The global expansion of the Web brings the global computing; and the increasing number of problems with increasing complexity & sophistication also makes collaboration desirable. In this paper, we presented a semantics-based framework for collaborative problem solving in agent grid by coupling joint intention and dynamic description logics (DDL), our previous work to extend description logics (DL) with a dynamic dimension to model the dynamic world. Capabilities and attitudes of agents were captured by actions, and formulas in DDL respectively. Thus representation components in our framework were conferred with well-defined semantics by relating them to some domain ontologies. We could employ reasoning on actions in DDL to help agents to find proper colleagues when collaboration is necessary, and the philosophy underlying Joint Intention to bind their actions to achieve their overall goal. The main strengths of our framework include: i) finding probably helpful agents in a semantically accurate way due to the employment of semantic information; ii) going much closer to industrial implementations while retaining the main express power of classical joint intention model.