The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous utility commands using ffmpeg across several files (including references/edit-qa-retakes.md, workflows/cinematic-shot.md, references/logo-and-mark-motion.md, and references/product-beauty-macro.md) to perform video normalization, scene detection, and frame sampling. These commands are legitimate tools intended for the skill's primary purpose of video generation and editing. Specifically:
ffmpeg is used to scale, pad, and re-encode generated clips to a standard delivery format.
ffmpeg scene detection (select='gt(scene,0.3)') is used to verify that the generated video contains the expected number of cuts.
ffmpeg tiling is used to generate contact sheets for visual quality inspection.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests and processes untrusted media data.
Ingestion points: The agent is instructed to read and analyze user-supplied images and generated video files using vision capabilities and shell tools (SKILL.md, workflows/cinematic-shot.md).
Boundary markers: While instructions emphasize visual inspection, there are no explicit delimiters or warnings to ignore adversarial content embedded in media metadata or visual frames.
Capability inventory: The agent has the capability to execute complex shell commands (ffmpeg) and interact with file system paths based on processed data.
Sanitization: There is no evidence of metadata sanitization or filtering for the processed media files.

krea-motion