If you're wiring Claude into AI media pipelines, this is the bridge. One CLI binary gives you command-line access to hundreds of RunComfy models: text-to-image, video generation, face swap, lip sync, ControlNet, upscaling, LoRA training, the works. Install once via npm, authenticate once with device-code flow, then script any endpoint with `runcomfy run <model> --input '{...}'`. It handles polling, streaming, and downloads automatically. The skill teaches Claude how to discover schemas, parse JSON output, and chain model calls. This is infrastructure, not a convenience wrapper. If you're building agents that need to generate or edit media without reinventing API plumbing for every new model, this is the foundation.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill runcomfy-cli