The Agent Skills Directory

[PERSISTENCE]: An automated security scan identified the file models/rvc_core/rvc_infer.py as potentially containing a backdoor (flagged as FileRepMalware [Bd]). This suggests a high risk of unauthorized access or hidden malicious activity within the skill's distributed files.
[DYNAMIC_EXECUTION]: The script scripts/rvc_infer_real.py loads AI models using the torch.load() function with the weights_only=False parameter. This is a significant security risk as it enables the deserialization of arbitrary Python objects, allowing an attacker to execute malicious code on the host machine if a compromised model file is loaded.
[COMMAND_EXECUTION]: Multiple scripts, including scripts/voice_change.py and scripts/rvc_process_long.py, rely on subprocess.run() to execute shell commands for audio processing tasks using FFmpeg and FFprobe. While these are used for functional purposes, the pattern of executing shell commands and spawning child Python processes increases the potential attack surface.
[REMOTE_CODE_EXECUTION]: The skill's documentation in models/RVC_MODEL_GUIDE.md instructs the user to clone a large external repository from GitHub (github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git) and download binary assets from third-party hosting services. Executing code and loading binaries from these unverified external sources presents a supply chain risk where malicious payloads could be introduced into the environment.

voice-changer