voice-changer
Warn
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script
scripts/rvc_infer_real.pyusestorch.load()with the parameterweights_only=False. This enables unsafe deserialization of PyTorch checkpoints (.pth files), which allows arbitrary code execution if a maliciously crafted model file is loaded by the user. - [COMMAND_EXECUTION]: The installation script
install_dependencies.shutilizessudoto install system packages (ffmpeg). Execution of commands with elevated privileges poses a risk of unauthorized system modification or privilege escalation if the script is subverted. - [COMMAND_EXECUTION]: The skill makes extensive use of
subprocess.run()to interface with external binaries likeffmpegandffprobe, as well as to execute its own internal Python modules. While these calls generally use list-based arguments to prevent shell injection, they grant the agent significant control over the host's media processing tools. - [DATA_EXFILTRATION]: The configuration file
config/voice_config.jsoncontains hardcoded absolute file paths (e.g.,/Users/m/.claude/skills/voice-changer/...). This configuration reveals specific host system details, including the local username ('m') and the internal directory structure.
Audit Metadata