Revisor de PR Automatizado com Claude
SYED-M-HUSSAIN/MLOps
Automatiza reviews inteligentes de PR via Claude no GitHub Actions
Ver no GitHubFramework modular integrado ao CI que automatiza revisão de pull requests usando Claude da Anthropic. Analisa diffs de código, aplica critérios de avaliação humano-cêntricas e posta feedback estruturado e acionável como comentários no PR.
Instalação
npx convez-ai SYED-M-HUSSAIN/MLOps1 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
- Automatizar revisão de código em repositórios com alto volume de PRs
- Garantir consistência de padrões de qualidade e segurança entre equipes
- Reduzir latência de feedback aos desenvolvedores em fluxo CI/CD
Veredito

Solução bem arquitetada e pronta para produção que integra Claude em workflows GitHub Actions. Ideal para times que buscam automatizar reviews com critérios transparentes, embora exija configuração inicial de prompts e critérios de avaliação personalizados.
README
A tradução para o português deste README ainda está na fila.
🤖 AI-Powered Code Review Automation with Anthropic Claude
Automate intelligent, consistent, and scalable code reviews directly in your GitHub pull requests using Anthropic's Claude API.
📌 Overview
This repository implements a modular, CI-integrated code review framework that leverages Anthropic Claude to automatically review pull requests. The system analyzes code diffs, applies human-centric evaluation criteria, and posts structured, actionable feedback as PR comments.
🧠 Core Objectives
- Automate high-quality PR reviews at scale
- Ensure consistency using criteria-driven analysis
- Enhance developer productivity by reducing review latency
- Utilize LLMs to identify bugs, suggest improvements, and uphold coding standards
🛠️ System Architecture
GitHub PR Event ─┬─▶ GitHub Actions Workflow
│
└─▶ Python App (scripts/)
├─ Fetch PR diff
├─ Load evaluation criteria
├─ Call Anthropic Claude API
├─ Parse response
└─ Post PR comment via GitHub API
📁 Repository Structure
MLOps/
├── scripts/ # Core logic
│ ├── main.py # Entry point
│ ├── reviewer.py # PR review pipeline
│ ├── github_utils.py # GitHub API interaction
│ ├── anthropic_utils.py # Claude API wrapper
│ ├── prompts.py # Advanced prompt engineering templates
│ ├── config.py # Environment and constants
│ └── utils.py # Helpers (e.g., markdown criteria loader)
│
├── evaluation_criteria/ # Human-aligned review principles
│ ├── README.md # Instructions for writing criteria
│ ├── code_quality.md
│ ├── performance.md
│ ├── documentation.md
│ └── security.md
│
├── .github/
│ └── workflows/
│ └── review.yml # GitHub Actions CI pipeline
│
├── .gitignore
└── README.md
🚀 Quick Start
✅ Prerequisites
- Python 3.10+
- GitHub Personal Access Token
- Anthropic Claude API Key
📥 Installation
git clone https://github.com/SYED-M-HUSSAIN/MLOps.git
cd MLOps
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
▶️ Run Locally
python scripts/main.py
📜 Writing Evaluation Criteria
Evaluation files are structured markdown documents under evaluation_criteria/, allowing modularity, explainability, and transparency of the review system.
Each file should include sections such as:
# Code Quality
### Positive Examples
- Uses meaningful variable names
- Follows single-responsibility principle
### Negative Examples
- Deeply nested conditionals
- Repetitive logic without abstraction
These files are parsed and embedded in the system prompt to guide Claude's analysis.
Refer to evaluation_criteria/README.md for detailed authoring guidelines.
🤖 Prompt Engineering
Prompt templates are constructed in scripts/prompts.py using structured prompt engineering best practices:
- System prompt defines the reviewer's role and tone.
- User prompt includes code diffs and target instructions.
- Prompts are contextualized using evaluation criteria.
This separation enables maintainable and testable interactions with the LLM.
🔁 GitHub Actions Integration
The .github/workflows/review.yml pipeline triggers the review on PR open or update.
on:
pull_request:
types: [opened, synchronize]
You can extend this pipeline to:
- Run tests on the Claude response
- Notify developers via Slack
- Auto-approve or label PRs
🧪 Testing and Validation
Unit tests (coming soon) will cover:
- Diff parsing logic
- Prompt formatting
- Claude response validation
- Markdown criteria loading
To run tests:
pytest tests/
🔐 Security Considerations
- Secrets such as API keys must be stored in GitHub Secrets or
.envfiles. - LLM responses are sandboxed and validated before GitHub posting.
- Avoid logging sensitive data.
🔄 Extending the System
You can easily:
- Add more criteria categories
- Support multi-file diffs
- Swap Claude with OpenAI/Gemini via modular prompt layers
- Track response accuracy with custom feedback loops
👥 Contributing
We welcome contributions in the form of:
- Improved prompts or criteria
- Support for additional models
- Enhancements to GitHub workflows
- Documentation improvements
Please follow Semantic Commit Messages and open a PR with a detailed description.
📃 License
This project is released under the MIT License. See LICENSE for more information.