The Agent Skills Directory

[COMMAND_EXECUTION]: The main script scripts/compare.py executes system commands via subprocess.run to invoke FFmpeg and FFprobe. These operations are fundamental to the skill's purpose of extracting video metadata and quality metrics. The implementation follows security best practices by using list-based arguments (avoiding shell=True), implementing strict path resolution via Path.resolve(), and enforcing timeouts to prevent resource exhaustion.\n- [EXTERNAL_DOWNLOADS]: The HTML report template (assets/template.html) includes references to external CSS and JavaScript files hosted on unpkg.com for the img-comparison-slider component. This is a well-known service recognized as a safe source for web assets, and the reference is limited to the generated report's visual presentation.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external video files provided by the user, creating a vulnerability surface for indirect prompt injection. However, the risk is categorized as low because the tool extracts binary-level pixel metrics (PSNR, SSIM) and technical metadata rather than natural language content that could influence agent behavior.\n
Ingestion points: Video files processed by scripts/compare.py.\n
Boundary markers: Absent for binary video data.\n
Capability inventory: FFmpeg binary execution, file writing to the output path, and temporary directory management.\n
Sanitization: Robust validation of file extensions against a whitelist, path resolution to prevent traversal, and file size limits (500MB).

video-quality-diff