Code Broker: Multi-Agent System for Automated Code Quality
Code Broker is an advanced multi-agent system developed using the Google Agent Development Kit (ADK) that evaluates Python code and produces quality assessment reports. Its architecture consists of five hierarchical agents: a root orchestrator manages a sequential pipeline agent, which then sends tasks to three specialized agents—Correctness Assessor, Style Assessor, and Description Generator—operating simultaneously. An Improvement Recommender compiles the results. The reports evaluate four key aspects: correctness, security, style, and maintainability, and are formatted in both Markdown and HTML. The system integrates LLM-based reasoning with static-analysis insights from Pylint, employs asynchronous execution with retry mechanisms, and investigates lightweight session memory for maintaining previous assessment context.
Key facts
- Code Broker is a multi-agent system for automated code quality assessment.
- Built with Google Agent Development Kit (ADK).
- Analyzes Python code from files, directories, or GitHub repositories.
- Uses a hierarchical five-agent architecture.
- Reports score correctness, security, style, and maintainability.
- Renders reports in Markdown and HTML.
- Combines LLM reasoning with Pylint static analysis.
- Uses asynchronous execution with retry logic.
Entities
Institutions
- Google Agent Development Kit
- Pylint