skills
scenario-labs/skills의 67가지 에이전트 스킬(시나리오·게임·영상 포함), npx skills add로 설치
개인 계정의 공개 저장소 소유자는 GitHub로 로그인해 운영자를 확인하고 소개를 직접 관리할 수 있습니다. 관리자 승인은 필요하지 않습니다.
로그인한 뒤 확인 파일을 만들어 이 저장소의 기본 브랜치에 30분 안에 커밋해 주세요. 다른 브랜치나 포크의 파일은 인정하지 않습니다. 파일을 올린 뒤 아래에서 확인하면 관리자 승인 없이 관리할 수 있습니다. 자격을 유지하려면 파일을 남겨 두세요.
로그인이 필요하면 로그인 화면으로 이동합니다. 로그인 후 이곳으로 돌아와 원하는 확인 버튼을 다시 눌러 주세요.
프로젝트 폴더에서 /nomorevibe verify를 실행하면 소유권을 확인하고 소개를 직접 관리할 수 있습니다.
1분 만에 확인하기AI agent skills for video, image, speech & music generation. Works with Claude Code, Cursor, Windsurf, OpenCode, ClawHub.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
varg — AI Agent Skills for Video, Image, Speech & Music Generation
A collection of Agent Skills for AI video, image, speech, and music generation using the varg platform. One skill to access Kling, Flux, ElevenLabs, Sora, and more — zero code, just prompt.
Works with Claude Code, Cursor, Windsurf, OpenCode, ClawHub, and any tool that supports the Agent Skills standard.
• VARG_API_KEY (get at varg.ai) -- required for all skills
• Cloud mode: curl only (zero dependencies)
• Local mode: bun runtime + ffmpeg
Each skill follows the Agent Skills specification:README 전문 보기 ↗A collection of Agent Skills for AI video, image, speech, and music generation using the varg platform. One skill to access Seedance 2.5, Kling 4, Wan 3, Veo 3.1, Nano Banana 2.1, GPT Image 2.5, ElevenLabs v4 and 200+ more models — zero code, just prompt.
Works with Claude Code, Cursor, Windsurf, OpenCode, ClawHub, and any tool that supports the Agent Skills standard.
# Install a specific skill
npx clawhub@latest install vargai
# Or install all skills from this repo
npx clawhub@latest install vargai
Installs the varg-ai skill and the hosted varg MCP server (https://mcp.varg.ai/mcp, sign in with OAuth on first use):
claude plugin marketplace add vargHQ/skills
claude plugin install varg@varg
Or inside Claude Code: /plugin marketplace add vargHQ/skills, then /plugin install varg@varg.
powers/varg-how-to connects the keyless varg how to MCP (https://api.varg.ai/v2/skills/mcp) in Kiro: Powers panel → Add Custom Power → Import power from GitHub → https://github.com/vargHQ/skills/tree/main/powers/varg-how-to.
# Clone into personal skills directory (available in all projects)
git clone https://github.com/vargHQ/skills.git ~/.claude/skills/varg
# Or symlink individual skills
ln -s /path/to/skills/varg-ai ~/.claude/skills/varg-ai
# Clone into any supported skills directory
git clone https://github.com/vargHQ/skills.git ~/.opencode/skills/varg
# Also works with:
# ~/.agents/skills/
# .opencode/skills/ (project-level)
npx skills add vargHQ/skills --yes
How to make any AI video: not sure which skill fits a video request? Start with how-to-make-ai-video. It searches varg how to (varg.ai/how-to), 2,000+ checked video skills and step-by-step tutorials, and lists the options to offer before anything is generated. No key needed. The same search is available as a public MCP server: claude mcp add --transport http varg-how-to https://api.varg.ai/v2/skills/mcp.
