The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx --yes hyperframes@latest to fetch the HyperFrames CLI from the npm registry.
[REMOTE_CODE_EXECUTION]: Executes code from the hyperframes package downloaded at runtime from a remote registry (npm).
[COMMAND_EXECUTION]: Runs system commands including ffmpeg, ffprobe, and npx to perform video editing tasks and environment verification.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted media files (video/audio) which are processed by ffmpeg and ffprobe. Maliciously crafted media files could potentially exploit vulnerabilities in these parsers or their metadata handling.
Ingestion points: Input media files specified in the asset manifest or generated by AI models.
Boundary markers: None explicitly defined for preventing injection through media metadata.
Capability inventory: Subprocess execution of ffmpeg, ffprobe, and npx, along with file system access.
Sanitization: Uses stable composition identifiers and validates title-safe margins, but lacks deep sanitization of media binary content.
[DYNAMIC_EXECUTION]: Uses npx to dynamically load and run the latest version of the hyperframes package, introducing a runtime dependency on an external source.

video-editing