Lobster is quite popular, but it doesn’t produce any new technology itself. It simply provides mainstream IM tools with the capability to leverage AI. So I thought it could also solve the daily publishing workflow for social media creators.
Existing Skills
First, let me introduce some existing skills. You can directly ask your AI to install these, and you can ask the AI for usage instructions.
- https://github.com/JimLiu/baoyu-skills/tree/main/skills Explanation: From Baoyu’s skills, I currently only use baoyu-cover-image (for generating article cover image prompts) and baoyu-post-to-wechat (for solving WeChat Official Account publishing issues).
For other platforms like Toutiao and Xiaohongshu, I’m currently developing them myself because they’re highly customized, so I won’t share them for now.
Principles of Social Media Publishing
For platforms like Xiaohongshu and Toutiao, since they have PC web interfaces, you can use Playwright to simulate browser operations. Of course, you could also use API interfaces, but they’re too easily blocked. I personally recommend directly manipulating the web pages.
For WeChat Official Account, since there’s an official API, the aforementioned Skill recommends using the official API method for publishing, as it’s more stable. Of course, Baoyu’s skill is compatible with both API and non-API approaches. Without an API, it naturally falls back to browser page operations.
Image Generation
Compared to GPT and Gemini, Gemini is much more powerful. Chinese character rendering is now good without obvious issues. Since API pricing is a threshold issue and costs are a concern, I’m using the web-based solution directly. That is, using Playwright to simulate browser operations, generating images directly on the web page, and downloading them locally. This way, you can continue with image-text publishing.
Writing Skills
Although there are some solutions online, whether for security or customization reasons, you always end up needing to modify them. So how to write efficiently? Here’s my approach: I use the Chrome Claude plugin to write directly on the target page. This way, Claude helps generate the rough code and also verifies its effectiveness. Then I can copy the basic logic and have OpenClaw re-encapsulate it.
Final Thoughts
So far, the most critical image generation and publishing are sorted out. This actually saves considerable time, and it also makes it easy to do multi-platform simultaneous publishing. So, let’s get to it!
