The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the use of the runcomfy CLI tool to execute image editing tasks. It provides templates for passing instructions and external image URLs to the service via command-line arguments.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes remote image content provided by users, which can contain adversarial instructions targeting the AI model.
Ingestion points: The skill accepts external HTTPS URLs in fields such as image_urls, images, image, and mask_image across its various routing options in SKILL.md.
Boundary markers: While the skill uses structured JSON to organize CLI inputs, the models themselves process the unstructured pixel data of the images where injections may reside.
Capability inventory: The skill executes the runcomfy CLI and performs file writing to a user-specified local directory.
Sanitization: The skill documentation acknowledges the risk of image-based prompt injection but does not provide mechanisms for sanitizing the visual content of the user-provided images.

image-edit