The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation explicitly references the storage location of authentication tokens at ~/.config/runcomfy/token.json. While this is standard configuration for the integrated CLI tool and is described alongside security recommendations (mode 0600), it identifies a sensitive path on the user's filesystem.
[PROMPT_INJECTION]: The skill handles untrusted external data in the form of image and mask URLs, which creates a surface for indirect prompt injection targeting the underlying image models.
Ingestion points: The skill accepts image_urls, images, image, and mask_image parameters across multiple routes, which are fetched by the remote model server.
Boundary markers: The skill documentation notes that user prompts are encapsulated within JSON strings before being passed to the CLI to prevent shell-based injection attacks.
Capability inventory: The skill utilizes a local CLI to communicate with a remote API and downloads generated image files to the local file system.
Sanitization: The CLI implementation includes a 2 GiB file size limit for downloads to prevent potential disk-exhaustion attacks from malicious or runaway model outputs.

image-edit