Video & Animation

Bring code-drawn visuals, Lottie animation, 3D scenes, or stabilized source video into Montagent as lossless footage, or an SVG still as one PNG, each with a rebuild recipe. Load for a look outside the effects list, generative visuals, SVG, Lottie or 3D model input, or source-video stabilization, reference-frame locking, ping-pong and slow-down.

QA a rendered launch film before anyone ships it. Use for "QA the video", "check the render", "is it ready to post", "review the film", "why does it look cheap", "check the captions/safe zones/loudness", after every render and before package. Reads the CLI's automatic report (`npx launchfilm qa <slug>`), contact sheets and stills at key timestamps, then checks text-at-time readability, contrast, platform safe zones, loudness (−14 LUFS, −1.5 dBTP), ffprobe encode specs, black and frozen frames, cuts on the beat, and the 12 amateur tells as blocking checks. Writes qa/<slug>/QA.md where every line cites a file or a measured number; a check that was not run says NOT CHECKED, never PASS. Use scene-build to fix picture problems it finds, sound to fix the mix, launch-film for the whole pipeline, reference-study for reference videos.

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。

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.

Timing, easing and layout craft for any Montagent video: beat grids from music, easing and hold lengths, safe areas, type scale, restraint. Load before judging whether a piece looks right.

MUST USE when user wants to 调研/research/搜索/search/查/找/look up anything on the internet — e.g. 全网调研 X / 帮我调研一下 X / 查一下 X / 搜搜 X / 看看大家怎么评价 X / X 上有什么讨论 / research this topic。 Also MUST USE when user mentions any platform or shares any URL/链接: 小红书/xiaohongshu/xhs, Twitter/推特/X, B站/bilibili, Reddit, Facebook, Instagram, V2EX, LinkedIn/领英/Boss直聘/招聘/求职/jobs, YouTube, GitHub code search, 小宇宙播客, 雪球/股票行情, RSS feeds, or any web URL. 16 platforms, multi-backend routing (OpenCLI / per-platform CLIs / APIs). Zero config for 6 channels. Run `agent-reach doctor --json` to see which backend serves each platform right now. NOT for: 写报告/数据分析/翻译等内容加工（本 skill 只负责从互联网获取内容）； 发帖/评论/点赞等写操作；已有专门 skill 的平台（先用专门 skill）。 【路由方式】SKILL.md 包含路由表和常用命令，复杂场景需按需阅读对应分类的 references/*.md。 分类：search / social (小红书/推特/B站/V2EX/Reddit/Facebook/Instagram) / career(LinkedIn/Boss直聘) / dev(github) / web(网页/文章/RSS) / video(YouTube/B站/播客) / finance(雪球/股票)。

Editing real video in Montagent: talking heads and interviews, green-screen keys, subtitles and word-by-word captions, lower-thirds and name bars, picture-in-picture, and music ducked under a voice. Use when a Montagent video starts from recorded footage.

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".

Router for all Remotion skills

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.

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.

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.

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。

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.

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.

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。

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.