mrolafsson

Herdr Vitals — Development skill for Claude Code

Development community

How your machines are doing, at a glance, before they slow your agents down.

How to install Herdr Vitals

This entry records only its repository, not the path inside it, so there is no exact command to give. Open mrolafsson/herdr-vitals and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Herdr Vitals does

How your machines are doing, at a glance, before they slow your agents down. A herdr plugin: CPU, load, memory, disks (and when they'll be full), network and the busiest processes, for this box and every server you ssh to. Toasts when one runs low, a day of history, nothing to install on the servers. Works with herdr-recap.

Alternatives in Development

Investigate Performance — Investigate a Nimbalyst performance problem (freeze, lag, idle CPU, slow op, memory growth) with measured evid 1.6k ★
Htb Writeups — The most comprehensive Hack The Box writeup collection - 500+ machines, 400+ challenges, interactive knowledge 244 ★
Paper Glance Skill — 📄 A Claude skill for comprehensive academic paper analysis — deep reports, mind maps, peer review, and promo 103 ★

README

herdr-vitals

**How your machines are doing, at a glance, before they slow your agents down.**

Agents run hot. A build, a test suite, a model training, five Claude sessions at once: the box they're on fills its disk, runs out of memory or spends the afternoon swapping, and you find out when everything slows to a crawl. herdr-vitals is a [herdr](https://herdr.dev) plugin that shows every machine you work on, the one herdr runs on and every server you ssh to, as a card: CPU, load, memory, disks, network and what's busiest, live. And it tells you when one runs low.

![Five machines as cards: atlas calm, gpu-box warning on CPU with /data full in two days, build-01 hot with its disk at 92%, a Mac calm, and edge-02 refused by ssh](docs/images/grid.png)

One look, every machine. A card each, with graphs of the last few minutes. Calm is quiet, each reading in its own colour; what wants a look turns yellow, what wants you now red, and the card says why in its corner (HOT /data, WARN swap).
Disks before they're full. Every filesystem's use is written down, and from how fast it's been filling comes when it will be full: 81% full 2d.
A toast when one runs low. CPU flat out for two minutes, memory above 90%, a disk 90% full or full within a day. Once, and not again until it eases off.
Load that means something. Shown a core: 1.00 is every core busy, more is work waiting. 34 on 32 cores and 4 on 4 read the same.
A day to look back on. z turns the graphs to the last 24 hours: the busy afternoon, the backup at three, when the disk started to fill.
Every process, the whole screen, by CPU or by memory (t).
Nothing to install on your servers. A few lines of shell over ssh, one connection a machine, reading /proc (Linux) or sysctl and vm_stat (macOS). Your ssh config and keys, as they are.
Graphs full the moment it opens. A sampler keeps reading in the background, lightly: about 0.1% of a core for