Publish your own skill: connect the varg MCP (claude mcp add --transport http varg https://mcp.varg.ai/mcp) and ask your agent to "publish this skill to varg how to" (install varg-publish-skill so every listing gets the same card). It reads the skill, shows you the card and sends it with publish_skill; the varg team reviews it and it becomes searchable at varg.ai/how-to. No agent: the form at varg.ai/how-to/publish. Skills that join this repo go through a pull request (see the README of each skill and vargHQ/skills-marketplace → Adding a skill).
| Skill | Description | ClawHub Slug |
|---|---|---|
| varg-ai | AI video, image, speech, and music generation | vargai |
| varg-previz | Plan multi-shot AI videos: story, anchors, a three.js grey-box previz with auto cameras, Seedance video-to-video packs | — |
| varg-video-editing | Which service to edit existing footage with, job by job (cuts, talking-head cleanup, Shorts, captions, AI restyle and swaps, dubbing, upscaling, QA, handing off to Resolve/Premiere/Final Cut), plus the rules that keep cuts frame-accurate and in sync | — |
| varg-video-edit-library | 206 named edits for footage you already have, each with the look, when to use it, the common mistake, timings and a recipe for FFmpeg, HyperFrames, Remotion and CapCut: cuts and pacing (J/L cuts, match cuts, speed ramps), transitions (CapCut-style, every FFmpeg xfade, Remotion TransitionSeries, HyperFrames and GL shaders), headlines, quotes, end cards, UI walkthroughs, cursors, device mockups, code and terminal, AI agent flows, numbers and charts, logo walls, grids and backdrops | — |
| media-download | Save videos/audio by link from X, YouTube, Instagram, Facebook, TikTok and 1800+ sites (yt-dlp), and Pinterest images by pin, board or search (gallery-dl). Local, no API key | — |
| short-form-ffmpeg-cookbook | Short-form FFmpeg cookbook: copyable commands for Reels. Crop to 9:16, burn captions (SRT/ASS), trim and join, music with ducking; every command tested. Local, no API key | — |
| paper-shaders |
VARG_API_KEY (get at varg.ai) or the hosted varg MCP (https://mcp.varg.ai/mcp) -- required for varg-ai, varg-previz and the format skills (varg-music-video, varg-character-keynote, varg-consistent-characters, varg-character-design and its category recipes, varg-podcast-video, varg-product-hook, varg-dropship-reel, the Kling 4 skills (kling-4-prompting, kling-4-keyframe-film, kling-4-action-montage), the occasion skills (varg-birthday-video, varg-occasion-video), ecommerce-video-ads, ai-ugc-ad-video, product-photo-to-video); media-download, short-form-ffmpeg-cookbook, paper-shaders, the motion craft skills (caption-styles, title-design, kinetic-type-art, social-overlays, app-ui-mockups, transition-craft, motion-formats, maps-and-data-motion, footage-effects), how-to-make-ai-video, varg-publish-skill (publishes with a varg login or key, no credits), video-skills, varg-video-edit-library, varg-video-editing, ecommerce-video-edit, ad-creative-testing, product-video-hooks (written hooks) and tiktok-shop-video (scripts) need no key (varg-video-editing uses varg only for the paid edits it recommends)npx hyperframes), or a Remotion 4 project; footage-effects needs only ffmpegEverything the plugin runs, sends or fetches:
https://api.varg.ai/v2/skills/mcp): no key. Sends the search text you or the agent type (search_how_to) and returns how-to cards. Searches are logged without IP or key (query, MCP client name, top results).https://mcp.varg.ai/mcp): OAuth sign-in to your varg account. Sends generation requests (prompts, uploaded or linked media) and spends your varg credits; results are stored in your varg account and served from s3.varg.ai.https://api.varg.ai): the skills' curl examples send VARG_API_KEY from your environment as a Bearer token to api.varg.ai only, never to any other host. Skills that need no key are listed under Requirements.yt-dlp on your machine against the video link you give it (X, YouTube, Instagram, TikTok, Pinterest and other sites yt-dlp supports) and saves the file locally.varg-previz/templates/previz/vendor/ so the previz page works offline.Privacy: https://varg.ai/privacy · Terms: https://varg.ai/terms · Security: SECURITY.md
Each skill follows the Agent Skills specification:
skill-name/
├── SKILL.md # Core instructions (required)
├── references/ # On-demand reference docs
│ ├── models.md # Model catalog, pricing
│ ├── components.md # Component reference
│ └── ...
├── scripts/ # Executable setup/helper scripts
│ └── setup.sh
└── assets/ # Static resources (optional)
mkdir my-new-skill
my-new-skill/SKILL.md with cross-compatible frontmatter:---
name: my-new-skill
description: >-
What this skill does and when to use it.
