The Agent Skills Directory

[COMMAND_EXECUTION]: The included utility scripts/review_frames.py executes system binaries ffmpeg and ffprobe using the subprocess module. The script is designed to process user-provided video files to generate review artifacts. While it uses a list-based argument structure to avoid shell injection, it remains a capability for local command execution driven by agent-provided inputs.
[DYNAMIC_EXECUTION]: The Python utility dynamically constructs command-line arguments and complex filter expressions for the ffmpeg binary based on time and frame parameters calculated at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external assets such as brand documents, reference videos, and product screens, creating a vulnerability surface.
Ingestion points: SKILL.md and references/review.md explicitly describe the agent reading and analyzing user-supplied video files, images, and design documents.
Boundary markers: The instructions lack specific delimiters or isolation techniques to prevent untrusted content within these assets from being interpreted as instructions by the agent.
Capability inventory: The skill includes the ability to execute shell commands via the bundled review_frames.py script.
Sanitization: The skill lacks content-level sanitization or security validation for ingested media and documents beyond basic file existence checks.

video-director