video-content-extractor
Warn
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONNO_CODEPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documentation instructs the agent to invoke system binaries (FFmpeg, Tesseract) and execute local Python scripts (e.g.,
scripts/extract_video.py) with user-supplied arguments. This creates a risk of command injection if arguments are not properly sanitized. - [NO_CODE]: The skill references a primary execution script
scripts/extract_video.pythat is missing from the provided package. This 'shadow code' prevents a security audit of how the skill handles subprocesses, file paths, or network requests, despite the author's claims of safety. - [PROMPT_INJECTION]: The skill ingests untrusted data from video files and converts it into text transcripts presented to the agent. This is an indirect prompt injection vector where text hidden in video frames could be used to override agent behavior.
- Ingestion points: Video frames processed by Tesseract OCR.
- Boundary markers: None mentioned in instructions.
- Capability inventory: Subprocess execution for FFmpeg and Python scripts.
- Sanitization: None described.
Audit Metadata