In the process of software development, software libraries are widely used as they can reduce development time and costs. Consequently, modern software projects contain code from different sources, which makes the systems highly complex and diversified. In addition, various risks come along with the usage of software libraries, such as low quality or security vulnerabilities, seriously affecting the quality of software projects. By analyzing the intensity of the coupling with software libraries, this study quantifies the complexity and diversity introduced by the dependence on the software libraries to the client code. For this purpose, a software boundary graph (SBG) model is constructed according to the method invocation relationships of the client code with the software libraries to distinguish their code boundaries. Then, a metric suite RMS for the complexity of the software library dependency graph is proposed on the basis of the SBG model to quantify the intensity of the coupling with the software from different sources. In the experiment, this study mines the data on all the historical versions of 10 popular software in the Apache open-source community and finally collects 7857 dependency defects among real-world projects. With the above-mentioned real-world data, empirical investigation based on hypothesis testing is conducted according to the proposed complexity metric suite RMS to discuss the following issues: H1: whether boundary nodes with higher risk factors are more likely to introduce more inter-project dependency defects; H2: whether boundary nodes with higher risk factors are more likely to introduce serious inter-project dependency defects; H3: what is the extent to which the value of the metric suite RMS affects the number and severity of introduced inter-project dependency defects. Experimental results show that according to the evaluation with the RMS, the boundary nodes exhibiting higher coupling degrees with the software libraries are more likely to introduce more inter-project dependency defects with higher severity. Moreover, compared with traditional complexity metrics, RMS greatly influences the number and severity of introduced inter-project dependency defects.