v1.0.1 and Claude Code

The original v1.0 was released on 5/22, and v1.0.1 on 5/25. There aren’t really any users yet—these were just some issues I ran into while using it myself, which I fixed in v1.0.1. For example, when typing in the input box, the cursor wouldn’t scroll automatically, so when entering a lot of text, you couldn’t see the last characters you’d typed.

By the way, I develop with Claude Code. Generally, when using Claude Code to build desktop applications (not web apps), you run into one difficulty: it’s hard to do automated verification. Right now, my approach is to create a probe test and verify using screenshots.

The method is pretty simple. As an example, the prompt would go something like this:

Create a probe test to run automated verification. First explore the codebase, then based on the login screen → display schedule use case, write Python to control mouse operations, with the goal of observing an empty schedule window at the end.

There’s one prerequisite, though: my codebase is layered according to clean architecture, so there’s a use case layer. This makes it easier for Claude to understand the code related to use cases.

That said, in practice the efficiency isn’t great, because taking and reading screenshots takes quite a bit of time. In terms of speed, it takes more than twice as long as a human would. But the upside is that once you’ve set it up, you can persist it into a proper test case for the future.

This kind of probe test works fairly well for clear-cut use case testing—it can indeed verify whether behavior has been broken. But it’s more troublesome if you want to use it to trace an issue, debug, or test performance. For example, there was a bug where the window would flicker briefly, about 0.1s. That’s hard to capture in a probe test screenshot, so I gave up and just verified it manually.