orchestrating-restoreassist-video-series

name

orchestrating-restoreassist-video-series

description

Use when producing, scripting, or shipping the RestoreAssist onboarding + walkthrough video SERIES as a batch (not one clip) — the setup-wizard / getting-started tutorial / onboarding-welcome videos need scripts written, narrated, rendered, captioned, and wired into the live app for distribution; when picking up where a prior session left off across many slugs; or when a polished walkthrough must land on YouTube and resolve in <VideoExplainer>.

Orchestrating the RestoreAssist Video Series

Overview

This is the conductor over the whole onboarding + walkthrough video series. Each individual video is produced with the per-video pipeline; this skill sequences the set — tracking every slug's state, driving the next ones to "done", and landing them in the live app on YouTube unlisted embeds for distribution.

Core principle: the series advances one status-derived wave at a time. You never start from zero — you read the repo to see what each video already has, then do only the missing steps. A video is "done" only when its slug plays in the running app, not when a file exists.

REQUIRED SUB-SKILL: Use producing-restoreassist-videos for the mechanics of any single video (the five artifacts, ElevenLabs voice, render-all wiring, caption gotchas). This skill does NOT repeat that — it orchestrates many runs of it.

The series (this is "all these videos")

Defined by registry slug → Remotion composition id → file stem. Stems are the MP3/MP4/VTT/public_id filename; slugs key the registry and analytics. Captions are the trap: getCaptionUrl(slug) looks up CAPTION_REGISTRY[slug] by the slug, but the file on disk is stem-named — and the registry is currently keyed inconsistently (onboarding by slug, wizards by stem). See the caption-key note below.

Wave	Slug	Composition id	File stem	Live surface
1 Onboarding	remotion-onboarding-welcome	OnboardingWelcome	onboarding-welcome	top of /setup
2 Setup wizard	setup-wizard-signin	WizardSignin	wizard-signin	setup wizard step
2	setup-wizard-signup	WizardSignup	wizard-signup	setup wizard step
2	setup-wizard-setup	WizardSetup	wizard-setup	setup wizard step
2	setup-wizard-dashboard	WizardDashboard	wizard-dashboard	setup wizard step
2	setup-wizard-integrations	WizardIntegrations	wizard-integrations	setup wizard step
2	setup-wizard-health	WizardHealth	wizard-health	setup wizard step
3 Tutorials	remotion-tutorial-login	TutorialLogin	tutorial-login	/dashboard/learn
3	remotion-tutorial-signup	TutorialSignup	tutorial-signup	/dashboard/learn
3	remotion-tutorial-setup-wizard	TutorialSetupWizard	tutorial-setup-wizard	/dashboard/learn
3	remotion-tutorial-dashboard	TutorialDashboard	tutorial-dashboard	/dashboard/learn
3	remotion-tutorial-inspections	TutorialInspections	tutorial-inspections	/dashboard/learn
3	remotion-tutorial-reports	TutorialReports	tutorial-reports	/dashboard/learn
3	remotion-tutorial-billing	TutorialBilling	tutorial-billing	/dashboard/learn
3	remotion-tutorial-team	TutorialTeam	tutorial-team	/dashboard/learn
3	remotion-tutorial-compliance	TutorialCompliance	tutorial-compliance	/dashboard/learn
3	remotion-tutorial-integrations	TutorialIntegrations	tutorial-integrations	/dashboard/learn
4 Help (note)	help-inspections … help-compliance (6)	— none —	help-*	/dashboard/help/*

Wave 4 has no Remotion compositions — the help-* entries are CDN screen-capture clips, not part of the Remotion pipeline. Do NOT invent compositions for them; flag them as a separate track and stop at Wave 3 unless asked. Waves 1–3 = 17 Remotion videos, the true scope.

Per-video status — derive it, don't guess

For each stem, the video has six gates. Read the repo to mark each ✅/❌ before doing any work:

Gate	Evidence in repo
1 Scripted (SSOT)	content/videos/<stem>.script.json exists, timed to the composition
2 Composition has <Audio>	remotion/compositions/<file>.tsx references staticFile('narration/<stem>.mp3'). Most series compositions import only AbsoluteFill, Sequence — they have NO Audio/staticFile yet. Copy the exact pattern from remotion/compositions/onboarding-welcome.tsx (import {Audio, staticFile} … + <Audio src={staticFile('narration/<stem>.mp3')} />).
3 In render list	<id> present in compositionsToRender in remotion/render-all.ts (all 17 already are)
4 Narrated	public/narration/<stem>.mp3 exists (≥10 KB). public/narration/ does not exist yet — it's created by the narration handoff step.
5 Captioned	VTT file at public/videos/captions/<stem>.vtt AND getCaptionUrl(slug) resolves it — i.e. caption-registry.ts has an entry keyed by the registry slug. A present VTT is NOT enough. (See caption-key note.)
6 Distributed	registry entry resolves to a working YouTube embed in the live app

Caption-key note (verified gotcha)

getCaptionUrl(slug) returns CAPTION_REGISTRY[slug] || null — keyed by the registry slug, not the stem.

