Video & Animation

Create or refine essay videos that feel like a live word-editor session, with expressive text edits, browser cutaways, and synchronized keyboard sound. Use for social launches of writing and the Kurate video style, rather than generic talking-head or montage-only edits.

Compress existing opaque videos with the user's standard MP4/H.264 profile while preserving resolution, frame rate and duration. Remove audio by default; preserve it when explicitly requested. Use for compression, not animation editing or export from design apps.

Plan, script, build and deliver any video in Remotion with an AI narrator, word-synced captions, the person's own footage, photos and screenshots, real royalty-free music, and a frame-by-frame check before delivery. Covers personal stories (funny or sincere), product promos, explainers, tutorials, recaps, greetings and announcements, vertical (Reels, TikTok, Shorts, WhatsApp Status) or landscape. Use whenever someone wants a video made, improved or re-cut, or says video-xoshaka, even if they don't mention Remotion, narration, captions or music.

Use when asked to make a short product launch / feature demo video (e.g. a 20-second 1080x1080 clip from real product screens) with a coding agent such as Claude Opus 5.5, using a code-driven Remotion pipeline plus optional open-source design, animation, 3D and export libraries.

Text-to-speech for voiceovers and spoken output. Renders narration for product, feature, and demo videos as per-scene audio files with exact durations, timing offsets, and captions, and speaks text aloud. Supports the built-in macOS/Linux voice, local Kokoro, OpenAI, ElevenLabs, Gemini, Cartesia, and Deepgram. Use when the user wants a voiceover, narration, or audio for a video or screencast, wants something read aloud or announced, wants an audio file (.wav/.mp3/.m4a), wants to compare or choose voices, or wants to set up TTS providers and API keys.

Use when an existing assembled video or cut needs revisions, recutting, captioning, enhancement, review, or export. For making a new video from an idea or raw footage, use creating-videos.

>- Make a short narrated film (an explainer, a product intro, a feature announcement, a walkthrough of code or a design) from a prompt. Every frame is drawn in code on a canvas, with a local AI voice (Kokoro), a score made in code, word-timed cues, burned-in captions, and .srt/.vtt files. The flow: prompt, then a storyboard page with the real voice and music that the user iterates on, then the full 1080p render. Real frames are sampled and checked at every step. Use when the user asks to "make a video", "make an intro/explainer/announcement video", "a narrated walkthrough of this PR/feature/repo", "turn this into a video", or wants to change the storyboard, script, voice, music or look of a film made this way.

Use when a user wants a new video made from an idea, research, supplied footage, a screen recording, a reference, or generated assets. Use for end-to-end production, including when raw footage is provided; use editing-videos when they only want an existing cut changed or reviewed.

Produce narrated, animated videos programmatically (any niche, format or visual style) — script → AI voice (ElevenLabs) → word-level alignment → deterministic animation (Remotion + SVG) → synced subtitles, SFX/music mix → validated MP4 + SRT + thumbnail + upload metadata. Use when the user wants to create, plan, restyle or batch-produce explainer/educational/kids/product/story/short-form videos, design a visual style from references, set up a video pipeline, or fix sync/subtitle/render problems in one.

Content strategist and manager for the maths video channel. It asks for the target audience, region, platforms, capacity and goals, researches the market (competitors, outlier videos, search demand, monetization rules, RPM by region, upload timing), and writes a growth and monetization strategy, a scored idea backlog, a content calendar and ready-to-produce video briefs (problem, angle, hook, outline, titles, thumbnail, description, tags, upload slot) that /manim-video turns into videos. It also reviews channel analytics and updates the plan. Use when the user says "plan the channel", "content plan", "video ideas", "what should the next videos be", "channel strategy", "grow the channel", "monetize", "make briefs", "review my analytics", or invokes /channel-planner.

Name, file, and track AI-generated media (images, video, audio) to a human-readable convention backed by a machine-readable sidecar, so assets are recognizable at a glance and reliably retrievable by an agent when orchestrating them into downstream workflows. Grounded in film/VFX shot naming (Netflix, MovieLabs), the MovieLabs Ontology for Media Creation (metadata separated from essence; structural vs functional), work-vs-publish separation (ShotGrid, ftrack), and C2PA/EU AI Act provenance. Use when downloading or filing generated assets, designing or applying an asset naming convention, building a project media library, or pulling assets into a lip-sync / image-to-video / edit workflow. Triggers on "name this asset", "naming convention", "file this generation", "asset library", "how should I name this", "sidecar", "content credentials", "provenance", "grab the asset for the workflow", "version this render".

Use when writing or auditing any written content — captions, carousels, newsletters, threads, LinkedIn posts, long-form — and as the FINAL filter on every piece. Reach for it whenever text feels generic, fluent-but-hollow, over-formatted, or "sounds like AI." Strongest on written prose; lighter on spoken reels.

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/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. 15 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) / dev(github) / web(网页/文章/RSS) / video(YouTube/B站/播客)。

Expert guide to discovering, tracking and critically evaluating the AI landscape (models, tools, papers, products) as of late 2026. Use when the user asks \"what's the best model for X\", \"what's new in AI this week\", \"which coding agent should I use\", \"is this benchmark claim legit\", \"compare GPT vs Claude vs Gemini\", \"best open-weight model I can self-host\", \"what AI tool fits my use case\", \"where do I track AI papers/releases\", \"explain this leaderboard\", \"is model X actually SOTA\", or wants a current frontier snapshot (labs, open weights, agents, coding tools, multimodal/video, on-device). Not for building/training models from scratch, general ML tutoring, or startup finance/strategy (use the start-up-tech-* skills).

