Recent AI news · Original report: 2026-10-03 (KST)

New AI agent skills on GitHub: Live Photo maker, motion design, Codex plugin (Oct 3, 2026)

This article was compiled with AI. This English translation preserves the original report's collection date, findings, and limitations. Please consult the linked sources for current information. Originally collected: 2026-10-03 10:10 (KST, UTC+09:00).

Translated into English: 2026-10-03 (KST). The translation date does not indicate a new source check.

At a glance

We picked four AI agent tool repositories created between September 27 and 30, KST, and summarized what their READMEs say they do, along with the requirements and limits they state themselves. Two of them make videos; the other two help you manage agents or have them check each other. We did not install or run them.

Repository overview

These figures were retrieved from the GitHub API at 10:11 on October 3 (KST). The commit column is the date of the latest commit in KST.

livecanvas: one sentence to a Live Photo

The README describes it as a skill that turns a sentence into an Apple Live Photo with a designed cover and animation. It runs in coding agents such as Codex and Claude Code.

• It works in this order: research and fact-checking, storyboard, animation in Remotion, then previews and Live Photo files. It can use images, emoji, timelines and charts.
• The README shows 15 demo GIFs and notes that the numbers in them are a snapshot from when they were made and do not update.
• Requirements: Node.js, Python 3 and FFmpeg. Packaging a Live Photo also needs macOS and the Xcode Command Line Tools. Live Photo packaging currently supports silent video only, and the files are not imported into the Photos app automatically.

claude-motion-design: video from code, no editing app

The README describes it as a Claude Code skill that makes motion design videos using only HTML, the browser automation tool Playwright, and ffmpeg.

• Every scene is drawn by a single piece of code computed from time, then captured frame by frame and stitched together. The README says the same input always gives the same frames, so you can fix one scene without touching the rest.
• The example is a 24-second launch video at 60 frames per second, which the author says was made with Claude Opus 5.5 and this skill in five iterations. Sound effects are placed on the music's measured peaks.
• Notes: Music and sound effects are downloaded separately from free sources and are not redistributed in the repository. Example numbers shown on screen are labeled "Example data."

skillscout: skills for 8 agents on one screen

The README describes it as a Mac app that finds every skill your coding agents load, shows which agent can see each one, and counts how often you use them.

• It knows the skill folders of 8 agents, including Cursor, Claude Code, Codex and Gemini CLI. It can link a skill to agents that are missing it, and lists skills unused for 60 days and pairs of similar skills.
• It looks through your chat history for requests you keep repeating and drafts new skills from them.
• Requirements: macOS 15 or later. The README says the app is not signed with an Apple Developer ID or notarized, so macOS shows a warning the first time you open it.
• Data: When you use the AI features, up to 2,000 recent messages (each cut to 220 characters) go to OpenAI or Anthropic through your own Codex or Claude account, according to the README.

claude-codex-coop: Claude and Codex review each other

The README describes it as a plugin installed in both the Claude and Codex desktop apps. Whichever AI you are chatting with can call the other one to review, look for counterexamples, or implement.

• The AI you are chatting with forms its own view first, checks the other side's evidence point by point, then answers with what is confirmed and what is still in doubt. Each task allows at most two rounds.
• Analysis and review are read-only; only implementation changes files, within the scope you allow.
• Requirements: It is a preview release. The README says it has been tested on Windows 10/11, while macOS support is experimental. Python 3.10 or later is required.
• Data: The other AI receives a list of files in your workspace (up to 400) and description files, and files that look like passwords or keys are skipped, according to the README.

What we could not confirm

• Actual behavior and security: We did not install, run or security-review any of the four repositories. All counts, requirements and data handling come from what the authors wrote in their READMEs.
• Whether data handling matches the description: We did not test whether skillscout and claude-codex-coop send only what their READMEs describe. Both read chat history or workspace folders, so on a computer with company data, check your own policies first.
• What the star counts mean: All four repositories are less than a week old, so star counts can change a lot within a day.
• Repositories left out: blueprint-animation, which had the most stars this week, and the video skill list awesome-claude-video-skills were left out because their licenses are unclear.

Sources

• GitHub, pengchujin/livecanvas https://github.com/pengchujin/livecanvas
• livecanvas English README https://github.com/pengchujin/livecanvas/blob/main/README.en.md
• GitHub, howseen-ai/claude-motion-design https://github.com/howseen-ai/claude-motion-design
• GitHub, flaviocopes/skillscout https://github.com/flaviocopes/skillscout
• GitHub, youagainchen/claude-codex-coop https://github.com/youagainchen/claude-codex-coop