Communitygithub.com

datassthou64-source/blue-motion-kit

Turn creator voiceover audio into researched, visual-first HyperFrames motion graphics. Use when anyone provides one or more audio files for a creator video and asks to transcribe, analyze, research the concepts or products online, build or revise pure motion graphics, preview locally, or render in the established continuous blue-gradient style.

blue-motion-kit란 무엇인가요?

blue-motion-kit is a Claude Code agent skill that turn creator voiceover audio into researched, visual-first HyperFrames motion graphics. Use when anyone provides one or more audio files for a creator video and asks to transcribe, analyze, research the concepts or products online, build or revise pure motion graphics, preview locally, or render in the established continuous blue-gradient style.

지원 대상✓Claude Code✓Codex CLI✓Cursor
npx skills add https://github.com/datassthou64-source/blue-motion-kit/tree/HEAD/.claude/skills/blue-motion

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

Blue Motion graphics

Build a factual visual explanation around the voiceover. Favor recognizable objects, product UI language, official logos, and transformations over captions or abstract text cards.

Load /motion-visual-sourcing first. It owns the two laws this style depends on — visual first, text second (delete every word; the idea must still land) and every named entity earns a real, sourced asset (fetch the true logo, real UI or a site capture for any company, product, site or repo the VO names — never a generic box or the name as plain type). This skill's blue gradient governs how it looks; those laws govern what goes on screen.

Where everything is

All paths below are relative to KIT, the root of the blue-motion-kit repo. bash <this skill folder>/scripts/kit-root.sh prints it (it resolves symlinks, so it is right whether this skill was loaded from the repo or through a link).

PathWhat
library/design.mdThe one visual spec — palette, type, objects, motion verbs, honesty rules, and the argument-shape → scene map. Wins any conflict. Read it before storyboarding.
library/scenes/13 config-driven templates from the approved Grok Bot build. Copy the .html + .motion.json into compositions/ and edit the CFG block at the top of its <script>; geometry regenerates from it.
library/patterns/The earlier static primitives. Still valid.
library/PATTERNS.mdIndex of both.
library/assets/logos/Frozen, verified official marks (Grok, Grok Bot, xAI, X, Cursor, Claude, Slack, Instantly, HeyReach, HubSpot, GitHub…). SOURCES.md is the provenance table. Check here before fetching.
reference/grok-bot/The approved reference build: scene source, contact sheets and two finished renders. Look, don't copy wholesale.
brand/creator.mdWho watches and how the creator talks.
KIT/editor.local.mdThis machine's editor preferences from the first-run interview (projects folder, NLE, approval gates). Read it if it exists.
references/build-recipe.md (this skill)The composition contract and the bugs that pass check clean. Read when authoring or debugging.
references/scene-patterns.md (this skill)Hand-built devices that are not templates.
scripts/ (this skill)bootstrap-project.sh (new project), render_passages.py (export), hlg_to_sdr.py (HDR footage fix), fetch-logo.py (logo resolver), kit-root.sh.

A template supplies the choreography, easing, palette and seek-safety. It supplies nothing about whether it is the right idea for this beat. Change the text and the metaphor to match what is actually said; re-research every newly named product and use its current official assets. Never ship a template's placeholder mark.

Workflow

  1. Read brand/creator.md, and editor.local.md if present.

  2. Use the mandatory HyperFrames, motion-graphics, media-use, and CLI skills.

  3. Inspect only the supplied audio. Ignore earlier work for that audio unless the editor explicitly asks to reuse it.

  4. Scaffold the project: bash SKILL_DIR/scripts/bootstrap-project.sh PROJECTS_DIR/<video-slug> AUDIO... (PROJECTS_DIR comes from editor.local.md; default KIT/projects). It copies the starter, gradient, logos and both template generations, and prints each audio file's duration.

  5. Transcribe every file with word timings (npx hyperframes transcribe assets/audio-01.<ext> --dir . or the current media-use path). Sanity-read it and correct obvious proper-name errors (Clay, n8n, Instantly, HeyReach, Prospeo, Apify, Grok, Cursor, Claude Code, Codex, MCP).

  6. For multiple files, preserve the originals and place them sequentially on one timeline. Use ffprobe for exact offsets.

  7. Turn the transcript into semantic beats: time range, meaning, visual subject, action, and research need.

  8. Research unfamiliar concepts and current product branding before drawing. Prefer official sources, save resolved assets locally, and record sources in RESEARCH.md. Logos: python3 SKILL_DIR/scripts/fetch-logo.py --manifest logos.txt assets/logos, then open every file it returns.

  9. Storyboard the full audio before authoring (STORYBOARD.md). Target one complete visual scene per 4–8 seconds. If editor.local.md asks for storyboard sign-off, stop here and show it.

  10. Put the blue-gradient video in the parent timeline, one unbroken clip per motion-graphics passage, and keep every scene transparent so the background never resets at cuts.

  11. Adapt the smallest semantically matching pattern. Do not force a topic into a template merely because it exists.

  12. Run npx hyperframes check, then snapshot the payoff frame of every scene and look at the images. Geometry bugs — an object landing at the wrong coordinates, one element sliding over another — pass check silently and only show up in a frame.

  13. Open the local preview (npm run dev) and give the editor the URL. Do not render until the editor explicitly approves (unless editor.local.md says auto-render).

  14. After approval, export with the default script. Do not run a full-timeline render.

    python3 SKILL_DIR/scripts/render_passages.py PROJECT_DIR
    

    It renders each motion-graphics passage to renders/MGNN-<label>.mp4. It converts any HLG footage to SDR first, and it checks every file for 1920×1080, 30 fps, frame count, audio and SDR BT.709. It then writes premiere-cuts.xml, which places each file at its real position on the VO timeline (imports into Premiere and DaVinci Resolve). Use --dry-run to list passages, --only 3,8 to re-render some, and --xml-only to rebuild the XML. Then look at a frame from each file; the script's checks cannot see a missing layer. Details and why: references/build-recipe.md §7.

  15. Finish with a short hand-off: project path, the renders, the XML, and any assets you could not verify.

SKILL_DIR is this skill's folder: KIT/.claude/skills/blue-motion.

Speed and token discipline

  • Reuse the exact gradient, style tokens, and verified scene primitives.
  • Research only nouns that affect factual visuals: products, logos, interfaces, workflows, or physical objects.
  • Prefer one authoritative source for a simple fact; add more only when ambiguous.
  • Keep BRIEF.md, RESEARCH.md, and STORYBOARD.md compact.
  • Reuse accurate geometry for laptops, repository windows, app tiles, and routes.
  • Show snapshots or the local preview instead of narrating CSS choices.

Non-negotiables

The look and honesty rules are library/design.md §9–10: no face cam, no captions or sentence cards, 1–3-word labels, real sourced marks on plates, and nothing fabricated. On top of those:

  • Laptop screens stay near 16:9 and settled devices never pulse or change dimensions.
  • Moving objects travel continuously. Do not pause halfway unless the narration describes a stop. Use smooth ping-pong motion for back-and-forth ideas.
  • Make each scene one dominant visual idea with a clear state change, using a tangible verb: split, copy, sync, pull, push, search, assemble, route, unlock, compare, transform.

Changing the kit itself

A new durable rule (a style call, a default asset, a new trap) goes into its one home — look in library/design.md, trap/procedure in this skill, creator facts in brand/creator.md — never into a project's notes or the agent's memory. Then commit it so every editor gets it (see CONTRIBUTING.md at the kit root).

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

관련 스킬