The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches models from the official Qwen organization on Hugging Face (e.g., Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice), which is a well-known and trusted source.
[EXTERNAL_DOWNLOADS]: Recommends cloning the official Qwen3-TTS source code repository from GitHub (https://github.com/QwenLM/Qwen3-TTS.git).
[EXTERNAL_DOWNLOADS]: Instructions involve installing official Python packages qwen-tts and flash-attn from public registries.
[COMMAND_EXECUTION]: Includes standard commands for environment setup using Conda and package installation with pip.
[COMMAND_EXECUTION]: Provides instructions to launch a local Gradio web interface. While the use of --no-ssl-verify and binding to 0.0.0.0 are noted as insecure configurations for public-facing servers, they are standard for local development and testing of this specific tool.
[SAFE]: The skill exposes an indirect prompt injection surface through text-to-speech generation. 1. Ingestion points: text parameter in generate_custom_voice and other methods in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: File write via soundfile.write and command execution via qwen-tts-demo. 4. Sanitization: Absent. This surface is inherent to the skill's intended text-to-speech functionality.

qwen3-tts