The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell command execution to process visual assets and render results. This includes using ffmpeg for extracting frames from reference videos and npx to run hyperframes and remotion for look test rendering. These commands use parameters derived from project metadata and user inputs.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it processes untrusted data from loglines, external source URLs, and reference videos/images.
Ingestion points: Untrusted data enters the agent context through loglines, external URLs, and reference media files specified in the project direction.
Boundary markers: The skill attempts to mitigate injection risks by instructing the agent in reference/style-bible-template.md to record explicit "take" and "do not take" lines, which helps distinguish source content from the intended style adaptation.
Capability inventory: The skill possesses significant capabilities including the ability to read and write local files (specifically ~/.motion-studio/taste.json) and execute shell commands (ffmpeg, npx).
Sanitization: The instructions do not specify any explicit sanitization, filtering, or escaping of the ingested external content before it is processed by the agent.

motion-look