Video & Animation

Cut-out character animation in Montagent: mascots and puppets built from a rig of parts, posing and gestures, lip sync to a voice line from visemes, blinks. Use when a Montagent video has a character that moves or talks.

用代码（不用 AI 视频或音乐生成模型）制作动画短片的通用规范：产品介绍、片头、宣传片、自我介绍等，画面、动效、转场、配乐、音效、排版全部用代码完成时使用。

Router for all Remotion skills

Make product videos from real screen recordings of the user's own app. A tutorial explains features chapter by chapter (chapter cards, numbered captions, key chips, synthetic cursor, click rings, zooms, recap cards, music and SFX, joined full cut). A promo is a 30-45 s 16:9 cut with cuts on the music's beat, short headline cards and an end card. Captures native macOS apps itself (agent-desktop + ffmpeg) and takes web footage from the web-video skill. Use for "tutorial video", "product walkthrough", "promo video", "launch video", "video hướng dẫn", "video quảng cáo", "video giới thiệu sản phẩm".

Build a narrated, ~30-minute technical video course with Remotion + ElevenLabs in an Omarchy / Tokyo Night terminal style (Waybar, tiled window, oh-my-zsh prompt, markdown headings), with every on-screen element revealed exactly when the narrator says it, sentence pauses, plain subtitles, fast remote rendering, and a chaptered web player. For hands-on topics it first builds a real lab (qemu/KVM VMs) and uses the lab's real command output on screen and in the narration. Publishing to the videolabs site (Cloudflare) is done by the videolabs_publish skill. Use when the user wants an explainer/course/training/lab video on a technical topic, or wants to edit, re-voice, re-render or republish one made this way.

Publish the whole videolabs site (https://videolabs.pages.dev): sync every lab in the repo's labs/ folder into the site (video to Cloudflare R2 only when it changed, player assets, lab resources, catalog), commit and push the videolabs repo so GitHub Actions deploys to Cloudflare Pages, then verify every lab. Use when the user wants to publish, deploy, add, update or re-upload video labs to videolabs.pages.dev. One run publishes all labs, not just one.

Ver, escuchar y analizar un video (URL de YouTube u otro sitio, o archivo local). Úsalo siempre que el usuario pida ver, resumir, analizar o extraer información de un video, por ejemplo una estrategia de trading explicada en YouTube. Convierte el video en hojas de fotogramas con hora, transcripción del audio y cortes de escena alineados en una línea de tiempo.

输入一个主题，自动生成一支「Vibe 知识大赏」风格的中文知识短片（暗底、光、细衬线中英双语字幕、无旁白、画面动作卡音乐节拍、每个视觉事件配真实拟音），直接渲染出 mp4。每个主题都按它自己的内容重新设计开场、画面世界、转场和声音。用户说"做一期 XX 的知识视频 / Vibe 风格视频"，或给出一个知识主题要求出片时使用。可选：换 BGM、指定时长。

Turn an interview, podcast, lecture, founder story, or other spoken video/audio file into a verified transcript, meaningful highlights, and an ordered, timestamped story cut list. Use whenever the user asks for the best moments, clips, an interview highlight reel, a rough cut, an edit decision list, or transcript-to-video storytelling, even if they do not mention StoryCut by name.

Motion graphics and titles in Montagent: launch spots, product and social ads, intros and outros, kinetic and typed text, logo reveals and loops, wipes, pops, colour changes, card flips and door swings in perspective, and trailer effects (slammed titles, glitch cuts, HUDs). Use when a Montagent video is built from graphics and text rather than footage or a character.

用 videobook 流水线把一个 YouTube / Bilibili 视频（尤其是课程录像）做成图文并茂的 Markdown + HTML 技术电子书：提取字幕（平台无字幕时用本地 faster-whisper ASR 兜底）、由大模型改写并插入 SCREENSHOT/SLIDE 占位、用已登录浏览器直接截取播放器高画质帧、渲染暗色主题 HTML 并起本地预览、可选抓取课程官方讲义与 4K 幻灯片增强、生成字幕校订对照稿、最后发布到 GitHub Pages 的 pages 分支。当用户给出视频链接或 BV 号并要求做成电子书/教程/图文/文档、要字幕纠错对照稿、要给文章配视频截图、或要发布已生成的 output 时使用。不负责单纯下载视频或字幕、视频剪辑，以及与本流水线无关的网页爬取。

Make motion design videos in code with Remotion, cut to music with sound effects that land on the frame. Use when someone asks for a promo, launch film, product video, explainer, social cut, landing loop or any "motion design video", for a product or not. It checks the machine, studies 2 to 5 videos the user likes (frames, spectrogram, volume), stops for the user's approval on look, claims, music and storyboard, builds from the product's own design, scores the film offline and will not call it done until a sync, loudness and picture gate passes. Optional modes: re-study the reference galleries for new entries, and generate photos or footage loops with Gemini when code cannot draw them.

Make a promo video with commercial polish (4K MP4 + web version) for a product, website, app, SaaS, open-source project, course, AI model, or event. The pipeline covers research, requirements interview, style sample frames, real 4K page screenshots, music with beat cuts, AI narration, mixing, and automated verification. Use it when the user says "make a promo video," "make a product intro video," "make a promotional video / trailer," "make a video for this project," or asks for a promo video, product trailer, or launch video.

