The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted user input (prompts and image URLs) and interpolates them into shell command templates. This provides an attack surface where maliciously crafted user input could attempt to influence the agent's execution or potentially bypass intended constraints during command assembly.
Ingestion points: prompt and images array in SKILL.md used for text-to-image and edit endpoints.
Boundary markers: None identified in the provided command-line templates.
Capability inventory: Local subprocess execution (runcomfy CLI), file system access (output directory writes), and network operations via the CLI.
Sanitization: No explicit sanitization or validation of the user-provided prompt or image URLs is present in the instructions.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @runcomfy/cli package from the NPM registry to function. It also fetches image generation results from *.runcomfy.net and *.runcomfy.com domains.
[COMMAND_EXECUTION]: The core functionality of the skill relies on executing the runcomfy CLI tool locally. It provides templates for running model tasks, managing logins, and downloading files, which are executed as shell commands by the agent.

gpt-image-2