research-and-capturelisted

# Research and capture Turn a topic into *evidence on screen*: the real announcement page, and footage that shows what the narration is describing. Without this a video is text slides read aloud. > Antigravity / Gemini: same steps, run the commands directly. ## 1. Research the topic first `WebSearch` for the primary source, then `WebFetch` it. Prefer the announcement, the paper, the changelog — not a blog summarising them. Pull out **concrete, quotable facts**: dates, figures, prices, names. Those become the script; everything else is filler. For deep academic or technical topics: - Pair with **`academic-deep-research`** to explore scientific literature, verify methodology, and extract empirical datasets without hallucinations. - Pair with **`academic-writing-flow`** to run evidence verification contracts, cross-checking every claim against peer-reviewed citations before scripting. Note the 2–4 URLs worth *showing*. A page is worth showing when the narration makes a claim a viewer might doubt. ## 2. Capture the evidence ```bash node scripts/media/screenshot.mjs \ --url "https://www.example.com/announcement" \ --out brain/<slug>/shots/announce.png --width 1280 --height 860 --wait 8000 ``` You don't need to run this by hand for a scene — a `media` block with `kind: "screenshot"` makes the render step do it. Run it manually when you want to *look* at the result first, which is worth doing for any page you haven't captured before. **Screenshots are evidence, not dec