ai-teaching-media
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the subprocess module to execute system commands such as ffmpeg, ffprobe, and macOS 'say' for media processing. It also executes 'npx hyperframes' for video rendering across multiple scripts including build.sh and build_video.sh.
- [EXTERNAL_DOWNLOADS]: The skill requires runtime installation of Python packages like 'edge-tts' and 'Pillow' via pip and relies on 'npx' to execute Node.js packages at runtime, which may download the packages if not present in the environment.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from user-provided technical articles and knowledge points, then interpolates this data into AI generation prompts without boundary markers or isolation instructions. This creates an indirect prompt injection surface. The skill possesses extensive capabilities, including shell execution and network requests, that could be exploited through malicious content in processed data.
- [DATA_EXFILTRATION]: The skill includes functionality to read local files and transmit their contents to external AI providers for processing, such as uploading audio for voice cloning in clone_voice.py or using local images as generation references in generate.py. This could be exploited to exfiltrate sensitive files if file paths are manipulated.
Audit Metadata