Yangguan AI News Video

扬关 AI热点资讯视频

Use this skill to produce a verified, sober, data-driven AI news video from recent public sources. The default output project is:

D:\codex_HyperFrames\yangguan-ai-news-video-template

Core Workflow

1. Clarify scope only if missing. Determine time range, video type, and target length. Defaults: recent week, weekly, 7-8 items, 16:9, 1920x1080, 110-120 seconds.
2. Research with browsing. Search domestic and international AI news within the requested range. Use official sources, government/agency sites, company blogs, reputable media, research institutions, papers, or standards bodies.
3. Return candidates first unless the user explicitly says to proceed without confirmation. Candidate lists should include title, source, date, link, summary, why it matters, suitability, and verification status.
4. After approval or explicit go-ahead, back up data. Copy data/current.json to data/archive/YYYY-MM-DD-current-backup.json.
5. Write data/current.json. Use the schema in references/data-schema.md. Mark all items 待核验 unless the user explicitly confirms they are verified.
6. Design content-specific backgrounds when visual quality matters. For polished videos, create one 16:9 background per news item, plus cover/summary/closing backgrounds as needed. Keep all Chinese text editable in HTML/CSS; never bake Chinese text into generated images. See references/content-driven-backgrounds.md.
7. Check and preview. Run npm run check, then start or reuse npm run dev -- --port 3019. Provide the Studio URL.
8. Render only after audit unless the user explicitly asks for a sample or final render. Require source/date/content review before formal MP4 export.

Research Rules

• Always browse for recent AI news; do not rely on memory.
• Cover both China and international news unless the user requests otherwise.
• Prefer 3-4 domestic items and 3-4 international items for an 8-item extended weekly video.
• Include at least one governance/safety item, one industry application item, and one infrastructure/model capability item when available.
• Do not invent news, links, dates, sources, companies, data, or quotations.
• Avoid vague subjects: write dated, attributed sentences such as "5月22日，国家发展改革委新闻发言人李超表示..." instead of "相关方面表示..." when the source supports it.

Writing Rules

• Each item must answer "what happened" without needing narration.
• Use dated, subject-led news sentences: "5月26日，国家能源局在深圳召开..."
• Avoid AI-ish filler such as "公开报道显示" unless no better attribution exists.
• Keep tone sober, institutional, clear, and non-promotional.
• Keep screen_text as a signal sentence and fact as a concrete dated event sentence.
• See references/content-rules.md for compression and audit rules.

Video Defaults

• Visual style: dark, restrained technology, radar/signal language, glass panels, cyan highlights.
• Prefer the content-driven background workflow for polished videos: the news content decides the background scene, and the video layout overlays editable Chinese text.
• Background images must contain no readable Chinese text, no logos, no fake UI labels, no official emblems, and no uniforms unless the user supplies authoritative assets.
• Cover: only "AI热点观察" and date; avoid repeating "扬关印记".
• Weekly extended flow: cover -> news items -> signal summary -> closing.
• Skip standalone "本期雷达" pages for extended weekly videos.
• News page priority: large headline and event explanation; source/date should be a small information row, not a large top bar.
• See references/video-style.md for layout decisions.

Useful References

• references/workflow.md: detailed end-to-end procedure.
• references/data-schema.md: JSON structure and examples.
• references/content-rules.md: source, safety, and writing constraints.
• references/video-style.md: visual and timing rules.
• references/content-driven-backgrounds.md: successful workflow for generating news-specific backgrounds and wiring them into the video.

Useful Script

Run this after editing data/current.json:

python C:\Users\admin\.codex\skills\yangguan-ai-news-video\scripts\validate_current_json.py D:\codex_HyperFrames\yangguan-ai-news-video-template\data\current.json

The script checks required fields, rough text length limits, item counts, and verification status.