FLARE: Agentic Coverage-Guided Fuzzing for LLM-Based Multi-Agent Systems

2026-04-06

AI Summary

This work addresses the challenge of effectively testing multi-agent large language model (LLM) systems, which are prone to failures such as infinite loops and tool invocation errors due to their non-deterministic interactions, rendering traditional testing methods inadequate. To overcome this, the authors propose FLARE, a novel framework that introduces coverage-guided fuzzing to this domain for the first time. FLARE automatically extracts agent specifications and behavioral spaces from source code to construct test oracles and leverages semantic analysis of execution logs to assess semantic correctness. Evaluated on 16 open-source applications, FLARE achieves 96.9% inter-agent and 91.1% intra-agent coverage, substantially outperforming baseline approaches, and uncovers 56 previously unknown faults unique to multi-agent LLM systems.

Abstract

Multi-Agent LLM Systems (MAS) have been adopted to automate complex human workflows by breaking down tasks into subtasks. However, due to the non-deterministic behavior of LLM agents and the intricate interactions between agents, MAS applications frequently encounter failures, including infinite loops and failed tool invocations. Traditional software testing techniques are ineffective in detecting such failures due to the lack of LLM agent specification, the large behavioral space of MAS, and semantic-based correctness judgment. This paper presents FLARE, a novel testing framework tailored for MAS. FLARE takes the source code of MAS as input and extracts specifications and behavioral spaces from agent definitions. Based on these specifications, FLARE builds test oracles and conducts coverage-guided fuzzing to expose failures. It then analyzes execution logs to judge whether each test has passed and generates failure reports. Our evaluation on 16 diverse open-source applications demonstrates that FLARE achieves 96.9% inter-agent coverage and 91.1% intra-agent coverage, outperforming baselines by 9.5% and 1.0%. FLARE also uncovers 56 previously unknown failures unique to MAS.

Problem

• Multi-Agent LLM Systems
• non-deterministic behavior
• agent interactions
• software testing
• failure detection

Innovation

• coverage-guided fuzzing
• multi-agent LLM systems
• test oracle generation
• behavioral space extraction
• failure detection