The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mentions a run_code tool intended for "read-only probing and diagnostics of fetched media" using a sandboxed ffmpeg environment. This is presented as a functional capability for media metadata inspection rather than arbitrary command execution.
[EXTERNAL_DOWNLOADS]: The skill utilizes a download_media tool to fetch assets from public URLs. This is a core functionality for the video editing use case, allowing users to import remote assets into their projects.
[PROMPT_INJECTION]: The skill processes untrusted external data, including video transcripts and asset metadata, which creates a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through read_project (transcripts and asset metadata) and download_media (external files).
Boundary markers: The skill does not explicitly define delimiters or specific "ignore" instructions for the processed media content.
Capability inventory: The agent has access to run_code, delete_project, create_motion_graphic_from_code, and import_media tools.
Sanitization: The instructions do not specify any escaping or filtering mechanisms for the ingested transcripts or metadata.

openchatcut-plugin-basics