video-content-extractor
Fail
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill requires the execution of an external script 'scripts/extract_video.py' which is not bundled with the skill. The instructions explicitly direct the user to download this script from an untrusted community GitHub repository ('github.com/274326424/video-content-extractor'). Executing unverified scripts from external sources represents a high risk of arbitrary code execution.
- [EXTERNAL_DOWNLOADS]: The skill documentation mandates fetching the primary execution logic from a remote repository, creating a dependency on external, unvetted content.
- [INDIRECT_PROMPT_INJECTION]: The skill performs OCR on video frames and includes the resulting text in a Markdown report. This creates a vulnerability where an attacker could embed malicious instructions as text within a video (e.g., hidden in a single frame or in the background). If the agent later reads the generated report, it may follow these injected instructions.
- Ingestion points: Video files provided via the 'video_path' parameter.
- Boundary markers: Absent; the skill does not specify any delimiters or warnings for the agent regarding the untrusted nature of the OCR-extracted text.
- Capability inventory: Subprocess execution (ffmpeg, tesseract) and file system write operations.
- Sanitization: No mention of filtering, sanitizing, or validating the text extracted via OCR before it is added to the report.
Recommendations
- AI detected serious security threats
Audit Metadata