After using Open Claw for over a month, I thought I was training the AI, but actually it was training me. I am now increasingly learning how to work with it. Here are my experiences from this month:

1. When many things get stuck, don't just blindly ask it to fix various bugs. Instead, have it carefully read the official documentation. AI often doesn't read the docs and just fantasizes about bugs, then fixes them to death. After reading the documentation, many bugs actually don't exist.
2. It only remembers the recent two days of conversation and a long-term memory document with summaries. So, even if you tell it something clearly, it might forget after a few days. The best way is for you to keep a summary of the key points yourself. Whenever you have a major breakthrough with it, ask it to update its memory system so that its long-term memory gets longer. Don't expect it to remember all your conversations. It will basically forget everything after two days.
3. When you find that Open Claw can't handle something, doesn't respond, or encounters various issues, directly ask your Claude code what to do. It can usually help you solve the problem directly. So, to use Open Claw well, remember to run Claude code in parallel and use them together.
4. You install various skills on it, thinking that it will have that skill now? It will use it on its own in the future? No. Some skills often require deployment, API access, or path setup after installation. Sometimes, if you don't ask your AI, it won't do any of these for you. It always fantasizes that it has the skill, but in reality, it never truly uses it.