<p>Future humanoid robots must be reactive and dexterous when operating in dynamic, human-centered environments. To address the challenges of task-space constrained manipulation and reactive motion planning resulting in locally optimal robot motion in complex unstructured environments, we propose a general framework for humanoid upper-body robots (HUBRs), which enables three distinct reactive motion modes, to fill the research gap about unifying tightly coupled coordinated bimanual manipulation, reactive motion, and overcoming the local minima of reactive motion planning: (1) Tightly coupled coordinated bimanual reactive collision-free motion for any type of obstacles, which is achieved by vector field (VF), cooperative dual task space (CDTS), and our proposed safety filter based on control barrier function (CBF). (2) The robot arms proactively escape from the local minima induced by concave obstacles and reactive motion via the proposed local VF modulation strategy that incorporates the novel force navigation kernel and the kernel parameter optimization based on sampling-based model predictive control (MPPI). This motion mode is automatically activated by evaluating an efficient scalar-valued activation function. (3) Reactive coordinated waist-arm motion based on the dual-arm reachability analysis for dual-arm workspace expansion. This motion mode is triggered when dual-arm manipulability or the tilt angle of the dual-arm end-effectors reach the threshold value. Various planning methods are thoroughly validated in cluttered and dynamic simulation scenarios. The proposed framework out-performs existing solutions in constrained manipulation tasks. We also demonstrate the benefits of the proposed framework in real-world experiments.</p>