Include trigger keywords for auto-discovery.
license: MIT
metadata:
author: vargHQ
version: "1.0.0"
openclaw:
requires:
env:
- VARG_API_KEY
anyBins:
- curl
- bun
primaryEnv: VARG_API_KEY
homepage: https://varg.ai
compatibility: >-
Describe environment requirements here.
allowed-tools: Bash(bun:*) Bash(curl:*) Read Write Edit
---
Your skill instructions here...
Add reference docs in my-new-skill/references/ (keep SKILL.md under 500 lines).
Push to main -- the GitHub Actions workflow auto-publishes all changed skills to ClawHub via clawhub sync.
| Field | Standard | Purpose |
|---|---|---|
name | Agent Skills | Skill identifier (must match directory name) |
description | Agent Skills | What + when (used for discovery in all tools) |
license | Agent Skills | License (MIT for ClawHub) |
metadata.author | Agent Skills | Author identifier |
metadata.version | Agent Skills | Semver version string |
metadata.openclaw | ClawHub | Runtime requirements, env vars, binaries |
compatibility | Agent Skills | Human-readable requirements description |
allowed-tools | Claude Code / OpenCode | Pre-approved tools (experimental in OpenCode) |
All fields are cross-compatible -- tools ignore fields they don't recognize.
# 1. Install the skill
npx clawhub@latest install vargai
# 2. Set your API key
export VARG_API_KEY=varg_xxx
# 3. Run setup to verify environment
bash varg-ai/scripts/setup.sh
# 4. Make your first clip (TSX/JSX videos are deprecated; edit clips with the ffmpeg tool)
curl -fsSL https://varg.ai/install | sh && vargai login
vargai generate_video --prompt "waves at dusk" --duration 5
MIT
| Frame-exact Paper Shaders backgrounds, textures and logo stings for HyperFrames and Remotion videos, plus 11 researched looks (liquid metal, god rays, marbled ink, contour lines, datamosh rain…) |
paper-shaders |
| how-to-make-ai-video | How to make AI video skills for Claude Code and AI agents: one search over varg how to gives the best skill with its install command, the options to show (with example videos) and the steps; also lists a skill you built on varg how to | — |
| varg-publish-skill | Turn your agent skill into a varg how to listing: checks it against the standard, fills the one card format every listing uses, shows it to you and publishes it for review through the varg MCP | — |
| video-skills | Finds the right skill for any AI video, image, voice or music job on varg how to and shows the options before anything is generated | — |
| varg-music-video | Direct an AI music video for a song: beat grid, cast and references, storyboard on downbeats, stills first, Seedance shots, lip-sync pass, lyrics in code, platform masters | — |
| varg-character-keynote | A 9:16 "character on stage" video in one render: a custom character delivers a punchy line at a keynote with native voice and applause | — |
| varg-consistent-characters | Multi-scene ads where the same actors, outfits and location stay consistent: profile cards, location card, stills for approval, native-audio clips, captions and QA | — |
| varg-character-design | How to make characters people remember: definition, details, clothing, character card, a sheet that stays consistent, reuse in Nano Banana, Kling 4 and Seedance 2.5; moderation-safe wording and famous looks without IP names | — |
| varg-character-grandma | Grandma and grandpa characters: age vs behaviour, real wrinkles, a loud signature outfit | — |
| varg-character-big-guy | Big guy characters: size shown with visible cues, a small delicate job, played straight | — |
| varg-character-cringe | Cringe deadpan characters: bizarre but believable, one outfit, one repeated action (the Jean Phil format) | — |
| varg-character-beautiful-woman | Very beautiful women who look like real models: casting details, real skin, fashion and swimwear styling, wording that passes filters | — |
| varg-character-fitness | Fitness characters: real muscle, a signature colour, catalog wording that gets sports bras past false blocks | — |
| varg-character-absurd | Absurd characters and mascots: food or object + animal + one human detail, in 3D, clay or photoreal looks | — |
| varg-character-cartoon | Original characters in famous cartoon looks (yellow sitcom, 3D family film, anime, paper cut-out, clay, puppet), described without IP names | — |
