I was curious how OpenClaw skills are built, so I gave it a try. After asking Gemini, I learned that the basic approach is to define the behavior in a SKILL markdown file. If needed, you can also include a simple script alongside it. Pretty straightforward.
While thinking about what to build, I decided to add a skill that sends messages to the company’s Dooray messenger via webhook. That way, N.I.C.K. would be able to contact me through our company messenger.
I had N.I.C.K. generate this markdown file as well. From defining the requirements to generating the file, it took less than 20 minutes. It even created a Python script on its own. I suggested using a venv if you need to install additional packages, but it replied that it would stick to the Python standard library, so it wasn’t necessary. Fair enough.
What actually took longer was getting the skill to be installed properly. Perhaps the path was tangled—N.I.C.K. couldn’t resolve the issue on its own. Gemini wasn’t much help either, so I eventually dug into the official documents and fixed it. Since the platform is evolving day by day, it seems the information the AI was referencing was already somewhat outdated.
Once you configure the channel name and its hook URL in the config file, you can send messages with a command like the following;
Send the message "Hello?" to the Dooray channel "Kirin"I’ve also uploaded it to ClawHub. At that point, I thought I was done—but I didn’t expect I should make an updated version just one day later. The day after uploading, I checked the listing and found it had been flagged as “Risky” in the security review. Wait… why??
I handed the ClawHub link to Gemini and asked what was wrong, and it pointed out a few issues. Looking back, I probably should have asked N.I.C.K. to handle it instead. Now that I better understand the structure, I’ll likely do that next time I build another skill.
Anyway, it’s now properly uploaded—please check it out.
Kirin's Technology Essay 