It is February 9, 2026, in the afternoon.
I'm feeling a bit unwell again today. When I had a cold a few days ago, I should have taken medicine promptly. As a result, I've been not only tired and easily fatigued these past few days. The reason I didn't take the cold medicine at the time was that I thought it would make me sleepy and waste time. The consequence of not cutting my losses was even more severe—I lost this bet.
Getting back on track, yesterday I used CZON + Opus 4.6 to create a version of the blog summary. Some links in it were broken, which ultimately came down to a bug in my prompt—I didn't give the correct instructions.
However, I believe we can't expect AI to follow all your instructions perfectly in one go. This is a physical characteristic of AI's non-mandatory coordination.
At first, I wondered whether CZON should support absolute path references relative to the project root directory. But later, I felt it still wasn't quite appropriate because I've already decided that the .czon/src/<lang>/ directory is also a fully legitimate fractal directory. Every reference between files within it should be completely valid, and images and hyperlinks inside should work properly.
Therefore, CZON should comprehensively reject the use of absolute path references relative to the project root directory and insist on using relative references. Although VSCode can support this referencing method, I think it's not standardized enough to precisely express the semantics of the reference. It's an unreliable referencing method and should be abandoned.
This rejection will bring some compatibility issues. For example, you might find that links in some older documents are broken. However, a relatively good solution is for CZON to provide a check command specifically designed to verify whether links in documents are valid. If invalid links are found, it can issue warnings, provide suggestions for fixes, or even perform automatic fixes (via the --fix option). The fix suggestions can be directly fed to an AI Agent for execution. After fixing, simply run the build command again.
Additionally, when running long runOpenCode tasks, I still encounter 600-second timeout errors. I've already used the short method of promptAsync + polling status, but if a task runs for more than 10 minutes, it still reports a 600-second timeout. I think this is an issue with OpenCode itself, and I will try to solve this problem next.
Furthermore, I think I need a TODO Summary feature to help extract all the TODOs from my blogs and generate a TODO List document for easy centralized viewing. This is an interesting feature point. I'll have AI read all the blog content, determine whether my TODOs are completed, and then I can check this TODO List daily to see what I should be working on.
Another interesting thought: since AI can already examine the value ranking in my mind, couldn't it also automatically help me prioritize these TODOs? Wouldn't that be wonderful?
