Stats

CPU, memory, disk, network and load as bar items, each its own item that stays off the strip until it has something to say, with a popover of the breakdown; the same facts as rows in a Stats palette; pal://stats/<item>.

Sources and what they cost

One sampler runs every interval seconds (3 by default, from the first render on) and pushes all five items with bar.update; the core diffs, so a strip that did not change costs nothing past the check. Measured on an M5 Max (18 cores, ~1100 processes), per call:

So a tick is about 5 ms every 3 s with nothing open, 27 ms with a process popover up. No root anywhere. No temperature: on macOS the sensors need a native reader (IOKit's HID sensor services on Apple silicon, the SMC on Intel), which the host has none of, and powermetrics needs root; the Battery extension shows a temperature when the optional power watcher's state file carries one. On Linux /sys/class/thermal would do without root; not read yet.

What Linux lacks: the memory segments are two (used and shared) rather than Activity Monitor's four, and there is no compressed figure; the startup volume is named /; the Wi-Fi interface is told apart from a cable only where nmcli answers the core's wifi.status().

The items

Each item renders its facts as states (below) and leaves presence and colour to the manifest's rules, which Settings > Bar shows per item and the user edits by id ([bar.items."stats/cpu".rules.quiet] when = "..."):

Two ways to keep an item on the strip at rest: [bar.items."stats/cpu"] show = "always" draws the same figure muted (the core's empty shape), or override the quiet rule (when = "false", or turn Hidden off in Settings

Bar) and it stays in colour. enabled = false takes it off entirely. An icon-only item is the core's show_title = false.

The popovers (view.ts, laid out for the 420 px popover), each with a sparkline of the last 60 samples as an SVG image in the theme's tag ink (the theme state picks light or dark; a shade that reads on both when it is unknown) and a keycap hint row; up/down (or j/k) move the ring, a click on a row moves it too, s opens the Stats palette:

• CPU: the share, the level badge, cores / load / uptime, the sparkline, one bar per core in two columns, the five busiest processes (cpu and memory columns); Enter Activity Monitor (the Processes palette on Linux), x kills the focused one (asks first), c copies the share, p the Processes palette.
• Memory: the share, the pressure badge, used of total and swap, the segments bar (app blue, wired violet, compressed amber, cached grey, free outlined) with a legend, the sparkline, the five largest processes; the same keys.
• Disk: the free space on the focused volume, then a card per volume (name, a read-only badge for a mounted image, free space in the level's colour, a bar, the mount point, used of total); Enter reveals it in Finder (open on the mount point), c copies the path.
• Network: the two rates large, a busy / active / quiet badge, both lines in one sparkline, then every interface that counts, has an address or moved a byte (the Apple-private awdl/anpi/llw ones only when they did), with its kind (Wi-Fi with the SSID from the core's wifi.status(), VPN for a tunnel), address and rates; Enter opens the Network palette (every address, the public IP), c copies the focused address. The totals count the physical links only (en*, eth*, wl*, ...), so a VPN's bytes are not counted twice.
• Load: the three averages as tiles coloured by their share of the cores, the per-core figure, the sparkline scaled to the core count.

The palette

Stats is live (listed on every show from the last sample, no tool runs; ⌘R samples now). Sections: Processor (CPU, load), Memory (memory, swap), Disks (a row per volume, free as the name), Network (a row per interface, then the totals), Busiest processes, Largest processes (five each), System (uptime). The value is the row's name and Enter copies it (a volume's mount point, an interface's address, a process's pid); ⌘O Activity Monitor, ⌘P the row's bar popover, ⌘R reveals a volume, ⌘A the Network palette, ⌘⌫ kills a process (asks first). The detail pane has the sparkline and every figure. A push with args: { section } (what pal://stats/cpu?palette=1 and the popover's s do) lists that section's rows only.

Marked rows in the palette copy their values one per line, reveal their volumes, or kill their processes; in the CPU and Memory popovers the process rows can be marked for one Kill over them all, in Disks the volumes for Reveal and Copy path, in Network the interfaces for Copy address.

Links and states

pal://stats/cpu, memory, disk, network, load open the item's popover (through pal://bar/stats/<item>); ?palette=1 opens the palette on that section instead.

States, all stats/<name>: cpu (percent), cpu_top (the busiest process's name), cores, memory (percent), memory_pressure (normal / warn / critical), swap (percent of the swap total), disk (the startup volume's percent used), disk_free (GB), disk_worst (the fullest writable volume's percent), net_down and net_up (KB/s), load1, load5, load15. A rule of your own reads them like the manifest's (stats.net_down >= 51200).

Settings, [extensions.stats]

Each item's own settings, [bar.items."stats/<item>".settings] (Settings › Bar, on the item's pane): one key, label, what the strip says.

Tests

host/test/extensions/stats.test.ts: the parsers on canned vm_stat, sysctl, df -kP, mount, netstat -ibn, /proc/meminfo, /proc/pressure/memory, /proc/net/dev, /proc/mounts and ps output; the level and colour logic; the five popovers through checkView on fixtures; the extension in the host against the real machine (shapes, not numbers); the label settings; the popover keys; the palette's rows, picks and section argument; the links; the loop's pushes.