Recent AI news · Original report: 2026-10-09 (KST)
New AI agent skills on GitHub: AI avatar from image, podcast to article, handouts (Oct 9, 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: 2026-10-09 10:10 (KST, UTC+09:00).
Translated into English: 2026-10-09 (KST). The translation date does not indicate a new source check.
In one line
A README-based look at four AI agent tool repositories created between October 4 and 7, Korea time. We did not install or run any of them.
The repositories at a glance
Figures are from the GitHub API at 10:10 a.m. on October 9, Korea time. "Commit" is the date of the latest commit in Korea time.
| Repository | What it does | Stars | Commit | License |
|---|---|---|---|---|
| mesh-avatar-studio | One illustration into a moving avatar | 507 | Oct 8 | MIT |
| mortiflix-oss | Video production approved stage by stage | 466 | Oct 9 | AGPL-3.0 |
| handout-remake | Class materials into tablet study handouts | 146 | Oct 7 | MIT |
| podcast-to-article | Podcast transcripts into long articles | 119 | Oct 8 | MIT |
mesh-avatar-studio: one illustration becomes a blinking avatar
The README says it turns a single illustration into a 2D avatar that blinks, talks and turns its head. A coding agent (Claude Code or Codex) prepares the avatar, and a person fine-tunes it by dragging dots in an editor.
- It has a feature that makes the avatar follow your face through a camera, and the README says camera video and microphone audio stay on the device.
- Requirements and limits: Node.js 22.17 or later is required. The README states that Codex draws the closed eyes and mouth shapes, and that Claude Code cannot generate images.
- Rights: The images of the sample character Miko are not covered by the MIT License and may not be redistributed on their own.
mortiflix-oss: video production where a person approves every stage
The README explains that a video is made in order (brief, script, style frames, animatic, final) and that the work only moves on once a person approves each stage.
- It can make four things: explainers (30 seconds to 2 minutes), vertical shorts (45 seconds to 3 minutes), logo animations (3 to 8 seconds) and instrumental songs.
- Requirements and cost: Node 20 or later and ffmpeg are required. Real work runs on the user's own Claude plan or API key.
- Limits it states itself: It has been tested mostly on Linux and not yet on macOS. A Windows version is promised by October 20.
handout-remake: class materials into tablet study handouts
The README says that when you give it lecture slides, PDFs or notes, the AI re-teaches the material in an order that suits a beginner and builds a PDF. The page size matches an 11-inch iPad held upright.
- One handout covers a single section, and 6 to 12 slides become 12 to 20 pages. Answers to practice questions sit on the very next page.
- Requirements: Output is in Traditional Chinese (Taiwan). Running it on your own computer needs Node.js 18 or later, Playwright with Chromium, and poppler.
- Cautions: The README says the AI can be wrong, so the textbook and the teacher come first, and that handouts should be for personal study only out of respect for the copyright in class materials.
podcast-to-article: podcast transcripts into something readable
The README describes a skill that turns the transcript of a podcast episode into a long article ready to publish. It rebuilds a transcript ordered by speaker and time into one ordered by topic and judgment.
- In the author's own example, a draft of 5,730 Chinese characters became 3,703, a 35% cut, and paragraphs went from 57 to 47.
- In an example with an 86-minute English interview, the first draft had 0 direct quotes, and after revisions the final piece was 2,573 characters with 18 direct quotes (23%), the README says.
- Requirements: It uses only the Python 3 standard library.
What we could not confirm
- Real behavior and security: We did not install, run or security-review the code of any of the four repositories. All figures and requirements are what the READMEs say.
- Korean results: The output of handout-remake and the examples for podcast-to-article are in Chinese. The READMEs do not say how they handle Korean material.
- What the star counts mean: All four are less than a week old, so star counts can change a lot even within a day.
Sources
- GitHub, shinshin86/mesh-avatar-studio https://github.com/shinshin86/mesh-avatar-studio
- GitHub, GTKottman/mortiflix-oss https://github.com/GTKottman/mortiflix-oss
- GitHub, xamjiang/handout-remake https://github.com/xamjiang/handout-remake
- GitHub, orange2ai/podcast-to-article https://github.com/orange2ai/podcast-to-article