voice-changer

总体看是一个与声明用途一致的本地媒体处理 skill，没有凭证收集、外传或异常网络数据流迹象，因此不像恶意技能。但其核心模型/环境来源不可从技能文档中充分验证，且额外模型通过第三方网盘分发，存在中等供应链风险。

The code appears to implement ordinary RVC voice conversion and contains no clear evidence of intentional malware. However, loading an untrusted checkpoint with weights_only=False creates a significant arbitrary-code-execution risk. Use only trusted model files or a safer, supported weights-only format.