In natural language we often use graded concepts, reflecting different intensity degrees of certain features. Whenever such concepts appear in a given real-life context, they need to be appropriately expressed in its models. In this paper, we provide a framework which allows for extending the BGI model of agency by grading beliefs, goals and intentions. We concentrate on TEAMLOG [6, 7, 8, 9, 12] and provide a complexity-optimal decision method for its graded version TEAMLOG(K) by translating it into CPDLreg (propositional dynamic logic with converse and "inclusion axioms" characterized by regular languages). We also develop a tableau calculus which leads to the first EXPTIME (optimal) tableau decision procedure for CPDLreg. As CPDLreg is suitable for expressing complex properties of graded operators, the procedure can also be used as a decision tool for other multiagent formalisms.