Fifteen skills, four of which clone someone else's repository, and no licence file at all

The first line says the project is unmaintained, in Chinese

The readme opens with an important notice, and it is written in Chinese: the project is no longer maintained and readers should move to a separate site. There is a language switcher at the top right offering an English version, and both files are in the tree, so an English reader has a choice about which one they read.

The repository's own archive flag is not set, so nothing on the hosting side prevents an install, and the last commit is dated 2026-09-28, recent enough that the notice is a decision rather than an abandonment. The rest of the readme is also Chinese, which matters more here than it would in a normal project. These are agent skills, and skills are usually matched by trigger keywords, so a skill whose triggers are Chinese will not be selected by an agent working from an English request no matter which readme you read.

The destination in the notice is not the same domain as anything else the repository points at, and the readme separately sends you to the author's profile page for contact, so the project is being steered off this repository entirely.

No licence file, and two skills credit other authors

The repository has no licence file. The licence metadata reports nothing at all, which is not the same as a permissive licence and not the same as no licence by accident; it means nothing is declared anywhere in the tree.

That sits next to two attributions inside individual skill descriptions. The interactive line-curtain skill says its engine is adapted from another author's dynamic ropes project and is used under a Creative Commons share-alike licence. The 3D Chladni skill credits its preview media to another author under a Creative Commons non-commercial licence.

So the collection contains derived work credited under share-alike and non-commercial terms, and nothing tells you what governs the collection itself or whether the two can be combined and redistributed. That question has a straightforward answer in principle and needs the actual licence texts to settle, and neither text is in this repository.

The contribution section makes the reason this matters concrete. Every preview animation has to be committed as a file, with a hard ceiling and a recommended ceiling, so the repository is a distribution channel for third-party imagery whether or not it intended to be.

Four of the fifteen skills clone a repository before rendering

Most of these skills describe generating an output directly: a ruler progress animation, a typing animation, a procedural fish, a spotlight reveal, a 3D photo ticker, a vinyl player. Four do not. Their instructions are to clone or update a separate GitHub repository and then render with it.

Two of those four point at the author's own repositories, one for the fisheye distortion effect and one for a 3D earth animation. The other two point elsewhere: one clones a third party's 3D Chladni figures project and uses its own deterministic exporter to produce video, and one clones a third party's WeChat chat project to render chat animations.

The difference between cloning your own project and cloning someone else's is the whole question here. In the first case a maintainer can break an upstream change and fix it in the same repository. In the second case an agent will fetch code from a repository that has no relationship to this one, and the skill text is what instructs it to do so.

That is also why the Chladni skill insists on reproducible seeds and a deterministic exporter, which suggests its author knew the render needed to be repeatable across machines.

Four rendering stacks under a Python language tag

The fifteen skills are not one technology. Several produce video through a component-based framework, including the candlestick chart, the 3D ticker, the vinyl player, the fisheye effect and the WeChat animation. Two produce interactive HTML instead, one of them a line-curtain simulation with a live physics panel, and one a spotlight reveal with configurable amplitude, lampshade scale and glow.

One drives a 3D earth scene through a headless browser to produce GIF and MP4 exports. One shells out to a foreign project's deterministic exporter and offers video or transparent video output. One traces a raster logo into an animatable vector and then produces brand motion, an interactive page, previews and a motion QA evidence set, which is the most pipeline-shaped skill in the collection.

Against all of that, the repository's detected primary language is Python. That is a strange reading for a collection of browser rendering pipelines, and it says something about how little code the repository actually holds: fifteen directories of skill definitions, one readme in two languages, and a gitignore.

Installation is interactive and asks which agent you are installing into

There is one install command, and it is not the usual non-interactive package install:

npx skills add vibe-motion/skills

The note under it explains the rest. It is an interactive script, you select which skills to install with the space bar, it recommends installing all of them, and you also have to choose the corresponding agent, with Claude Code given as the example. The reason given for that last step is that different agents store skills in different paths.

That is a real design constraint and it is worth respecting rather than working around. An agent skill is not a library import; it is a set of instructions dropped into a directory the agent reads, and the directory is agent-specific. It also means the installer is a script from a package registry that will write into your agent configuration, which is a higher-consequence action than installing a dependency into a project.

Two of the skills are named after their own tooling, one for typing animations in a specific command line agent's style and one for the animation principles taught in a film course, so the collection is not a coherent toolkit but a grab bag with a shared installer.

Two skills reach the network for their input

The candlestick skill fetches market data from a commercial finance site and draws it per frame onto a canvas in a dark terminal style. That is a live third-party data source being pulled into a render, which means the output depends on the state of someone else's site at render time, and the terms that govern that data are not discussed here.

The brand video skill starts from a company website or official brand materials, and it is the most prescriptive in the collection. It verifies the logo, fonts, colours, images and product information first, then designs a narrative, a frame-by-frame timeline and a real product demonstration. Defaults are stated as English, sixteen by nine, fifteen seconds, with custom duration and aspect ratio supported, and the stated range is fifteen to thirty seconds.

Its example credit lines name two real companies, a fast food brand and an AI lab, as works produced with the skill. That is a claim of prior work with identifiable brands attached to it, and anyone reusing the skill should know the names are already on the page.

A one megabyte ceiling on every committed preview

The contribution section is four sentences and all of them are about file size. Before opening a pull request, compress every animation. Each one must be under one megabyte, defined precisely as one million bytes, and the guidance is to aim for under a quarter of that, two hundred and fifty thousand bytes.

The method given is to reduce preview resolution, frame rate and palette size. The acceptance check afterwards is that the full duration, the looping behaviour and any key text or subject must still be legible.

Those numbers explain a lot about the shape of the repository. Fifteen skills each need a preview, previews are binaries, binaries are committed rather than linked, and a quarter-megabyte budget is what that produces: short loops at reduced resolution and few colours. It is a sensible rule for a repository of visual work and a direct constraint on anyone who wants to add a longer or higher-fidelity demo.

One entry in the readme is explicitly not a skill

The final entry in the collection is not part of it. A boids fish-school simulation built on a 3D library is included under a miscellaneous heading with an explicit note that it is not a skill and is shown only as a standalone project, with its address pointing at a different repository in the same account.

Everything else in the readme is a skill, and the count is consistent: fifteen skill descriptions and fifteen skill directories in the tree. That internal consistency is worth noting because this is a small, hand-maintained collection rather than a generated index.

What the readme does not contain is any statement about versions, compatibility, or which skills depend on which other skills. There is no changelog, no release history and no support statement beyond the notice at the top. For a collection where several entries clone external repositories at render time, the absence of any pinning or version guidance is the gap a reader is going to hit.

Editorial conclusion

Treat this as a directory of techniques rather than as a library, which is what an agent skill collection is. Two things decide whether any of it is usable for you. The first is the deprecation: the notice at the top of the Chinese readme says the project is no longer maintained and sends you to a separate site, so the installer and the skill directories may not be there in the same form later. The second is provenance. Four of the fifteen skills do not generate anything on their own, they clone or update a checkout of another repository first, and two of those four are repositories this project does not control. Before you adopt any skill that pulls an external repository at render time, read what it will execute, because the skill text is what tells your agent to run it. The missing licence file is a separate and simpler problem: with attribution to other authors inside the descriptions and nothing covering the collection itself, redistribution rights are not stated anywhere.

Frequently asked questions

Official sources

1. Issues
2. README
3. vibe-motion/skills on GitHub