The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/qwen_audio_studio.py executes system commands via subprocess.run to invoke ffmpeg and ffprobe for audio validation and metadata extraction. These calls use argument lists (shell=False) and include validation for user-supplied output filenames to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill connects to Alibaba Cloud's Beijing region endpoints (aliyuncs.com) to generate and download audio files. This is a well-known service and essential to the skill's primary function.
[CREDENTIALS_SAFE]: The skill implements best practices for secret management by requiring DASHSCOPE_API_KEY and SFM_WORKSPACE_ID to be stored in environment variables. The Python script includes a comprehensive redact function that scrubs these credentials and Base64-encoded audio data from all logs and generation reports.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text to construct complex prompts for the TTS model. It maintains a low risk profile by enforcing a 3000-character limit and utilizing the processed text exclusively for audio generation, which has limited potential for influencing the agent's execution environment.
Ingestion points: User input via --prompt or --prompt-file in scripts/qwen_audio_studio.py.
Boundary markers: Mode-specific instructional templates are used, though explicit boundary delimiters are not added to the interpolated user text.
Capability inventory: File system access for audio references, network access via requests for the API, and execution of ffmpeg/ffprobe via subprocess.
Sanitization: Character count validation and regex-based filtering of output filenames.

z-qwen-audio-studio