omnivoice-tts
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches the OmniVoice package and source code from its official GitHub repository and installs PyTorch dependencies from the official PyTorch wheel index. It also includes instructions for using a community mirror (hf-mirror.com) to download model weights if Hugging Face is inaccessible.
- [COMMAND_EXECUTION]: Provides CLI commands for running local web demos and performing single or batch inference tasks using the installed omnivoice package.
- [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: Processes user-supplied text, reference audio files, and speaker attribute instructions (e.g., gender, accent) for synthesis in SKILL.md.
- Boundary markers: None identified; user input is passed directly to the model's generation function, which supports inline non-verbal tags.
- Capability inventory: Includes file system writes (saving audio via torchaudio), network operations (fetching model weights), and local shell execution of project-specific CLI tools.
- Sanitization: No explicit sanitization or validation of the input text or instruction strings is demonstrated before they are processed by the TTS model.
Audit Metadata