The Agent Skills Directory

[SAFE]: The skill utilizes a restricted execution environment by specifying allowed-tools: Bash(runcomfy *), which prevents the AI agent from executing unauthorized shell commands outside of the specific vendor CLI.
[EXTERNAL_DOWNLOADS]: The skill references the installation of the @runcomfy/cli package via NPM. This is a standard and expected download from the official package registry for the vendor's own tools.
[PROMPT_INJECTION]: The documentation explicitly acknowledges the potential for indirect prompt injection from untrusted image data or metadata.
Ingestion points: The image, mask_image, and prompt fields in the JSON input for the runcomfy run command.
Boundary markers: Inputs are structured within a JSON object, providing architectural separation from the command execution context.
Capability inventory: The skill is strictly limited to image processing and generation tasks via the restricted runcomfy command set.
Sanitization: The documentation notes that the CLI does not perform shell expansion on input content, reducing the risk of command injection, and advises processing only user-provided URLs.

image-inpainting