docs(product-launch-video): use real screenshots for site showcases

skills/product-launch-video/SKILL.md

@@ -54,6 +54,8 @@ Classify the input and choose the path. Explicit URL -> capture it and use the s
 
 Run capture with: `npx hyperframes capture "<URL>" -o ./capture`
 
+For a site tour or show-it-as-is brief, the captured page is the visual source of truth. Use the real screenshot instead of rebuilding the full website in HTML. If the shot needs internal movement, keep the screenshot as the base and overlay real captured assets at measured positions, or rebuild only the one component that moves. For a scroll shot, use a 2x full-page capture and animate the viewport over it. Recreate the whole page only when the user explicitly asks for a stylized interpretation or the capture is unusable.
+
 If `GEMINI_API_KEY`, `GOOGLE_API_KEY`, or an OpenRouter key exists, capture auto-captions assets into `capture/extracted/asset-descriptions.md`. This is not a review gate. Without a vision key, use DOM context and continue.
 
 No-capture path: create `capture/extracted/tokens.json`, `capture/extracted/visible-text.txt`, `capture/extracted/asset-descriptions.md`, and `capture/assets/` by hand. `tokens.json` should be `{ "title": "", "description": "", "colors": [], "fonts": [] }`; fill title/description from the brief when possible. `visible-text.txt` contains the full brief or script. `asset-descriptions.md` should say no assets were captured unless the user gave asset notes.

skills/product-launch-video/sub-agents/frame-worker.md

@@ -12,4 +12,6 @@
 
 **Place each candidate by its `roles`** (the `focal` is the hero): a `cutout` is a foreground subject — respect the 83% keep-out, lay text around it, not over its face; a `background` is full-bleed and dimmed ~30–50% so foreground content stays legible. Frame files are sub-compositions. Audio remains orchestrator-owned: never author `<audio>` in a frame. An approved `[video]` candidate may be declared as a frame-local `<video data-frame-video="approved" data-start="..." data-duration="..." data-track-index="...">`; `assemble-index.mjs` hoists it to the host root and translates its timing. Give every approved video explicit host geometry with numeric `data-frame-video-x`, `data-frame-video-y`, `data-frame-video-width`, and `data-frame-video-height`; optionally set `data-frame-video-fit="cover|contain|fill|none|scale-down"` (default `cover`). The assembler converts only those values to host CSS: frame-local classes and inline styles do not cross the hoist boundary. Do not use this declaration for audio, unapproved URLs, or videos without explicit timing and geometry. If no approved video is supplied, use an explicitly supplied static still/key art (`[video-still]` or another image candidate) as `<img>`; do not extract a frame, fabricate a URL, or silently embed an unapproved clip.
 
+When a frame showcases a captured website, use the supplied screenshot as the page. Do not rebuild the full site in HTML: even a close recreation can change the real layout, spacing, or branding. If the shot needs motion inside the page, overlay the supplied real assets at measured positions or rebuild only the moving component.
+
 Brand text comes from your frame's `scene` / narrative — never from `frame.md` (a style spec, not the product's content). Place the named assets, and never invent new ones.