vox-explainer
Pass
Audited by Gen Agent Trust Hub on Jul 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill makes legitimate network requests to
api.atlascloud.aito utilize various AI models including xAI TTS (voiceover), Seedream (keyframes), Gemini (animation), and MiniMax (music). All media generation is handled via a standard submit-and-poll asynchronous pattern. - [COMMAND_EXECUTION]: Extensive use of local binary tools
ffmpegandffprobeis required for the video assembly stage. The skill provides detailed shell recipes for conforming video clips, mixing audio with sidechain compression, and burning subtitles into the final MP4 output. - [DATA_EXFILTRATION]: No sensitive local data is accessed. The skill correctly instructs the agent to use an
ATLASCLOUD_API_KEYstored in environment variables rather than hardcoding credentials, following security best practices for API integration. - [PROMPT_INJECTION]: The skill's primary function is processing user-supplied topics to generate creative content. It lacks specific boundary markers or sanitization logic for these inputs, presenting a standard surface for indirect prompt injection (Category 8).
- Ingestion points: User-provided topic string used to derive the narrative script (SKILL.md).
- Boundary markers: Absent in the provided prompt templates.
- Capability inventory: Local shell execution via
ffmpeg(references/ffmpeg-assembly.md) and network operations via Pythonrequests(references/atlas-cloud-api.md). - Sanitization: No explicit escaping or validation of user-provided strings before inclusion in generated scripts or shell command strings.
Audit Metadata