The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data through multiple image ingestion parameters, creating a surface for indirect prompt injection if images contain malicious text or metadata.
Ingestion points: The skill accepts external content via image_url, image_base64, overlay_url, and overlay_base64 parameters across various actions (e.g., blur, composite, draw) as defined in SKILL.md and schema.md.
Boundary markers: There are no instructions or delimiters provided to the agent to treat image content as untrusted or to ignore instructions found within processed images.
Capability inventory: The skill provides significant image modification capabilities, including drawing text, shapes, and composite layering, which could be exploited in a prompt injection scenario.
Sanitization: The documentation does not describe any sanitization or validation of the content or metadata of the processed images to prevent injection attacks.

image-editor