Confusion with code-testing-agent

Something in the metadata seems to be confusing agents into thinking the code-testing-agent skill is the name of another agent they can call. I have had this happen a few times in VS Code, where instead of loading the skill, they try to run it as an agent instead. It seems to be particularly bad with GPT-5.6 Luna, but other models also have the same problem sometimes. Heck, when I first installed the plugin, I thought that was an agent name and was confused as to why I couldn't select it in the agent list.