Machine Monitor

machine-monitor

A [Claude Code](https://claude.com/claude-code) skill — and a standalone CLI — that renders a live, htop-style dashboard of a macOS machine: real die temperature, honest memory accounting, live per-process CPU, every listening port mapped to the project it came from, Docker, disk, battery, and an automatic WindowServer diagnosis when compositing goes hot.

Built for Apple Silicon, where the usual answers are wrong. Zero dependencies — system `python3`.


Generated, not recorded — the machine above is fictional, so the frame shows the tool rather than whatever the author happened to be running. `python3 tests/ansi_to_gif.py`.

Why it exists

Three things about macOS make "why is my Mac hot / slow" harder than it should be, and the obvious answer is wrong in all three.

• ps aux's %CPU is a lifetime average since the process started, not current load. A dev server that has been up for four days shows a low, flat number while pinning a core right now. This tool differences each process's CPU time over a real sample window, which is what top does internally.
• "Memory used" is usually computed as active + wired, which ignores compressed memory — the single largest category on a machine under pressure. Measured on the development machine: active + wired read 5.7 GB while the true figure including the compressor was 12.3 GB of 16 GB, with swap 88% full. The naive number said there was plenty of headroom while the machine was thrashing.
• WindowServer at high CPU gets guessed at. ProMotion, video decode, screen recording, transparency and display count are all plausible, and people pick one. This runs a live check for each and labels them ← CAUSE, ← FACTOR, or ✓.

On temperature

Earlier versions of this tool stated that Apple Silicon exposes no reliable die temperature and deliberately printed none. **That was wrong**, and it is worth