voicebox-voice-synthesis
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs users to download pre-built binaries for various operating systems from 'https://voicebox.sh'. It also directs users to clone a source code repository from 'https://github.com/jamiepine/voicebox.git'. These domains are external to the predefined trusted organization list.
- [COMMAND_EXECUTION]: The installation guide includes commands to install the 'just' task runner ('brew install just' or 'cargo install just') and execute scripts from the cloned repository ('just setup', 'just dev'). These scripts perform automated environment configuration and dependency installation on the host system.
- [INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface where external input can influence the behavior of the TTS models. (1) Ingestion points: The 'text' field in POST requests to the '/generate' endpoint. (2) Boundary markers: Absent; the skill explicitly teaches the use of embedded tags and instructions (e.g., '(whisper)', '[laugh]') which are processed by the model to alter output style. (3) Capability inventory: The skill uses curl, fetch, and httpx to interact with a local API, and includes commands for reading local files ('curl -F') and executing tasks ('just'). (4) Sanitization: There is no mention of sanitizing or validating the input text to prevent malicious or unexpected control instructions.
- [COMMAND_EXECUTION]: Several curl examples demonstrate how to interact with the local API, including patterns that read from the local file system to upload data, such as '-F "file=@/path/to/voice-sample.wav"'.
Audit Metadata