• remotion-onboarding-welcome is keyed correctly (slug → onboarding-welcome.vtt). ✅
• The six setup-wizard-* slugs are missing from CAPTION_REGISTRY — only their stem keys exist ("wizard-signin", not "setup-wizard-signin"). So getCaptionUrl("setup-wizard-signin") returns null and the VTT, though committed, never loads. ❌

So gate 5 for a wizard video requires adding a slug-keyed entry, e.g. "setup-wizard-signin": "/videos/captions/wizard-signin.vtt". Mark gate 5 ✅ only after the slug (not the stem) resolves.

Fast status sweep:

for s in onboarding-welcome wizard-signin tutorial-login …; do
  printf "%-22s script:%s mp3:%s vtt:%s\n" "$s" \
    "$([ -f content/videos/$s.script.json ] && echo y || echo n)" \
    "$([ -f public/narration/$s.mp3 ] && echo y || echo n)" \
    "$([ -f public/videos/captions/$s.vtt ] && echo y || echo n)"
done

What you do vs what you hand off

Plan-and-handoff — credential/GPU steps cannot run in the sandbox. Do everything sandbox-safe, then emit ONE run-list for a human/dev host.

You do (sandbox-safe)	Hand off (real env)
Write each *.script.json (scripting)	ElevenLabs narration MP3 (ELEVENLABS_API_KEY)
Add <Audio>, register in render-all.ts	pnpm run render:tutorials (headless Chromium)
Generate VTT captions (pure) + registry	YouTube unlisted upload (manual — OAuth)
Patch video-registry.ts once youtubeId returns	—

Maintain a living tracker at docs/superpowers/plans/<date>-video-series-status.md (the existing …-onboarding-welcome-video-production.md is the worked template). Update it every wave so the next session resumes instantly.

Distribution = YouTube (the precedence trap)

VideoExplainer resolves sources in strict order (VideoExplainer.tsx): cloudinaryUrl → localPath → youtubeId. Almost every series entry ships with cloudinaryUrl AND localPath already set, so adding youtubeId does nothing — the iframe branch is never reached.

To flip a slug to YouTube, the registry entry must become youtubeId-only:

"remotion-tutorial-login": {
  youtubeId: "ABC123xyz",            // unlisted id returned by the human upload
  title: "Signing in to RestoreAssist",
  durationSec: 45,
  category: "getting-started",
  // cloudinaryUrl + localPath REMOVED — else they win and YouTube never shows
},

scripts/video-upload.ts --host youtube is an OAuth placeholder — it does not upload. The real upload is a human step; you only wire the returned id. To patch the registry deterministically once the id is back:

npx tsx scripts/video-upload.ts --slug <slug> --youtube-id <id> --title "<title>"

Then confirm the entry is youtubeId-only (no cloudinaryUrl/localPath left). And add the slug-keyed caption entry (caption-key note) in the same change.

Orchestration loop

1. Build/refresh the status table for the wave (sweep above).
2. For each video missing gates 1–3 & 5: do the sandbox-safe work via producing-restoreassist-videos.
3. Emit the consolidated handoff run-list (narration → render → "upload these N MP4s to YouTube unlisted, return ids").
4. When ids return: patch the registry to youtubeId-only, run pnpm tsc --noEmit components/setup/video-registry.ts style check, commit.
5. Verify in the live app before claiming the wave done (gate below).
6. Update the tracker; advance to the next wave.

Verification gate (do not skip — required by repo rules)

A green render and a registry entry are NOT proof. For each shipped slug:

1. Open its live surface (e.g. /setup, /dashboard/learn) as the relevant user state.
2. The embed plays, narration is audible (one voice), CC track works, duration badge matches.
3. Network panel: no 404 for the YouTube thumbnail/embed; no VideoExplainer console error.
4. Provide the per-slug checklist: where to check, how to get there, what to see, what NOT to see (dead player / silent audio / missing CC), confirmation prompt.

Common mistakes

• youtubeId added but cloudinaryUrl/localPath left in place → YouTube never renders (precedence). Remove the other two.
• Captions keyed by stem, not slug → getCaptionUrl(slug) returns null, CC silently absent. The six setup-wizard-* slugs are missing from CAPTION_REGISTRY today; add slug-keyed entries. A committed VTT alone does NOT mean captioned.
• Treating help-* as Remotion → no composition exists; that's a different track.
• Scripting drift → timings in *.script.json must match the composition's scene timing, or captions/audio desync. One transcript is SSOT.
• Claiming the series done after rendering → render ≠ distributed ≠ verified. Only the live-app check closes a video.
• Re-deriving the per-video pipeline here → use producing-restoreassist-videos; keep this skill about sequencing.
• Losing state between sessions → if you didn't update the tracker doc, the next session redoes work. Update it every wave.