Luxury Tech

Rules: ${CLAUDE_PLUGIN_ROOT}/references/rules.md · CLI: ${CLAUDE_PLUGIN_ROOT}/references/cli.md · Design system (read it before authoring graphics): ${CLAUDE_PLUGIN_ROOT}/presets/luxury-tech/DESIGN.md. ss = "${CLAUDE_PLUGIN_ROOT}/bin/shorts-studio".

Input: $ARGUMENTS

0. Engine check

The commands below run the Shorts Studio engine. Installed as the Claude Code plugin it is already present and shorts-studio is on the Bash PATH. If shorts-studio doctor is not found (for example, this skill was added with npx skills add), ask the user before installing anything, then:

• Claude Code: /plugin marketplace add bloxy-studios/shorts-studio → /plugin install shorts-studio@shorts-studio.
• Other agents: git clone --depth 1 https://github.com/bloxy-studios/shorts-studio ~/.shorts-studio/engine and read every ${CLAUDE_PLUGIN_ROOT} path below as ~/.shorts-studio/engine.

Then run shorts-studio doctor --setup once (project-local runtime; never sudo or global installs).

1. Start from a timed edit

Luxury Tech sits on top of an edit with final word timings:

• Creator recording (camera + clean mic): ss creator <project> --camera FILE --mic FILE --preset luxury-tech (sync, takes and assembly run first; the Luxury layout is applied automatically).
• Existing clip: ss plan <project> --clip <id> ... then continue below.

Inspect analysis/<camera>/sheets/ov*.jpg. Record where the presenter looks away from the lens: ss analyze note <project> --kind gaze --at 00:12.4 --end 00:14.0 --detail "looks away to the left, reading" --confidence 0.8. (Head-pose CV is evidence, not measured eye gaze; your observations override it.)

2. Write the director graphics brief (source-specific, word-anchored)

ss luxury brief <project> --clip <id> writes clips/<id>/luxury/graphics.json listing every line's words, the item kinds and icons. Fill it in:

• intents: per line — hook (→ split), opinion (→ full camera when looking at the lens), screen (→ full visual), card/list (→ split), point, outro/cta (→ full camera).
• title: a truthful cover title from the hook if the user gave none (≤ 2 lines at ~110 px).
• cta: only if the speech or brief contains one: {"keyword": "<real word>", "requestedBy": "speech"}.
• runs[]: one spec per graphics run, keyed by lines (+ part when a line was split): one dominant subject + at most one supporting cue; items in world px (split content zone y 40–900, full visual y 120–1230); camera moves anchored to the naming word ({"kind":"step","to":"car","when":{"word":"cars"}} — step starts 0.25 s before the word, whip midpoint lands on the word, ≤ 1 whip per run, push ≤ 1.6×). Animate the action being explained (typing, click, answer, check); never a word row that repeats the narration.
• Facts/claims: use a real capture (image item, sources: [{url, kind: "capture", disclosed}]); label reconstructions; never fabricate an official screenshot or capability.

Then: ss luxury plan <project> --clip <id> → prints the layout-runs table (run, time, layout, source/take, words, graphic subject, caption zone, events) and saves a new plan version. Fix every warning (items outside the zone, too many subjects, unknown words) and re-plan.

3. Preview before rendering

ss luxury preview <project> --clip <id> snapshots each graphics run (landing frame, after each move/event, last frame). Read every contact sheet. Check: sharp composed landing frames, no overlap with the caption zone, hierarchy, nothing below 30 px. Adjust the brief and re-plan.

4. Render, inspect, repair

ss render <project> --clip <id> (graphics, head mattes for split pieces, captions, synth audio, 60 fps encode, measured QA). Read qa/contact.jpg and qa/cuts.jpg, and pull full-resolution frames of each split run (ffmpeg -ss <t> -i <video> -frames:v 1 f.png) to inspect the head/card join. Record what you saw with ss qa note. Then ss repair <project> --clip <id> for any fixable QA finding (bounded; see the director brief's stop criteria).

Error handling and limits

• Presenter never looks at the lens → camera layouts are withheld and logged; full visuals carry the edit.
• Matte unstable/misaligned (QA IoU/alignment probe) → that run falls back to a contained card; reported.
• Background removal is ~1.6 s/frame on CPU here: only split-run source frames are matted, and cached.
• 30 fps camera footage is repeated to 60 fps (no new motion detail); graphics are genuinely 60 fps.
• CTA keyword not in the speech with requestedBy: "speech" → planning fails; never invent a CTA.

Example

"Use Luxury Tech for this one and do the whole edit yourself" (camera.mp4 + mic.wav): run ss creator … --preset luxury-tech, inspect sheets, write the brief (hook card, radar → systems → car whip, checklist, car/plane tiles), ss luxury plan, ss luxury preview, read sheets, ss render, inspect, ss repair, then ss export.

Completion check

Done only when: the layout table uses all three layouts (or logs why not); QA shows luxury.static-footage, luxury.landing-sharp, luxury.highlights, luxury.sfx-rules passing and audio.true-peak passing; you inspected real frames of every split run and recorded it (ss qa note); the report lists every exception and fallback.