Large Language Model (LLM) agents have demonstrated remarkable capabilities in complex reasoning tasks, yet their application to offensive cybersecurity remains constrained by the scarcity of labeled vulnerability datasets and the rapidly evolving threat landscape.