ASPIRE: A Multi-Agent Framework for Execution-Free Code Analysis and Repair

The proliferation of Large Language Models (LLMs) and agentic systems for code generation introduces new challenges in ensuring the reliability and correctness of generated programs. Errors in dynamically generated code can cascade through workflows, leading to inefficiencies or failures in critical systems. We present ASPIRE, a multi-agent framework designed to address these challenges by providing execution-free static analysis and laying the foundation for program repair. ASPIRE’s agents collaboratively simulate runtime traces, predict program states, and evaluate code correctness without requiring actual execution, making it particularly suited for LLM-generated code and agentic workflows. Experimental results on a curated dataset from Codeforces show ASPIRE achieving ~56% accuracy in verdict prediction, with a 33% improvement when employing multi-agent collaboration. While program repair remains an area of future work, ASPIRE demonstrates significant potential as a safeguard for dynamic, LLM-driven code generation systems.