EC2 Klaus had a CPU problem. Ollama and the gateway process were fighting for cycles on a 2-vCPU box. It took some work to sort out.
So when HP Klaus started acting up, I described it the natural way: “I’m having similar issues with HP Klaus as we were having with EC2 Klaus.”
That sentence did a lot of work before I noticed.
Klaus started establishing access. Tailscale into EC2, then a hop onward to HP’s local address. I watched it go for a few minutes while it built a theory around the last incident: resource contention, Ollama competing with the gateway process. That was the frame before a single log line had been read.
Then I pasted the actual error.
openclaw gateway start returning MODULE_NOT_FOUND for openclaw.mjs.
The actual problem was a missing file. The gateway process wasn’t competing for resources; it wasn’t starting at all. Ten minutes of access setup and a diagnostic posture built for the wrong problem, to find a corrupted npm package.
The analogy had picked the investigation before any evidence arrived. “Similar to X” is how the symptom felt to me, not a description of the cause. The cause was completely different. With the literal error in the first message, we’d have skipped most of the access setup and started with the right question.
I would have opened the terminal, grabbed the literal error, and led with that.