The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external media data which acts as a potential vector for indirect instructions.
Ingestion points: Processes external images and videos via URLs (prompt_image, video_uri) and local files (e.g., kettle-product.png) in SKILL.md.
Boundary markers: Absent. The instructions do not define specific delimiters for separating user-supplied prompts or media content from agent instructions.
Capability inventory: The skill has capabilities to write files to the local system (open("espresso.mp4", "wb")) and perform network operations to the Runway API.
Sanitization: Absent. No explicit sanitization or filtering of external content is implemented before interpolation into API payloads.
[EXTERNAL_DOWNLOADS]: Fetches resources and generated media from Runway's official developer service.
Evidence: Communicates with the official API endpoint https://api.dev.runwayml.com/v1 and downloads generated output from URLs returned by the service.
[COMMAND_EXECUTION]: Executes standard package manager commands to set up the development environment.
Evidence: Instructions include the execution of pip install runwayml and npm install @runwayml/sdk to install required dependencies.

runway-ml