Video & Animation

>- Find what's really working in the user's niche on YouTube, rank it by how far each video beat its own channel, then name the formula. Use for "what's working right now", "find viral videos in my niche", "why did this blow up", competitor research, or a swipe file.

MUST be used when breaking down large text content into TTS-appropriate chunks. Intelligently splits stories, meditations, or long content while preserving emotion markup, maintaining narrative flow, and optimizing for Cartesia voice synthesis. Handles sentence boundaries, emotion tags, and chunk size limits. Triggers on: chunk text, split for TTS, break up text, TTS chunks, divide content, split story, chunk meditation, text chunking, prepare for voice, TTS processing, voice synthesis prep.

>- Write the description, chapters, tags and search-facing text for a YouTube video, aimed at what a real person would type. Use for "write my description", "add chapters", "timestamps", "tags", "SEO", "help this video get found", or "nobody can find this".

Produce a narrated, chaptered 1080p MP4 demo or walkthrough video of any browser-based product (web app, dashboard, internal tool, embedded artifact). It drives the real UI in Chrome with an on-screen cursor, highlight rings, notes and title cards, voices a scene script with TTS (Amazon Polly or macOS say), times every action to the narration, cuts loading waits, and mixes, normalises and chapters the result. Use it when asked to "create a demo video", "record a walkthrough", "make a screencast with voiceover", "product tour video" or to re-record or revoice an existing one.

Write the caption, hashtag set, and first comment for a reel or post. Use when the user says "caption", "hashtags", "write the description", "what do I write under this", "first comment", or "post copy".

Get the full text of a YouTube video (or any yt-dlp-supported video, or a local audio/video file) as a clean, proofread Markdown transcript with chapter headings. Downloads audio with yt-dlp, transcribes it locally on macOS with Spokenly (NVIDIA Parakeet TDT 0.6B v3), then polishes it with parallel editor subagents. Use this whenever the user shares a youtube.com / youtu.be link and wants what is said in it — a transcript, "текст из видео", "расшифровка", "транскрипция", subtitles as text, quotes — and as the first step when they want a summary, notes or answers about a video's content, even if they never say "transcript". Also for podcasts, lectures or meetings given as local audio/video files.

制作、续作或修复高质量角色替换 PV，保留原画风、分镜、关键姿势与节奏，处理逐稿生成、遮挡、字幕和连续性，并分别验证编码与美术质量。适用于低帧率手绘 PV 和角色重绘动画；不用于普通换脸或单张立绘。

>- Use when hosting, deploying, or operating an app on the Prisma Platform: projects, branches, preview environments, services and their versions, Prisma Postgres databases, object-store buckets, environment variables, custom domains, logs, or the GitHub integration. Triggers on "Prisma Platform", "Prisma Compute", "Prisma Postgres", "Prisma Storage", "preview environment", `prisma deploy`, `prisma dev`, `prisma auth`, `prisma project`, `prisma branch`, `prisma service`, `prisma postgres`, `prisma bucket`, `prisma git`, `prisma.config.ts`, promote, rollback.

Analyze authorized reference videos and turn their evidence-backed pacing, typography, semantic overlays, caption motion, product-proof structure, and retention devices into a production-ready fine-cut specification for English- or Spanish-language TikTok/Reels/Shorts edits. Use when the user asks for 精剪, reference-driven editing, product-demo ads, talking-head explainers, pros/cons lists, tier lists, animated captions, timed product cutouts, or a more sophisticated cut than basic trimming and subtitles. Pair with auto-video-editor for local renders and OpenChatCut for layered Remotion timelines. Do not use to copy protected creative expression shot-for-shot, fabricate product evidence, identify an exact burned-in font, or claim conversion performance.

Turn user-provided local video, image, and audio assets plus natural-language editing requirements into an isolated, auditable edit job, a validated JSON timeline, a rendered MP4, and a bounded local technical QA report. Use when the user asks Codex to automatically edit, trim, reframe, caption, sequence, score, or package uploaded media. Do not use for knowledge-base construction, social-platform publishing, or pure reference-video prompt writing.

>- Use when deploying or managing an app that uses Prisma Composer (`@prisma/composer`): wiring its services and Modules, running it locally, testing composed services, or standing up / tearing down an environment. Triggers on "prisma composer", "@prisma/composer", "prisma app", `prisma deploy`, `prisma dev`, `compute()`, `module()`, `contract()`, `service.load()`, `mockService`, `bootstrapService`.

Scaffold a motion-studio film project (seek(t) renderer, springs, beat grid, synthesized sound, critique tools), install it, set up the browser, run doctor. Use for 'set up the motion studio', 'new film project', 'init studio'.

Make motion feel physical with closed-form springs from lib/motion.js: presets, springFromFeel, track() for multi-target values, loopTrack loops, indicators, text swaps. Use for 'feels linear', 'add bounce', 'easing to springs'.

The viral one-line 'showreel for a résumé' prompt: why it works (anatomy), the credited quote, and seeded variants that avoid look-alike reels; runs it as an engine test. Use for that prompt, its variations, or a showreel.

Ship every format from one timeline, then package it: layout() reframing (never crop), render all formats, per-format checks, deliver.mjs, a brand skill, a service offer. Use for 'export 9:16 and 16:9', 'deliver', 'make a skill'.

Build, debug or speed up the seek(t) render engine: render contract, canvas vs page capture, workers/subframes/chunks, determinism fixes, Remotion or HyperFrames handoff. Use for 'render is slow', 'frames differ', 'use Remotion'.

Turn a reference (a frame, a video or an image library) into docs/style_guide.md and a beat-grid shot list: measure cuts, palette and pace, take grammar not content, wait for OK. Use when the user supplies or names a look.

Make Claude watch its own frames: critique.mjs evidence, harsh 7-axis scores, timestamped P0/P1/P2 problems in docs/review_log.md, fix the 3 worst, re-check, pass the gate. Use for 'critique it', 'review the frames', 'is it good'.

Plan long-form or overnight films: director's brief (PLAN FIRST or GO, budget, definition of done), ANIMATION_GUIDE + STORYBOARD, chapter subagents, gates, chunked render. Use for 'overnight film', 'music video', '2-minute film'.

Brand-asset stage of a product film (motion-reel runs the whole film): capture real screenshots, logo, colors and fonts from a URL, never redraw UI; brand story beats; voice + mascot. Use to grab or fix a brand's assets.

Make a launch video, app or product reel, explainer or motion ad from code, end to end: brand assets from a URL, beat-grid shot list, synced sound, critique loop, all formats. Default for new video requests; not footage edits.

Use when... a user wants auditable short-video scripts grounded in live Qianchuan reference scripts and customer-confirmed facts.

Record a genuine screen recording of a short video and verify it frame-accurately. Use when asked to capture, record, or screen-record a video URL, or to process a batch of them from a queue file. Handles the full download-to-verified-output pipeline on a headless Linux box.

Use when... a user provides an original short-video script and requests an auditable rewrite grounded in Qianchuan references and confirmed customer facts.