Direct, generate, edit, compare, and review visual assets with current Google Gemini image models. Use for image creation, image editing, reference-based consistency, product and character visuals, text-bearing graphics, grounded diagrams, video-derived images, and multi-model image portfolios.

Turn local voice memos, meeting, interview, lecture or podcast recordings into transcripts, notes, summaries or action items, privately on the user's computer, using the Yap MCP tools (transcribe_audio, search_dictations). Use when the user mentions an audio or video file, a recording, a voice memo, or something they dictated earlier.

Build a dynamic talking-head edit in Remotion from pre-cut footage. Punch-ins chosen from the transcript, a side layout with content cards, a screen scene with a camera picture-in-picture, rotoscoped depth (hero words and objects passing behind the speaker), path-traced 3D hero type and 3D logos, fixed captions with phrase emphasis, subtle SFX tied to visible events, then a measured audio sync trim. Use when the user asks to make a talking head "more dynamic", add punch-ins or zooms, put graphics beside or behind them, show a screen walkthrough, or "edit this like a pro". Expects PRE-CUT footage (a raw take goes to /autocut first). Proposes a plan for approval before rendering anything expensive.

Turn a raw talking-head recording into a clean pre-cut file. Measures audio/video sync from the speaker's lips and fixes it, transcribes word by word, finds stumbles, false starts and restarts with a verbatim pass, removes silences, proposes retakes, renders with intact word edges and re-verifies the cut. Use when the user says "cut this", "clean this recording", "remove the silences and mistakes", "autocut", "prep this for dynamic-edit", or drops a raw recording and asks to get it ready. Plan first, render only after the human approves.

Điều phối — nhận input, xác định loại, chuyển tiếp tới skill anh em trong repo — doc (tài liệu), media (video/audio/ảnh), paper (báo cáo), web (đọc web/mạng xã hội/tìm kiếm) — và mở cổng câu hỏi bắt buộc trước khi viết báo cáo. Dùng khi chưa biết skill nào xử lý input đang có, hoặc khi cần quy trình phối hợp nhiều skill.

将主题或材料经搜索核实制成苹果 Live Photo 信息动效卡。用 Remotion 编排图片、PNG、emoji、场景、时间线和可选图表，适用于知识解释、产品比较、变化过程及数据故事；制作可预览动效与配对实况资源，说明接收端兼容性。

Apply, switch or regenerate a project's design system (the one source of colours, fonts, logos, chart, email, document and video styles for every output), check its contrast in light and dark, and add a new design system to the framework from a Claude Design System artifact. Uses scripts/design-system.mjs to generate tokens.css, tokens.ts, a Fluent UI theme, a model-driven app theme, email and chart palettes, the video-tour brand block and an Office theme mapping. Use when the user says "design system", "apply the brand", "use our house style", "switch theme", "rebrand", "regenerate the tokens", "check contrast", or "add a design system".

Xử lý media — ffprobe thông số, ffmpeg cắt/ghép/transcode/GIF, trích frame, tách audio, phụ đề (shift/convert/burn), timecode SMPTE, phát hiện shot, OCR frame/ảnh, tải từ URL (yt-dlp), nhạc (thể loại/mood/BPM/lyrics). Dùng khi đưa video/audio/ảnh/URL media và muốn tóm tắt nội dung, trích text, cắt ghép, làm phụ đề, lấy khung hình, hoặc tải xuống.

Find everything worth keeping from this AI session (the main build, side discoveries, techniques, the feedback the user taught you), rate each by rarity on evidence, save them to LOOT.md, and turn the rare one that earns it into finished public work: a GitHub repo with docs and a logo, an agent skill, a blog post, a gist, a demo video. Also reads past Claude Code and Codex sessions. Use when the user says "loot", "/loot", "what's worth shipping from this", or "turn this into a repo / skill / blog post".

Write, structure and fix prompts for Kling 4.0 and Kling 4.0 Flash video generation, following Kling's official 4.0 user guide. Covers text-to-video, Omni Reference (@Image/@Video/@Element roles), first/last frames and up to 10 keyframes, 30-second multi-shot timelines, video editing ("Edit @Video1"), creative recreation of a reference video, dialogue in any supported language or accent, burned-in subtitles and on-screen text, and stylized looks. Use this skill whenever the user mentions Kling, Kling 4, Kling 4.0, Kling Flash or kling.ai and wants a video prompt written, expanded, converted, translated or debugged, even if they just describe a scene ("make a Kling prompt where…", "turn this idea into a 15s ad for Kling", "my Kling video keeps adding music / the face drifts / the camera ignores me"). Also use it to convert prompts written for Kling 3.0, Seedance, Veo, Sora or other video models into Kling 4.0 format.

Distill a book, long-video transcript, podcast, course, or interview into a coherent set of executable skills. Use when the user asks to "拆书" / "蒸馏一本书" / "把 XX 书做成 skill" / "把这个视频/播客/课程蒸馏成 skill" / "turn a book or video into skills" — i.e. wants the frameworks, principles, and methodologies in long-form content extracted into atomic, reusable Claude skills that an agent can invoke in real-world situations. NOT for simple summarization, book reviews, or role-playing as the author (that is nuwa-skill's job).