Skip to content
Conduit

skill

server-discovery

Agent skill: read-only discovery of servers over ssh, host cards, fleet index and network map for a devops harness

EvilFreelancer/server-discoverycommunitylistedunreviewed
Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/EvilFreelancer/server-discovery itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 26, 2026

server-discovery

An agent skill that reads servers read-only over ssh and keeps their host cards in a fleet harness repository up to date. It owns the scripts the whole set of fleet skills runs on: the probes, the card writer, the validator, the host index and the network map.

It is one of four skills of a universal devops harness, a git repository that mirrors a fleet of servers as code:

SkillDoes
server-discoveryread a box, compare with its card, refresh the card, validate the fleet, regenerate the index and the map
server-onboardingadd a new box: placement, card, README, mirrored configs, Topology, GPU hand-off
server-healthcheckis a box or a namespace actually healthy, read-only
gpu-server-setupNVIDIA or Intel Arc driver, Docker GPU access, serving stacks, power caps

The harness it expects

fleet.yaml                       optional settings: password policy, mirror layout, extra tags, map
README.md                        generated list of every namespace between <!-- hosts:start/end -->
<namespace>/README.md            namespace card (frontmatter), ## Topology (Mermaid), host list
<namespace>/<host>/README.md     host card (frontmatter) + narrative
<namespace>/<host>/...           configs mirrored from the box
.agents/skills/<skill>/          the skills

The card schema, the namespace card, the Topology notation and fleet.yaml are documented in references/: host-card.md, namespace-card.md, harness.md.

What it does

  • scripts/discover.sh - a POSIX probe that prints ### section blocks: identity, OS, kernel, CPU, memory, disks, sudo, sshd auth, Docker and its runtimes, nvidia-smi, NVIDIA cards on the PCI bus from sysfs, Intel discrete GPUs (Arc, Battlemage) from sysfs and debugfs. Runs on Debian, Ubuntu, busybox (OpenWrt) and Synology DSM.
  • scripts/discover.py - sends the probe over ssh in batch mode (or parses a saved run) and prints the card facts as JSON, plus probe keys that flag an incomplete NVIDIA stack (cards on the bus without a driver, Docker without the nvidia runtime); --apply writes the facts into a card and leaves the hand-written keys and the body alone.
  • scripts/discover-net.sh - the network counterpart: addresses, routes, listening ports, Docker publish ports and compose projects, wireguard, OpenVPN, /etc/hosts, ~/.ssh/config, DNAT rules (iptables, nft, OpenWrt uci).
  • scripts/render_index.py - validates fleet.yaml and every card and rewrites the generated host lists; --check is the commit gate.
  • scripts/render_map.py - when fleet.yaml turns it on, one self-contained HTML page that draws the fleet as a graph from the cards and the Topology diagrams, fonts inlined, no network needed.

The skill compares before it writes: a read that lost a GPU, a disk or a known value blocks the apply, and another hostname answering on the card's ssh command is reported, not written.

Install

The skill is distributed through the rpa-skills catalog together with the other fleet skills.

Claude Code: add the catalog once, then install the set from it:

/plugin marketplace add EvilFreelancer/rpa-skills
/plugin install server-discovery@rpa-skills
/plugin install server-onboarding@rpa-skills
/plugin install server-healthcheck@rpa-skills

Codex: codex plugin marketplace add EvilFreelancer/rpa-skills, then install from /plugins.

Coddy: the catalog is built in, coddy skills sync installs and updates the whole set.

Inside a harness repository, as a git submodule next to the other fleet skills:

git submodule add https://github.com/EvilFreelancer/server-discovery.git .agents/skills/server-discovery
ln -s ../../.agents/skills/server-discovery .claude/skills/server-discovery   # Claude Code

As a plain skill folder, copy or symlink this repository into a skill root (its SKILL.md lives at the repo root); the directory name must match the name field:

ToolPath
Coddy~/.coddy/skills/server-discovery/
Claude Code~/.claude/skills/server-discovery/
Cursor~/.cursor/skills/server-discovery/
OpenAI Codex~/.codex/skills/server-discovery/

The commands in the skills assume the project-level install .agents/skills/; with another location the agent uses that prefix.

Requirements on the machine the agent runs on: Python 3.10+, PyYAML, OpenSSH. On the boxes: a POSIX sh.

Usage

  • "Re-scan node1, the driver was upgraded."
  • "Which kernel and GPUs does every box in the lab namespace have now?"
  • "Validate the cards and regenerate the index."
  • "Draw the network map of the fleet."
python3 scripts/discover.py --ssh "ssh node1"                      # JSON, writes nothing
python3 scripts/discover.py --input node1.txt --apply lab/node1/README.md
python3 scripts/render_index.py --check
python3 -m unittest discover -s tests -v

License

MIT © Pavel Rykov. The bundled Inter and JetBrains Mono fonts are under the SIL Open Font License 1.1, see templates/fonts/.

↑↓ navigate openesc close