The Agent Skills Directory

Prompt Injection (LOW): The skill uses authoritative language such as 'MANDATORY' and 'CRITICAL REQUIREMENT' to ensure the agent creates sidecar metadata files for every download. This is a functional requirement for license compliance and attribution tracking rather than an attempt to bypass safety protocols or override core agent behavior.
External Downloads (LOW): The skill utilizes curl to download image and video files from Pexels servers (images.pexels.com, videos.pexels.com). These are binary media assets and the downloads are targeted at a legitimate, well-known media provider.
Indirect Prompt Injection (LOW): The skill processes JSON data from the Pexels API and uses it to construct local metadata files.
Ingestion points: JSON responses from api.pexels.com containing photographer names, alt text, and URLs.
Boundary markers: Absent in the processing workflow.
Capability inventory: File writing via shell redirection (>) and network access via curl.
Sanitization: The workflow uses jq to build the metadata JSON, which provides inherent escaping for string values, mitigating the risk of basic injection into the resulting file structure.
Command Execution (LOW): The skill provides bash snippets for API interaction. These commands use standard environment variables for authentication and do not exhibit dangerous patterns like piping remote content directly into a shell.

pexels-media