Abstract

High-quality fuzz harnesses are essential for effective gray-box fuzzing. While Large Language Models (LLMs) offer promise for automating this task, existing one-turn generation methods suffer from hallucinations and inadequate coverage due to coarsegrained function targeting and misaligned generation workflows. We present SynapseFlow, an automatic harness generator that addresses these limitations through two key innovations: dataflowaware function aggregation and a staged, rollback-enabled generation workflow decomposition. SynapseFlow first analyzes source code to construct Structural Flow Graphs and extract coherent Function Triplets. It then synthesizes harnesses via a decomposed fourstage process governed by a staged rollback algorithm to ensure correctness. We evaluated SynapseFlow on 25 real-world opensource software projects. The experimental results indicate that SynapseFlow outperforms state-of-the-art tools (OSS-Fuzz-Gen, CKGFuzzer, PromeFuzz), achieving 3.07, 1.71, and 4.26 higher branch coverage, and 1.77, 1.51, and 1.36 higher bug detection rates, respectively. Most importantly, SynapseFlow discovered 7 previously unreported bugs (5 assigned CVEs), demonstrating its practical effectiveness in real-world bug discovery.