video-analysis
Pass
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes external CLI tools to process media. It uses
python3 -m yt_dlpto download video content andffprobeto determine video duration. It also invokes thepostplusCLI for billing and credit confirmations during the analysis workflow. - [EXTERNAL_DOWNLOADS]: The skill downloads social media videos (TikTok, Reels) from external URLs to the local environment. These downloads are performed by
yt-dlpbased on source URLs provided in a manifest file. - [DATA_EXFILTRATION]: Local video files and metadata are uploaded to a hosted Gemini API service for analysis. This process involves transmitting session tokens and content to the PostPlus platform, which is consistent with the skill's stated purpose of providing cloud-based video insights.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where untrusted source URLs from a manifest are interpolated into the Gemini prompt in
scripts/run_video_analysis_batch.mjs. Ingestion points include the manifest-derived URLs; no boundary markers or sanitization are present to isolate the URL string from instructions. Capabilities include subprocess execution (yt-dlp, ffprobe), file system writes, and network operations via the PostPlus API. This finding is assessed as safe given the specific constraints of the prompt and the skill's utility.
Audit Metadata