| varg-podcast-video | Podcast and two-character dialogue videos: studio set, wide master shot, close-ups and cutaways from it, 5 s scenes with native dialogue audio, stitched episode | — |
| varg-product-hook | The first 5-15 s of a product video: find a hook that already sells this kind of product, then remake the idea as realistic phone or security-cam footage with Seedance 2.5 from your product photos (backyard camera, torture test, problem → solution, satisfying close-up, gift reaction) | — |
| varg-dropship-reel | The 60-second faceless dropshipping guide reel end to end: product hook, laptop on a zigzag tablecloth with a hand pointing a marker at cost, price and demand, the steps, a night-skyline iPhone ending with "Comment GUIDE", word-by-word lime captions and the comment-to-DM funnel; scripts cut and caption the reel | — |
Turn product photos into an AI ASMR unboxing video · varg-unboxing-video | Turn product photos into an unboxing or ASMR video: hands-only, timed VISUAL/ACTION/SOUND blocks on Seedance 2.5 with the box and product as references, native ASMR sound, no music | — |
Turn product photos into AI video that keeps the logo and shape · varg-product-lock | Keep the product identical in every AI shot: lock sheet, multi-angle reference pack, camera-risk ladder, drift check with video_analyze, real packshot composited back | — |
Turn an AI video ad into a TikTok Shop and Meta compliance report · ai-ad-compliance-check | Check an AI video ad against TikTok Shop, Meta, YouTube, FTC, New York and EU AI Act rules: video_analyze pulls every claim, a versioned rules file scores pass / fix / block with sources | — |
Turn winning competitor ads into new AI video ad angles · varg-ad-library-remix | Turn winning competitor ads (Meta Ad Library, TikTok Creative Center, Kalodata) into new angles: winner signals, video_analyze breakdown, angle × format × awareness gaps, originality check | — |
Turn one winning video ad into 20 AI ad variations · varg-ad-variants | Turn one winning ad into 20 Andromeda-safe variants: concepts × single-variable iterations, new hooks on varg, a tested FFmpeg hook-swap script with a manifest, naming and test plan | — |
Turn a product into an AI singing mascot jingle ad · varg-mascot-jingle-ad | Turn a product into a singing mascot ad: claymation character with a lip-sync face, Lyria jingle trimmed to 15-30 s, Seedance 2.5 singing to the song, master track laid back on top | — |
Turn one video ad into 8 languages with AI dubbing · varg-ad-localize | Turn one video ad into 8+ TikTok Shop markets: transcreation with an offer card per market, same-voice ElevenLabs, lip-sync only where faces talk, re-cut on the new words, native review | — |
Turn a flat-lay photo into an AI virtual try-on video · varg-tryon-video | Turn a flat-lay of clothing into a try-on video: Bria virtual try-on still, fit-check motions on Kling 4 (turn, tug, walk, mirror), honest size captions | — |
How to prompt Kling 4 for cinematic AI video · kling-4-prompting | Prompt Kling 4 so it doesn't look like AI slop: key frames from GPT Image 2.5 plus a short director's prompt, reference-to-video inputs (@Image, @Element, @Video, @Voice), multi-shot with seconds, "no slow motion", dialogue in quotes, the failure fixes and the varg call with the reference route pinned | — |
Make a short film or music video with Kling 4 · kling-4-keyframe-film | Beat sheet, cast and look references, every key shot as a still on GPT Image 2.5, one 30-second Kling 4 take per beat, then cut the usable seconds like live-action takes; dialogue scenes and the fix pass | — |
Turn key frames into a 30-second action montage with Kling 4 · kling-4-action-montage | Breakdance, BMX, skate or chase montages from 3-4 stills and a one-line prompt: the moves, "no slow motion", the end state as the last still, picking the seconds that land | — |
Turn photos and a name into a personalized AI birthday video · varg-birthday-video | Turn a name, an age and a few photos into a birthday video that is theirs: a song that sings the name (ElevenLabs Music), their photo blowing out the candles (GPT Image 2.5 still, Seedance 2.5 or Wan 3), a retro camcorder tape, then-and-now, a pet's birthday, a milestone poster; the formats that spread on X, 27 looks, and a script that exports for WhatsApp | — |
