Routes video region edits through RunComfy's prompt-driven endpoints, defaulting to Wan 2-7 edit-video for spatial language tasks like "remove the watermark in the bottom-right" or "clean up the passing person in the background." The skill picks between three models based on intent: Wan for prompt-driven regions, Lucy Edit for identity-stable swaps, and Seedream for frame-by-frame stacks. It's honest about the tradeoff: these are all prompt-based, not pixel-precise mask propagation. For surgical edits you need the ComfyUI workflows with SAM2 tracking, which the CLI can't reach. Good default for conversational video cleanup, just don't expect pixel-perfect masks from text alone.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill video-inpainting