With the increasing demand for intelligent and sensor-driven systems, traditional monolithic agent architectures face critical limitations in scalability, adaptability, and seamless integration with heterogeneous sensing resources. The objective of this study is to develop a scalable intelligent task management framework capable of coordinating multiple sensing and reasoning components in dynamic environments. To address these challenges, we propose a hierarchical multi-agent architecture consisting of a Task Dispatching Layer, Domain Agent Layer, and Modular Tool Layer. Inter-agent collaboration is standardized through a custom Agent-to-Agent protocol, while sensing and tool functionalities are modularized through a Modular Communication Protocol. In addition, a dynamic negotiation mechanism is incorporated to support conflict resolution and multi-objective task optimization. A prototype system was implemented and evaluated using sensor-driven task scenarios. Experimental results demonstrate up to a 50% improvement in task success rate and significant gains in operational efficiency compared with conventional single-agent architectures. Despite these promising results, challenges remain in large-scale deployment, dynamic sensor reliability, and adaptive learning from long-term interactions. Future work will focus on integrating machine-learning-based negotiation strategies, adaptive sensor management mechanisms, and intelligent resource allocation to further enhance system robustness, scalability, and real-time decision-making capabilities.