rebuild-site

name	rebuild-site
description	Rebuild a whole landing page / multi-section website from a screen recording that scrolls or walks through it. Use when the user has a recording of an ENTIRE page (not a single animation) and wants the full page recreated — layout, sections, copy, design system, and scroll animations. Orchestrates analyze (measured timing + section structure + curated frames) + asset generation + a full multi-section build. For one isolated animation, use /motiscope:recreate instead.
argument-hint	[path-to-video] [gsap|css|framer]
allowed-tools	Bash, Read, Write, Edit, AskUserQuestion
user-invocable	true

motiscope: rebuild-site

Rebuild a whole landing page from a walkthrough recording. This is analyze + recreate applied to an entire page, and it leans on the same division of labor:

• The numbers give you the WHEN + the structure — motiscope measures section timing, transitions, and per-section entrance easing, and curates the frames that show each section.
• You give the WHAT — from the frames you reconstruct the sections, layout, copy, design system, and what kind of animation each uses (open vocabulary — scroll reveals, pins, masks, draws, whatever you see).
• Imagen fills the photos — generate the photographic assets the design needs.

It's a team effort: measured timing + curated frames + your vision + generated assets → a full page. Be honest about which is which in the final summary.

Resolve scripts + preflight

SCRIPTS="${CLAUDE_PLUGIN_ROOT:-}/scripts"; REFS="${CLAUDE_PLUGIN_ROOT:-}/references"
[ -f "$SCRIPTS/ingest.py" ] || { SCRIPTS="<abs dir of this SKILL.md>/../../scripts"; REFS="<...>/../../references"; }

(Windows: python.)

Preflight

python3 "$SCRIPTS/mvsetup.py" --check   # exit 0 = ready; else hand to /motiscope:doctor

Resolve the video from $1, else scan animations/ (as in analyze).

Step 1 — analyze the whole walkthrough

python3 "$SCRIPTS/ingest.py" "<video>" --preset landing

landing gives readable 1280px frames and auto-decomposes onto the motion. The result:

• hold segments = the sections the walkthrough paused on.
• fade-out/cut segments = the transitions between sections.
• per-section entrance timing/easing (the move beats + their cubic-bezier), the stagger timing, and curated frames covering each section.

For a long clip, note it may exceed one pass; you can also re-run focused on a range (--start/--end) to get denser frames for a specific section.

Step 2 — build the site plan (from the frames)

Read report.md (the measured timeline) and every curated frame (parallel Reads). Then reconstruct — this is your vision doing the work — and present a site plan to the user:

1. Design system: palette (hexes), typography (name the actual fonts if you can — e.g. Martian Mono), spacing/scale, brand/logo, light/dark.
2. Sections in order: for each — a name (hero / features / how-it-works / pricing / FAQ / footer / …), its layout, its copy transcribed verbatim from the frames, and the imagery it contains.
3. Motion per section: what animates and how (scroll-reveal, pin+scrub, stagger, mask, draw, parallax — name what you see), wired to the measured timing (durations, bezier, stagger each_ms) from the segment table. Landing walkthroughs are almost always scroll-driven — assume ScrollTrigger unless the frames say otherwise.

Show the plan and let the user correct it before building.

Step 3 — assets

List the photographic assets the design needs (hero shots, feature photos, backgrounds — anything that isn't UI you can build in CSS/SVG). Then ask the user (AskUserQuestion): (a) point at their own files, (b) generate, or (c) placeholders. For (b), generate with the real provider:

python3 "$SCRIPTS/assetgen.py" --check   # confirm a key (gemini) is set
python3 "$SCRIPTS/assetgen.py" generate --type image --provider gemini \
  --prompt "<match the shot + the design's grade>" --out assets/hero.png --aspect-ratio 16:9
# then shrink for the web: ffmpeg -i assets/hero.png -vf scale=1200:-2 -q:v 5 assets/hero.jpg

Match each prompt to what the frame shows and the design's color grade. If no key, hand to /motiscope:doctor. Never write keys into output or commit them.

Step 4 — build the page

Default target GSAP + ScrollTrigger ($2 overrides: css | framer). Read "$REFS/easing-map.md" and "$REFS/targets/<target>.md".

For GSAP, defer to the official gsap- skills* (gsap-timeline, gsap-core, gsap-scrolltrigger for the scroll reveals/pins, gsap-react if the project is React, gsap-utils for stagger).

Build one standalone page (index.html + assets/) with:

• the design system (tokens for palette/type; embed or link the identified fonts),
• a nav + every section in order, copy verbatim, assets wired in,
• scroll-driven motion per the plan, using the measured durations / cubic-bezier / stagger — not guessed values,
• the section transitions you saw (fades, mask wipes, pinned zooms),
• responsive layout and a prefers-reduced-motion: reduce guard (elements resolve to final state).

Keep the timing measured; build the effects from what you saw.

Step 5 — output, preview, credit

Write to ./motiscope-output/site/. Then:

• tell the user how to preview (open motiscope-output/site/index.html),
• give an honest breakdown: timing = measured; layout/copy/design = reconstructed from frames; photos = generated (Imagen); fonts = identified. Flag what to fine-tune.
• Attribution & rights: if the source is someone else's design (a Dribbble shot, another company's site), add a prominent credit in the footer (link the original), frame it as a design-to-code study, and confirm the user has the right to publish before it goes public. Don't publish a clone of a third party's brand without their say-so.

Two worked examples:

• Cadence — an original page, walked through and measured. Every scroll boundary lands on the authored millisecond, and the two things motiscope gets wrong are published with the reason.
• Alterfx — a real Dribbble concept rebuilt from a 10s recording, credited on the page and framed as a design-to-code study. This is what the rights step below looks like when the answer is "yes, with attribution".

Rights come first. If the recording is of someone else's site or concept, say so before you build: offer to recreate it as a study with visible attribution, or to rebuild the structure and motion under the user's own art and copy. Do not publish a clone of someone else's design.