When soft external short circuit (ESC) occurs on a lithium-ion battery module, all cells in the module exhibit fault characteristics similar to those of an internal short circuit (ISC), thereby posing a higher safety risk. However, due to the lack of differences between cells and the absence of obvious fault behavior, existing methods are not suitable for diagnosing soft ESC. To address this challenge, we use the cumulative current integration between different charging cycles as features, combined with clustering algorithms, to diagnose soft ESC on the module. Based on experimental data, the validation results indicate that this method can effectively identify soft ESC faults without being limited by operating conditions. It is also not affected by other module anomalies, such as ISC, state of charge consistency anomalies, and module capacity degradation, and is not influenced by data noise. Additionally, it is compatible with various clustering algorithms, demonstrating strong generalization and robustness.