Turn photos into an AI invitation, holiday greeting or tribute video · varg-occasion-video | Invitations and save-the-dates that unfold with the names and date printed right (Seedance 2.5 first and last frames), Christmas, New Year, Diwali and Eid greetings, old photos brought to life, a love story across the eras, calm memorial and retirement tributes | — |
| ecommerce-video-ads | Ecommerce video ad skills for Claude Code and AI agents: the plan and the rules for product videos that sell on TikTok, TikTok Shop, Reels and Meta: four questions, angle, hook, format, casting, second-by-second script, product fidelity, platform rules (AI labels, Shop bans, FTC), cost gate, variants | — |
| product-video-hooks | Turn a product into scroll-stopping video hooks: three agreeing layers (frame, line, text), 70-line hook bank, claim checks, one MP4 per hook | — |
| tiktok-shop-video | Turn a product link into a TikTok Shop video: Shop styles, 18–35 s script, phone shot list, and the 2025–2026 rules (real presenter, no AI voice, no slideshow, no mismatched prices) | — |
| ai-ugc-ad-video | Open-source alternative to Arcads: AI UGC ads with an AI actor holding your real product, face, voice and label locked, phone-real prompts, no AI testimonials | — |
| product-photo-to-video | Turn a product photo into a product video ad: truth sheet, cut-out, storyboard sheet, Seedance PRODUCT LOCK, repair pass, real packshot on the last second; ASMR unboxing, macro, 360° loop, try-on | — |
| ad-creative-testing | Turn one winning ad into a creative testing plan: decode the winner, concepts Meta and TikTok treat as different, how many and when to judge, funnel diagnosis, kill/scale rules, naming | — |
| ecommerce-video-edit | Turn raw product footage into a TikTok ad with FFmpeg: hook moved to second 0, dead air cut, text hook and CTA in the safe zone, -14 LUFS, one MP4 per hook, 4:5 cut; tested commands, local, no API key | — |
| caption-styles | 17 animated caption looks (karaoke, word pop, sliding pill, word tiles, glass, neon, serif italic…) as exact presets: font, size per format, sync 1-2 frames before the voice, 9:16 safe box, contrast; from the Netflix and BBC subtitle guides and short-form editors. Local, no API key | — |
| title-design | Lower thirds, speaker plates, show opens and headline, CRT, chrome and glitch title reveals with broadcast rules: hold time from word count, safe areas, type scale, plate opacity, easing. Local, no API key | — |
| kinetic-type-art | 17 kinetic typography pieces on the beat (typeface strobe, word slams, word tunnel, manifesto, riso and photocopy posters, decrypt and glitch wordmarks) with beat maths, spacing, free font pairings and palettes. Local, no API key | — |
| social-overlays | CTA pills, follow and subscribe prompts, review cards, LIVE badges, comment callouts and counters in one 9:16 safe box from TikTok, Reels and Shorts specs, timed after the payoff, real data only (FTC). Local, no API key | — |
| app-ui-mockups | Message threads, team chat, coding-agent terminals, code walkthroughs, 3D code slabs and exploded UIs rebuilt as HTML at measured typing and reading speeds, with theme colors from MIT sources. Local, no API key | — |
| transition-craft | Shake-distort cuts, Droste infinite zooms, tile-flip mosaics, bokeh, collage and 3D gallery transitions: frame counts, beat placement, log-space zooms, decaying shakes. Local, no API key | — |
| motion-formats | Seven formats with beat structures and presets: analog-horror website trailer, VHS bouncing logo, pixel-art runner on a CRT, wave wordmark, DAW one-shot music video, launch trailer, kinetic lyric video. Local, no API key | — |
| maps-and-data-motion | Globe-to-country dives, satellite flyovers, route and watercolor maps, self-drawing charts, count-ups, versus timelines and audio visualizers, with map credits and tile licences built in. Local, no API key | — |
| footage-effects | Glow and halation, glitch, CRT, chromatic aberration, halftone, acid photocopy, zoom blur and heat haze on real footage: 34 tested ffmpeg chains plus HyperFrames and Remotion versions. Local, no API key | — |