The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external media files (images, videos, and audio) via user-supplied URLs. These assets are passed to a remote model and could potentially contain adversarial content intended to influence the agent or the underlying model's behavior.
Ingestion points: The image_url, video_url, and audio_url fields in the input schema for the bytedance/seedance-v2/pro model.
Boundary markers: The skill utilizes JSON-structured input when calling the local CLI, providing a degree of separation between data and command structure.
Capability inventory: The skill uses the runcomfy CLI to submit generation requests and downloads results to a specified local directory.
Sanitization: Documentation states that the CLI does not perform shell expansion on the prompt, but there is no mention of sanitizing or validating the content of the external media files.
[EXTERNAL_DOWNLOADS]: The skill instructions direct users to install the @runcomfy/cli package from NPM and to add the skill from the agentspace-so/runcomfy-skills repository on GitHub.
[COMMAND_EXECUTION]: The skill relies on executing the runcomfy command-line tool locally to interface with the model API and manage file downloads.

seedance-v2