What is the best agent skill stack for video, images and audio?
Reviewed . Scores and counts come from the directory snapshot of .
Short answer: image from coreyhaines31/marketingskills (88/100) creates and edits marketing images with the generator you choose, remove-background from Bria-AI/bria-skill (86/100) cuts them out with Bria RMBG 2.0, and vox-director from Alisa0808/vox-director (85/100) turns one topic into an explainer or ad video with script, motion, voice-over, music and captions. capcut-edit edits the result and cuts long-form into shorts, blog-audio narrates posts, creative-director supplies the concept, and claude-real-video-for-agents lets the agent watch what it made.
The stack
Seven skills, in score order. Scores come from the directory snapshot of October 22, 2026, and install counts are the figures skills.sh published on November 2, 2026.
| Skill | Repository | Score | What it does |
|---|---|---|---|
| image | coreyhaines31/marketingskills | 88/100 | Creates, edits and optimizes marketing images such as blog heroes, social graphics, product mockups, banners and OG images, with tools like Flux, GPT Image, Gemini and Recraft. |
| remove-background | Bria-AI/bria-skill | 86/100 | Removes backgrounds from images to produce transparent PNGs, cutouts and masks by segmenting foreground from background, using Bria RMBG 2.0. |
| vox-director | Alisa0808/vox-director | 85/100 | Turns one topic into a Vox-style paper-collage explainer or ad video, automating script, motion, voice-over, music and captions via Atlas Cloud and ffmpeg. |
| capcut-edit | renezander030/capcut-cli | 85/100 | Edits CapCut and JianYing video projects, including subtitles, timing, speed, volume, templates, animations, and cutting long-form video into shorts. |
| blog-audio | infrasity-labs/dev-gtm-claude-skills | 82/100 | Generates MP3 narration of blog posts with Google Gemini text-to-speech, including summaries, full read-alouds and two-speaker podcast mode, with HTML5 embed code. |
| creative-director | smixs/creative-director-skill | 81/100 | Generates creative concepts using SIT, TRIZ and lateral thinking, scores them against six weighted criteria, and refines them recursively to a 9+ threshold. |
| claude-real-video-for-agents | HUANGCHIHHUNGLeo/claude-real-video | 80/100 | Installs and uses the crv tool so an agent can watch videos by extracting deduplicated scene-aware keyframes and transcribing audio. |
How to use it
- Start with the idea, using creative-director. It generates concepts with SIT, TRIZ and lateral thinking, scores them against six weighted criteria and refines them until they clear a 9+ threshold. A weak concept is an expensive render, so spend the time here; every later step multiplies whatever you decide now.
- Make the stills with image: blog heroes, social graphics, product mockups, banners and OG images, generated with the tool you already use, such as Flux, GPT Image or Gemini. Then run remove-background for transparent PNGs, cutouts and masks with Bria RMBG 2.0.
- Produce the video with vox-director. One topic in; a Vox-style paper-collage explainer or ad out, with the script, motion, voice-over, music and captions automated through Atlas Cloud and ffmpeg. Treat the first render as a draft, not a deliverable.
- Edit with capcut-edit: subtitles, timing, speed, volume, templates and animations in CapCut and JianYing projects, and cutting long-form video into shorts for each platform. Cut the shorts from the long version rather than rendering each one separately.
- Add the audio version with blog-audio. It generates MP3 narration of blog posts with Google Gemini text-to-speech, as summaries, full read-alouds or two-speaker podcast mode, with HTML5 embed code. A post people can listen to is a post people finish.
- Review with claude-real-video-for-agents. It installs the crv tool so the agent can watch a video by extracting deduplicated scene-aware keyframes and transcribing the audio. Have the agent check its own output before you do.
Install the stack
npx skills add https://github.com/coreyhaines31/marketingskills/tree/main/skills/image
npx skills add https://github.com/Bria-AI/bria-skill/tree/dev/skills/remove-background
npx skills add https://github.com/Alisa0808/vox-director
npx skills add https://github.com/renezander030/capcut-cli/tree/master/skills/capcut-edit
npx skills add https://github.com/infrasity-labs/dev-gtm-claude-skills/tree/main/writing-skills/blog-audio
npx skills add https://github.com/smixs/creative-director-skill/tree/main/creative-director
npx skills add https://github.com/HUANGCHIHHUNGLeo/claude-real-video/tree/master/skills/claude-real-video-for-agentsThe skills CLI works for Claude Code, Codex, Cursor and other agents, and each skill's page in the directory lists the copy-by-hand routes if you would rather not run npx.
What to check before you install
- Licenses: six of the seven are MIT; creative-director is CC-BY-4.0, which requires attribution.
- Services: image works with the image generator you pick, blog-audio uses Google Gemini text-to-speech, remove-background uses Bria RMBG 2.0, vox-director uses Atlas Cloud and ffmpeg, and capcut-edit works on CapCut and JianYing projects. That is several accounts and at least a few metered APIs, so budget for them and keep the keys out of the repository.
- Read each SKILL.md before you enable it. claude-real-video-for-agents installs a tool on your machine, and the directory has not security scanned any skill.
FAQ
Will I get a finished video from one prompt?
vox-director's summary says it automates the script, motion, voice-over, music and captions from one topic. Whether what comes out is finished is a judgment you make after watching it, which is why claude-real-video-for-agents and capcut-edit are in the stack: review, then cut.
Do these skills work in Codex and Cursor?
Yes. All seven use the Agent Skills format, so the same folder works in Claude Code, Codex, Cursor and any other agent that reads a SKILL.md. The skills CLI installs for each of them, and every skill page in the directory shows the commands.
What about music?
vox-director adds music as part of its automated pipeline. None of the seven picks composes music on its own, so if you need a specific track, bring it to the edit in capcut-edit, which handles volume and timing. image, at about 91,000 installs recorded by skills.sh, is the most installed pick here; the video skills have far fewer installs recorded, so read their SKILL.md files with that in mind.