The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external media assets, such as images, audio files, and videos, which are fetched from remote URLs. This introduces a risk where instructions embedded within these third-party files (e.g., in metadata, visual text, or audio signals) could influence the agent's behavior.
Ingestion points: The skill accepts image_url, audio_url, and video_url as inputs for various video generation and extension tasks as defined in SKILL.md.
Boundary markers: While the skill uses JSON formatting for CLI inputs to mitigate shell injection, there are no specific prompt delimiters or instructions provided to the agent to ignore content-based directives within these media files.
Capability inventory: The skill has permission to execute the runcomfy CLI tool and perform file writes to the local ./out directory.
Sanitization: The documentation notes the untrusted nature of these URLs but relies on the agent's internal reasoning rather than technical sanitization or validation of the media content.
[EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to install the @runcomfy/cli package globally via NPM, which is a necessary dependency for the skill's functionality.
[COMMAND_EXECUTION]: The skill performs shell command execution by invoking the runcomfy CLI tool to interact with remote video generation models.

ai-video-generation