TW was developed to be hackable by non developers. This means things like filters, macros, etc. that are more declarative. The downside is that they are not standard.
Today, anyone with an agent can be a developer. But now using agents to develop TW is a problem, since the models didn’t encounter a lot of TW examples while training compared to the mountains of examples of Javascript, React, etc.
It would be good if there was a way to build custom views in a standard way. Maybe React components and some JS API to filter tiddlers (or maybe SQL). Hacking TW will then become much easier for the agents, and as a side benefit will allow tapping to the eco system of these technologies.
Hi, can you elaborate on how to do that?
You mean clone the GitHub repository and let the AI read the folder structure?
I guess another approach is to feed the AI with tiddlers of https://tiddlywiki.com/ because most of them explain with examples how to use wikitext, filters, and so on
Yes. If you have a local development environment and start a LLM CLI from there, it has full access to all that info. It works quite well. The only problem is, that you have to use a relatively new and powerfull version. They usually have good knowledge about TW source code and wikitext and filters.