Video Content Script Generator

Problem Solved

In short-video pipelines, topic data and script production are often mixed together: upstream provides a few hot_topics_*.json files, while downstream needs publishable, shootable content. This skill confines its role to consuming existing hot topics and generating scripts. It does not scrape pages, handle logins or anti-bot flows, or maintain data sources.

How It Works

It reads fields such as topic_id, title, region, and video_brief, then extracts core facts, timelines, impact analysis, editorial angle, and visual elements. It selects style templates and platform strategies by topic type, then builds a six-shot structure: hook, setup, core information, depth, twist, and CTA. The output includes Chinese and English voiceovers, subtitles, visual directions, IP character actions, sound effects, and platform-specific titles, descriptions, hashtags, cover text, and pinned comments. Optional search enrichment, reference-image supplementation, and LLM enhancement can improve output, but failures fall back to templates and mark llm_skipped instead of blocking.

Boundaries

The skill does not produce placeholder-only drafts, and it must not invent facts, data, or quotes. Missing information should be marked for human follow-up. It is not a topic-discovery tool, so it is not suitable when hot topics must be collected first. Downstream video agents can consume video_assembly_brief.json and platform_variants.json for assembly.