Ten animation skills, one folder contract, and a release script with no target

Ten skills, one contract: SKILL.md, references, assets

The collection is a set of folders under skills/, one per animation type, and every folder follows the same four-part layout: SKILL.md with the instructions the agent executes, README.md written for a person, references/ holding prompt references, and assets/ holding template HTML. The result is always the same shape. The agent reads a description, activates the matching skill, and emits one complete HTML file. The overview names ten of them: ppt-animation, flowchart, network-protocol-viz, dynamic-archify, scholar-notes, card-theater, video-shot-demos, phone-ui-demos, stacked-data-cards and win11-ui-demos. No bundler, no framework dependency and no build step appear anywhere in that layout. The point of assets/ is that the output is a file you can open, record or embed, which is what lets a skill folder sit inside an agent's skill directory without dragging a toolchain along.

The release script points at a directory the tree does not contain

The root package.json defines one script and nothing else:

  "scripts": {
    "release": "node scripts/release/update-readme.mjs"
  },

The path it calls, scripts/release/update-readme.mjs, is not among the top-level entries, which are .gitignore, CONTRIBUTING.md, LICENSE, README.md, README_EN.md, UI/, original-prompt.txt, package.json, prompt.md, skills/ and web_animation/. So the one command the project defines has no target in the published layout. The same file sets private to true, which keeps the package out of any registry, and pins the version at 0.1.0. Together with the absence of GitHub releases, that leaves no version to track a skill folder against: adding or editing a skill is a commit, not a release.

Prompts by another name, with the earlier files still at the root

The one-line summary shown for the project calls its contents a set of AI prompts for generating animated HTML pages, covering animation effects, 3D visualization, PPT-style presentation and UI beautification. The overview file at the root describes something narrower and more structured: a set of Agent Skills for WorkBuddy, Claude Code and Cursor, with the prompt material split into SKILL.md, references/ and assets/. Two root files, original-prompt.txt and prompt.md, sit next to that structure, which is what the older prompt-collection framing looks like after it has been folded into skills/. The primary language recorded for the repository is HTML rather than Markdown, and the default branch is master rather than main, so any script that assumes either default has to name the branch itself.

Three skills get prose, seven get a table row

The overview opens with a summary line for each of the ten skills, and those lines are where the specific claims live. video-shot-demos is credited with 29 rotating styles, camera moves, WebAudio sound effects and character reactions. phone-ui-demos stages a lock screen, chat, settings and app views on a 3D phone. stacked-data-cards cascades argument cards on a millisecond timeline. win11-ui-demos reproduces a file explorer, start menu, system dialogs and a blue screen repair, recorded full screen. The gallery that follows goes into detail for only three skills: ppt-animation with its theme names and preview table, flowchart with its chart types and interaction modes, and network-protocol-viz with protocol coverage and control requirements. The remaining seven appear in the gallery table with a category and a use case and no further prose. The visible gallery ends inside the network-protocol-viz preview table, right after the row for HTTPS application layer encryption.

network-protocol-viz makes play, pause and reset mandatory

This is the one skill whose overview states rules rather than adjectives. A packet or frame is drawn as layered rectangles with hover explanations per field; an exchange is drawn as a client and server sequence diagram with packets moving between them; a routing story is drawn as a topology with an animated path. Play, pause and reset controls are required in every output. Layers keep fixed colors: gray physical, blue link, green network, orange transport, purple application. Packets travel with a glowing trail, and a security-education mode appends a disclaimer at the bottom of the page. Covered protocols include TCP/IP, IPv4, Ethernet frames, routing, switches, DHCP, HTTPS and PPP. Requiring the controls and offering the disclaimer mode is a different instinct from the usual demo page, where the animation just runs.

flowchart and dynamic-archify are positioned against each other

The overview states that two of the ten are complements rather than alternatives. flowchart leans educational: dark background, glowing nodes, arrows that flow, data particles, and a style built for explaining. It covers seven chart types, flowchart, concept diagram, principle demo, sequence diagram, comparison, timeline and system overview, and it keeps AI and machine learning diagrams intact, naming RNN, LSTM, GRU, MLP, Word2Vec and GPU one by one. dynamic-archify leans engineering: system architecture, flow, sequence, data flow and state machines, with flowing animation plus export to multiple formats. The three interaction modes, autoplay, manual stepping and hover highlighting, belong to flowchart. The overview credits dynamic-archify with export but never says which formats, so an author who needs the diagram in a specific file type has to find that out elsewhere.

A Chinese overview, a separate English file, and no releases

The overview at the root is written in Chinese, and README_EN.md sits beside it as the English version, so the two are maintained as separate files rather than generated from one source. Licensing is unambiguous, MIT in the root LICENSE and MIT again in package.json, with the author field naming UncleC. The repository has 1,426 stars and two open issues. The last push landed on 2026-10-01, and there are no GitHub releases, so the promise of ten skills can change without a version marker. For someone copying a single skill folder into a project, the check that matters is whether assets/ carries the template that skill needs, because the folders are self-contained and nothing in the collection pins a version of anything else.

Editorial conclusion

Judge AI_Animation as a prompt library that happens to be packaged as agent skills. The folder contract is the useful part: SKILL.md for the machine, references/ for the raw material, assets/ for the template, and an output that is always one HTML file you can open or record. What a reader cannot rely on is versioning. There are no GitHub releases, the root package is marked private at version 0.1.0, and the single script it defines points at a path the repository does not contain. Copy an individual skill folder when you want one animation style, and check that assets/ holds the template you need; treat the rest of the collection as a set of examples rather than a dependency. For anything on a production schedule, the open items are the install path, the export formats behind dynamic-archify, and the promises the summary makes that the detailed entries never repeat.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Unclecheng-li/AI_Animation on GitHub