The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external media files (provided via the source argument or SDI ingest) as its primary input. This creates an indirect prompt injection surface where maliciously crafted media metadata could attempt to influence the agent's behavior during the probing and analysis steps.
Ingestion points: Source media files and SDI streams processed by ffmpeg-probe and media-mediainfo (SKILL.md, Step 2).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when processing source data.
Capability inventory: The skill utilizes file system access, complex command execution (ffmpeg, hdr-dovi-tool), and network upload capabilities (media-cloud-upload).
Sanitization: No explicit sanitization or validation of the media source's internal metadata is described.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of specialized media processing tools (e.g., decklink-tools, ffmpeg, hdr-dovi-tool, vfx-oiio). These commands are documented for professional mastering tasks and do not show signs of arbitrary command injection or malicious intent.
[EXTERNAL_DOWNLOADS]: The skill documents delivery workflows involving uploads to legitimate targets such as Netflix (via Aspera) and Amazon (via S3) using the media-cloud-upload tool. These operations are aligned with the skill's stated purpose for broadcast delivery.

workflow-broadcast-delivery