Came across @op7418's post and took the time to compile all the Claude Code / Codex Skills he has open-sourced.

First, the numbers:

- Humanizer-zh: 5600🌟

- NanoBanana-PPT-Skills: 2100🌟

- Claude-to-IM-skill: 1800🌟

- Youtube-clipper-skill: 1700🌟

- Document-illustrator-skill: 300🌟

- Video-Wrapper-Skills: 200🌟

Feedback from several friends who have used them: the IM bridging skill is the real deal. Once installed, you can directly interact with Claude Code or Codex within Telegram, Discord, Lark, QQ, and WeChat. Someone asked in the comments, 'What's the difference from the official Telegram integration?' In short: the official one uses a Web API for demonstration, while this one directly integrates the AI into your IM. Tool calls, file reading/writing all work, and permission confirmations are handled right in the chat window. Some people run code remotely by sending commands via WeChat—very practical.

The Youtube-clipper skill, with 1700 stars, didn't get them for nothing. It's not a simple, brute-force tool that splits videos by time. Claude first analyzes the subtitle content, understands the semantics, and then generates meaningful chapter divisions (typically 2-5 minutes per segment), letting you choose which clips you want. Translation is also batch-processed, handling 20 sentences at once, reducing API calls by 95% compared to translating line by line. The generated subtitles are bilingual (Chinese-English), with customizable font positioning, styles, and can even be burned directly into the video.

I've used the Video-Wrapper-Skills when making videos. It automatically adds effects similar to those in 'Xiao Lin Shuo'—highlight cards, keyword explanations, guest info bars, chapter titles, quote effects, etc. It supports four visual themes: Notion style, Cyberpunk style, Apple style, and Aurora gradient style. For rendering, you can use Playwright+HTML+Anime.js for high-quality animations, or pure Python PIL for faster speed.

The Humanizer-zh skill with 5600 stars is truly outrageous. You connect it after Claude's output, and it helps remove the 'AI-generated tone.' The approach involves deconstructing the content first, then reorganizing it—a two-stage process. The resulting text feels more like it was written by a human, avoiding the clichéd openings like 'First,' 'Secondly,' 'In summary.'

The Document-illustrator-skill automatically analyzes documents to generate accompanying images, supporting both 16:9 and 4:3 ratios. The cover images I often use are generated by this tool.

NanoBanana-PPT-Skills is a tool for AI-generated PPT images and videos, supporting smart transitions and interactive playback.

Someone replied on Twitter saying, 'Master Cang hits the pain points with precision every time he open-sources something, with quality and practicality maxed out.' That's not an exaggeration. The common trait of these tools is: install and use, no fluff.

If you're also a Claude Code or Codex user, these repositories are worth checking out.