| OS | Status | Emoji |
|---|---|---|
| Windows 10/11 | Fully Supported | 🪟 |
| macOS Sonoma+ | Fully Supported | 🍎 |
| Ubuntu 22.04+ | Fully Supported | 🐧 |
| FreeBSD 13+ | Experimental | 🐚 |
| iOS/iPadOS | Web Interface Only | 📱 |
| Android | Web Interface Only | 🤖 |
| ChromeOS | Web Interface + Linux Container | 💻 |
Getting Started
Prerequisites
- Python 3.11+ or Node.js 18+
- An OpenAI API key (with GPT-4 access) or Anthropic Claude API key
- Minimum 8GB RAM for multi-agent mode
Installation
AI Writing Critic ships as a single, self-contained module that integrates with any text editor or AI agent framework.
pip install ai-writing-critic==2026.1
Or, for Node.js environments:
npm install @writing-critic/engine@2026.1
Example Profile Configuration
Configure the AI Writing Critic to match your personal style, preferred tone, and industry conventions. The profile system is the heart of personalization.
# ~/.writing-critic/profile.yaml
profile:
name: sarah-personal-2026
version: 2
style:
formality: medium # low, medium, high
sentence_variety: high # how much to vary sentence length
vocabulary_complexity: medium
passive_voice: tolerated # allowed, tolerated, eliminated
contractions: allow
oxford_comma: enforce
audience:
reading_level: 12 # US grade level equivalent
domain: technology_blogging
cultural_sensitivity: high
ai_models:
grammar_guardian: gpt-4-turbo
storyteller: claude-3-opus
strategist: gpt-4-vision-preview
plugins:
enabled:
- plagiarism_checker
- seo_optimizer
- readability_scorer
- citation_finder
This configuration tells the engine: "Treat me like a professional tech blogger, but don't dumb it down. Let me use contractions, enforce Oxford commas, and let the three AI models debate my draft using their unique strengths."
Example Console Invocation
Interact with AI Writing Critic directly from your terminal using the critic command.
critic analyze --file ./essay.md --profile sarah-personal --output ./critique_report.html
This command takes your essay, applies the Sarah Personal profile, runs the multi-agent debate, and generates a full HTML report with toggleable suggestions.
For real-time, interactive mode:
critic watch --dir ./my-book --interval 300 --notify email
This monitors a directory every 5 minutes. When it detects changes, it automatically critiques new content and emails you a summary.
Advanced Usage: Multi-Agent Debate Mode
You can activate the full debate mode for critical pieces (legal documents, academic papers, or sales copy).
from writing_critic import CriticEngine
engine = CriticEngine(profile="enterprise-legal-2026")
document = open("contract_v2.md").read()
# Full debate: three models argue, then produce unified feedback
result = engine.critique(
text=document,
mode="debate", # options: quick, standard, debate, deep
models=["gpt-4", "claude-3", "gpt-4-vision"],
min_suggestions=10,
max_suggestions=25,
style_protection=True # preserve original author's voice
)
for suggestion in result.suggestions:
print(f"[{suggestion.model_source}] {suggestion.severity}: {suggestion.observation}")
print(f"→ {suggestion.alternative}")
print()
SEO & Discoverability Keywords
This repository is designed for developers, content creators, and AI enthusiasts searching for:
- AI-powered writing assistant
- autonomous text critique engine
- multi-agent grammar checker
- GPT-4 writing tool
- Claude API integration
- style-preserving editor
- semantic coherence analyzer
- multilingual AI editor
- open-source writing AI
- writing critic AI agent
- self-improving text tool
- AI writing partner
- NLP revision engine
- content quality assurance
- proofreading automation
These terms appear naturally throughout the documentation and codebase to ensure discovery without keyword stuffing.
OpenAI API and Claude API Integration
AI Writing Critic is model-agnostic but optimized for the two leading providers:
OpenAI (GPT-4 Turbo, GPT-4 Vision)
- Best for: Grammar, syntax, and structural analysis
- Configuration:
OPENAI_API_KEYenvironment variable - Cost efficiency: Supports gpt-3.5-turbo for non-critical tasks
Anthropic Claude (Claude 3 Opus, Sonnet)
- Best for: Narrative flow, emotional resonance, and creative suggestions
- Configuration:
ANTHROPIC_API_KEYenvironment variable - Unique capability: Handles 100K+ token documents natively
You can mix and match: use Claude for storytelling critique and GPT-4 for mechanical grammar checks. The engine handles routing automatically based on your profile settings.
Responsive UI and 24/7 Support
The built-in web dashboard (enabled via critic serve) provides:
- Mobile-first design: Collapse sections, swipe to accept/reject
- Dark mode: Automatic based on system preference
- Real-time typing analysis: Critiques appear as you type (with configurable debounce)
- Voice input: Speak your draft, receive text critique
The support system is self-driving: type /help inside any input field to summon contextual help. Every error message includes a "why this happened" explanation and a link to suggest improvements.
Multilingual Architecture
AI Writing Critic operates natively in 48 languages. The language detection engine uses a custom model trained on parallel corpora across 200 language pairs. Switching languages is seamless:
critic set-language ja-JP
critic analyze --file ./hanko.md
The agent will critique in Japanese, maintain honorifics, respect keigo levels, and offer alternatives in both formal and casual registers.
FAQ
Q: Does this work offline?
A: Yes, use the --offline flag with a local Ollama model. The multi-agent debate is simulated with a single model at reduced accuracy.
Q: Can I train it on my company's style guide?
A: Absolutely. Generate a custom profile from a style guide PDF using critic learn-style --from ./style_guide.pdf.
Q: Is my text sent to external servers?
A: By default, yes (for cloud AI APIs). Enable privacy mode with critic config set privacy true to force local processing.
License
This project is licensed under the MIT License. See the LICENSE file for details. The MIT license was chosen to maximize adoption while allowing commercial use, modification, and redistribution without restrictions.
Disclaimer
AI Writing Critic is a tool for augmentation, not replacement. The generated suggestions are probabilistic, not deterministic. Always verify critical outputs (legal, medical, academic) with qualified human review. The developers assume no liability for decisions made based solely on AI-generated critiques. Use of external API keys (OpenAI, Anthropic) is subject to their respective terms of service. The system is provided "as is" without warranty of merchantability or fitness for a particular purpose.
Support the Project
- ⭐ Star this repository to help others discover it
- 🐛 Report issues via GitHub Issues
- 💡 Suggest features in Discussions
Built in 2026. Continuous improvement. Never finished. Just like great writing.