I previously shared my OpenClaw mechanism. Today, on my way home from work, I happened to use OpenClaw on my phone to complete a system mechanism optimization development, from fixing a bug to refactoring a service, and establishing a global OpenClaw mechanism. The vibe coding process, with screenshots throughout, shows the process almost seamlessly. A bit of background: I’ve been using OpenClaw to build various capabilities, so many services that use model capabilities are either called via CLI (main Agent calling command line) or hardcoded in scripts. As more scenarios arose, I hit a snag yesterday. Due to a lack of control over model usage and calls, my OpenClaw became unusable, and because of a retry death loop requesting large model services, I lost about tens of dollars. Today was full of meetings. After patching the vulnerability in the morning, I started fixing this issue on my way home from work. The following is essentially the entire conversation, and you can see the conversation times, which is quite interesting. Whether you’re a novice or looking to learn vibe coding, consider this a case study, and a callback to my previous article.
Finally, almost home, casually sending a ✌️(＾▽＾)✌️