What problem does it solve? Calling hosted generative AI models usually requires writing custom API clients, handling authentication, polling job status, and downloading outputs by hand. This Skill teaches the agent to do all of that through the single runcomfy CLI, giving one install and one login for hundreds of image, video, lip-sync, face-swap, and LoRA-training endpoints. ## Core Features & Use Cases - Unified model invocation: Submit any RunComfy model with runcomfy run <vendor>/<model>/<endpoint> --input '{...}', then poll status and download results automatically. - Scriptable automation: Use --output json, --no-wait, and runcomfy status to batch prompts, submit jobs without blocking, and pipe results into jq or shell loops. - Predictable error handling: Documented exit codes (64, 65, 69, 75, 77, 130) distinguish bad input, retryable timeouts/429s, and auth failures so agents can retry or re-authenticate correctly. - Use Case: A user asks to generate 20 product images from a prompts file. The agent loops over the file, calls runcomfy run blackforestlabs/flux-2-klein/9b/text-to-image per prompt, and downloads each result into its own output directory. ## Quick Start Install the CLI with npm, sign in using runcomfy login, then ask the agent to run a RunComfy model such as openai/gpt-image-2/text-to-image with your prompt.
