fix(metrics): CPU usage shown inverted on Linux (~99% when idle)

[flavioduque]

Problem

On Linux the dashboard reports ~99% CPU while the host is idle.

getCpuUsage() runs:

top -l 1 -n 0 | grep 'CPU usage' || mpstat 1 1 | tail -1 | awk '{print 100 - $NF}'

top -l is macOS-only, so on Linux the mpstat branch runs. That awk already prints usage (100 - %idle), but the code treats the number as idle and computes 100 - value again. The result is inverted: an idle server shows ~99%, a busy one shows a low value.

Also, when measurement throws, the route returned Math.floor(Math.random() * 30) + 10, which is a made-up number shown as real data.

Fix

• The mpstat branch now prints %idle (awk '{print $NF}'), which is what the parsing code expects. The macOS path is unchanged.
• If the command fails, fall back to the existing os.cpus() calculation instead of a random value.

Verification

Ran inside the running panel container (Alpine, BusyBox top, sysstat mpstat) with the command string read straight from route.ts and the same parsing:

	command output	panel shows	real idle (mpstat)
before	1.5	99%	97.24%
after	99.37	1%	99.00%

Not covered by an automated test: the function shells out to mpstat/top. I can extract the parsing into a pure helper with a tests/*.test.ts case if you prefer.

🤖 Generated with Claude Code