Level of Autonomy (LoA) switching is pivotal for cultivating trust and credibility in human-machine collaborative decision-making between manned and unmanned systems. While existing research has concentrated on the timing design of LoA switching, the applicational integration with autonomous decision-making algorithms remains underexplored. This paper proposes a modular collaborative framework that couples LoA switching with autonomous algorithms via a behavior tree structure, enabling dynamic integration and scaling of various target allocation algorithms (e.g., ant colony optimization) and behavioral planning algorithms (e.g., reinforcement learning). Additionally, the framework incorporates a screening mechanism for various autonomous decision-making models to select the ones with better performance for current situation. In the context of UCAV formation autonomous airborne interception scenarios, the collaborative decision-making performance of different LoA modes is evaluated by win rates and time cost. The proposed method could enhance the synergy between human and machine while improving the controllability and interpretability of autonomous systems.