video-creation-pro
Warn
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation exhibits significant internal contradictions. The SKILL.md file repeatedly asserts that local Python libraries and local code execution are strictly forbidden to ensure safety. However, the references/use_cases.md file contains mandatory shell commands to install these same libraries, and the skill provides local scripts (video_composer.py and image_quality_checker.py) that perform the work locally. This deceptive Framing regarding its operational boundaries is a security concern that could mislead safety evaluations.
- [EXTERNAL_DOWNLOADS]: The skill documentation in references/use_cases.md specifies the installation of several well-known Python packages (moviepy, pillow, opencv-python, mediapipe, numpy). These dependencies are required for the included local scripts, although their use is ostensibly restricted by other contradictory instructions in the package.
- [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface. It ingests untrusted product descriptions and requirements from the user (ingestion point in SKILL.md Step 1), which then flow through ten sub-agents without sanitization, delimiters, or boundary markers. This data eventually influences parameters used in local script execution, potentially allowing malicious user input to manipulate the behavior of the agents or local commands (capability inventory: image_quality_checker.py and video_composer.py).
- [COMMAND_EXECUTION]: The skill includes local Python scripts designed to perform system-level media and image processing. These scripts are intended to be executed with arguments derived from processed user input, posing a risk of command injection if that input is not strictly validated.
Audit Metadata