video-to-superprompt
Fail
Audited by Gen Agent Trust Hub on Jul 19, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute
ffprobeandffmpegshell commands using the$VIDEOvariable, which is populated by user-supplied local paths or URLs. Direct interpolation of untrusted input into shell commands without explicit sanitization steps creates a high risk of command injection, where a malicious user could execute arbitrary system commands by providing a crafted file path or URL. - [PROMPT_INJECTION]: The skill processes untrusted external data (video content and metadata) to generate builder-ready prompts. This establishes a surface for indirect prompt injection where instructions hidden within the video's metadata or visual content could manipulate the agent's output or behavior.
- Ingestion points: Local file paths, uploaded files, and external URLs (referenced in
SKILL.md). - Boundary markers: The skill lacks delimiters or explicit instructions to ignore embedded commands within the analyzed media.
- Capability inventory: Shell command execution (
ffmpeg,ffprobe), file system operations (mkdir), and Git repository interactions (commit). - Sanitization: There are no mechanisms described for validating, escaping, or filtering the input used in shell commands or prompt generation.
Recommendations
- AI detected serious security threats
Audit Metadata