Knowledge-Guided Multi-Agent Framework for Application-Level Software Code Generation
Qian Xiong, Bo Yang, Weisong Sun, Yiran Zhang, Tianlin Li, Yang Liu, Zhi Jin
TL;DR
The paper tackles the challenge of applying large language model–driven code generation to complex, application-level software, hindered by context fragmentation, grounding gaps, and lack of iterative refinement. It proposes the Knowledge-Guided Application-Level Code Generation (KGACG) framework, which orchestrates three agents—Code Organization & Planning Agent (COPA), Coding Agent (CA), and Testing Agent (TA)—in a closed-loop process to transform SRS and ADD into executable code, augmented by external knowledge sources and a feedback mechanism. Through a Java Tank Battle case study, KGACG demonstrates iterative collaboration: COPA derives detailed implementation plans and project structure, CA implements and refines code with API knowledge and self-debugging, and TA designs and evaluates unit tests with traceability to requirements. The framework aims to advance automation and maintainability in large-scale, real-world software development by integrating structured planning, coding, and testing within a knowledge-guided, multi-agent system.
Abstract
Automated code generation driven by Large Lan- guage Models (LLMs) has enhanced development efficiency, yet generating complex application-level software code remains challenging. Multi-agent frameworks show potential, but existing methods perform inadequately in large-scale application-level software code generation, failing to ensure reasonable orga- nizational structures of project code and making it difficult to maintain the code generation process. To address this, this paper envisions a Knowledge-Guided Application-Level Code Generation framework named KGACG, which aims to trans- form software requirements specification and architectural design document into executable code through a collaborative closed- loop of the Code Organization & Planning Agent (COPA), Coding Agent (CA), and Testing Agent (TA), combined with a feedback mechanism. We demonstrate the collaborative process of the agents in KGACG in a Java Tank Battle game case study while facing challenges. KGACG is dedicated to advancing the automation of application-level software development.