The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates video processing tasks such as normalization, concatenation, and frame extraction by invoking ffmpeg. Analysis of scripts/assemble_edit.py, scripts/sample_qa_frames.py, and scripts/_common.py confirms that these subprocesses are initiated using list-based arguments rather than shell strings, effectively mitigating risks of command injection.
[EXTERNAL_DOWNLOADS]: scripts/poll_video_jobs.py contains logic to download completed video assets from remote URLs. These URLs are retrieved from metadata generated by the associated Krea MCP service, forming a standard part of the asset retrieval workflow for the animation pipeline.
[DATA_EXPOSURE]: The skill manages internal production data, including script beats, shot manifests, and asset URLs, within a local project directory structure. No access to sensitive system paths or hardcoded credentials was detected.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied creative inputs (storyboard details and motion prompts) which are interpolated into payloads for AI video generation tools. The workflow explicitly requires human approval at multiple gates (story spine, asset sheets, and keyframes) before final video generation, providing significant protection against unintended model outcomes.

krea-animation