Design, diagnose, and lock believable photorealistic adult human character identities that avoid generic, averaged, or visibly AI-generated faces. Use when creating adult actors or characters for film, video, games, or visual stories; when a portrait feels like an "AI face," stock model, beauty render, or character sheet; when a face must feel sympathetic, ordinary, distinctive, or consistent across scenes; or before turning a character identity into cinematic, emotional, or continuity-controlled images and video.

Write and revise prose so it reads like a specific, thoughtful person wrote it, not an AI. Use this skill for any writing the user will send, post, publish, or say out loud, even when they don't mention AI or tone: business emails, Slack or Teams messages, memos, reports, proposals, procedures, LinkedIn and other social posts, captions, bios, YouTube scripts, video narration, hooks, outlines, and stories. Also use it whenever the user asks to humanize or de-AI text, make it \"sound like me\", check whether something sounds like AI, review a draft for AI tells, tighten or simplify prose, or rewrite in plain English. Combines Orwell's rules and plain-English discipline, Wikipedia's Signs of AI Writing field guide, a channel-aware humanizer pattern list, and StoryScope research on the narrative habits that give away AI-written stories.\n

Convert handwritten, typed or OneNote-exported course notes, homework solutions and exams into accessible Markdown that builds both a tagged PDF (PDF/UA-1) and a website (WCAG 2.1 AA), with every figure redrawn as described SVG and every equation typeset in real notation with a spoken form. Bundles the build scripts, LaTeX class, checks, a notation lint and a project scaffold. Use this whenever someone uploads or points to lecture notes, scanned or OneNote PDFs, homework solutions or teaching material and mentions accessibility, ADA Title II, WCAG, PDF/UA, screen readers, alt text, or transcription, and also for any later review round on such a project (fixing figures, labels, math notation, captions, glyphs missing from a PDF, reorganizing the repository), even if they do not say "accessible".

Connect a social media account or channel to PostEverywhere. Use when the user wants to add Instagram, TikTok, YouTube, LinkedIn, Facebook, X, Threads, Pinterest, Bluesky, Telegram, Discord or WordPress, or when a post can't go out because no account is connected.

Connect a social media account or channel to PostEverywhere. Use when the user wants to add Instagram, TikTok, YouTube, LinkedIn, Facebook, X, Threads, Pinterest, Bluesky, Telegram, Discord or WordPress, or when a post can't go out because no account is connected.

Read-only ads audit (Meta, Google, LinkedIn, TikTok, Microsoft, YouTube, Apple) from exports, scored. Use for: granska mina annonser, annonsgranskning, audit my ads, PPC audit.

Connect a social media account or channel to PostEverywhere. Use when the user wants to add Instagram, TikTok, YouTube, LinkedIn, Facebook, X, Threads, Pinterest, Bluesky, Telegram, Discord or WordPress, or when a post can't go out because no account is connected.

Make finished static charts with plotnine, in Chinese or English, for slides, documents, weekly/monthly business reviews and journal papers. Picks the chart from what the user wants to compare (ranking, trend, change between two points, composition, contribution to a change, target attainment, metric decomposition, distribution, relationship, estimates with intervals, KPI summary), renders it at the exact output size (slide / doc / journal column) as SVG, PDF and PNG, and checks the layout automatically. Business-review charts follow reporting conventions (▲▼ instead of red/green, colour only for anomalies, basis captions, period-aligned comparisons). Use when the user asks to draw, plot, chart or visualize data, make a figure for a deck, report, business review or paper, or fix a chart that looks "busy but says nothing". 用于画图、做图表、汇报/周报月报/论文配图、改图。Not for interactive charts (plotly/echarts) or dashboards.

Theo dõi fanpage Facebook, kênh YouTube/TikTok và tin ngành của đối thủ để lập báo cáo thị trường có nguồn, nêu khoảng trống nội dung và ý tưởng hành động. Dùng cho bản tin tuần/ngày hoặc hồ sơ đối thủ tùy ngành; không dùng để tương tác hay vượt giới hạn nền tảng.

Stop the generic admin-dashboard screen on any product. Use on every new or changed UI, page, list, settings view, reading view, or generated layout. Bans obvious subtitles, the eyebrow-title-subtitle stack, decorative metric heroes, equal KPI cards, card-in-card, decorative row chrome, and a narrow centered column with huge side gutters. The first screen is this product's job at full content width. Secondary facts get a named place, not a delete. Do not strip a manager down to a database grid, and do not hide a list action behind Open. A product with several areas gets a persistent sidebar. Personality comes from composition and the existing accent, not from a display font. Do not reuse the objects, verbs, or layout of a previous screen. Triggers: UI, tela, dashboard, cards, KPI, métricas, subtítulo, layout, "muita informação", "não sei onde olhar", /lean-screen. Task rules stay in ux-best-practices. Visual identity stays in the design system or frontend-design.