The input is a full novel, article or short story outline. The agent confirms visual style, runtime, aspect ratio, tone and the core hook, then writes a four-act script in Chinese screenplay format, conflict first, with a cliffhanger at the end of each 15-second episode. It numbers assets as character (C), scene (S) and prop (P) codes with image prompts for models such as GPT-Image-2, Seedream or Nano Banana Pro, then writes a Seedance 2.0 storyboard prompt per episode with a timeline, asset slots and an ending-frame note, chaining episodes through the video-extend feature. The skill does not generate media itself (the README suggests pairing it with the Jimeng CLI), and its instructions are in Chinese.