Code Broker: A Multi-Agent System for Automated Code Quality Assessment

Samer Attrah has introduced Code Broker, a multi-agent system built on Google's Agent Development Kit (ADK) that automates code quality assessment for Python. The system employs a hierarchical architecture with five specialized agents: a root orchestrator coordinates a sequential pipeline agent, which dispatches three agents in parallel—a Correctness Assessor, a Style Assessor, and a Description Generator. A final Improvement Recommender synthesizes findings into actionable reports. Code Broker scores code across four dimensions—correctness, security, style, and maintainability—and outputs results in both Markdown and HTML formats.

Code Broker combines LLM-based reasoning with deterministic static-analysis signals from Pylint, using asynchronous execution with retry logic for robustness. It also explores lightweight session memory to retain and query prior assessment context. The system can analyze code from local files, directories, or GitHub repositories. Preliminary qualitative evaluation on representative Python codebases suggests that parallel specialized agents produce readable, developer-oriented feedback. However, limitations include evaluation depth, security tooling, large repository handling, and reliance on in-memory persistence. All code and reproducibility materials are publicly available.