The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.check_output to execute local commands such as ffmpeg, ffprobe, and internal Python scripts. All subprocess calls are implemented using argument lists rather than shell strings, which effectively mitigates command injection risks from user-supplied text or file paths.
[DATA_EXPOSURE]: The skill manages local audio assets and voice profiles within a dedicated project directory (defaulting to ~/dev/VoiceCloning). It follows best practices by treating source recordings as immutable and requiring user authorization for profile creation. No access to sensitive system paths or credentials was detected.
[REMOTE_CODE_EXECUTION]: The skill relies on locally installed MLX libraries and pre-downloaded model weights. It does not perform any network operations to fetch external scripts or executable content during runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text to be synthesized into speech. While this represents a data ingestion surface, the process is limited to local text-to-speech generation and re-transcription for verification. The user input does not influence the logic of the shell commands or the agent's operational constraints.

voice-clone-skill