agentleFS
Lukk17Lukk17/agent-standards/.agents/skills/homelab-services/SKILL.md

homelab-services

Skill1 starsChanged 3 days agoMITView on GitHub

Running a Docker Compose homelab on one VM, every change asked of the owner first: AdGuard Home, Uptime Kuma, Telegraf, Samba, File Browser, Plex, Homepage, Nginx Proxy Manager, a local CA for .internal names, and data disks. Use when the user says "add a .internal name", "https for the .internal names", "deploy the compose change" or "the VM disk is full". Not for the Proxmox host, use `proxmox`.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every service carries a json-file log limit, max-size and max-file, and older containers are recreated to pick it up.

    line 62
  2. Very specific:

    Create every mounted folder as the login user before the first start.

    line 63
  3. Very specific:

    Never mount NTFS with force.

    line 68

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Homelab Services

How an agent runs and repairs a small self-hosted stack: one Ubuntu Server VM, one Compose file, and the services a home network leans on, DNS first. The Compose file and the files it mounts live in a git clone on the owner's machine, and the VM holds a plain copy. Every change to a server, a service or the router is proposed to the owner first, with what can go wrong and how to undo it, and nothing changes before a yes.

Written for AdGuard Home 0.107, Uptime Kuma 2, Telegraf 1, Samba 4, File Browser 2 (gtstef/filebrowser), Plex Media Server, Homepage 1 and Nginx Proxy Manager 2 as containers on an Ubuntu Server VM under Proxmox VE, with OpenSSL 3 in a throwaway container for the local CA. Facts carry a VERIFIED or UNVERIFIED mark, explained in conventions.md.

You are about toOpen
Name a host, user, folder or container: setup facts live in the owner's notes, examples use <placeholders>. Or judge a VERIFIED markconventions.md
Copy a change to the VM, start a service, create its folders, cap container logs, write a $ into Composedeploy-and-logs.md
Change AdGuard Home rewrites, upstreams, retention, ignored domains, or AdGuardHome.yamladguard-home.md
Change Uptime Kuma monitors, /metrics, the labelset log flood, or its databaseuptime-kuma.md
Run or read the Telegraf ping, DNS, HTTP and speed probes and their CSV filestelegraf-network-watch.md
Set up the Samba share and its login filesamba.md
Set up File Browser, read or reset its admin passwordfile-browser.md
Run, claim and lay out Plex with read-only mediaplex.md
Configure the Homepage dashboard and its allowed hostshomepage.md
Add a proxy host for a .internal name, or fix one that answers 500nginx-proxy-manager.md
Give .internal names https: make, extend, renew, trust or revoke the local CAlocal-ca.md
Attach, mount and protect a USB or NTFS data disk, or clear a dirty NTFS volumedata-disk.md
Explain a full disk, find what filled it, free spacedisk-full.md

When to activate

  • Adding a service, or copying a changed Compose file or Telegraf config to the VM
  • A .internal name that does not resolve, or does not reach its service
  • Changing how long AdGuard Home keeps its query log or statistics
  • Changing monitors in Uptime Kuma, or a container log that grows by gigabytes
  • Reading the network watch files after an internet drop
  • Attaching a data disk, sharing it over Samba and File Browser, or pointing Plex at it
  • A VM whose root file system is full, or services that stopped writing
  • The user says "keep 30 days of query log", "change the monitor interval", "Added label is not included in initial labelset" or "mount the NTFS disk"

When not to activate

  • The Proxmox host, its API token, or growing the VM's virtual disk and its LVM volume, use proxmox
  • Syncthing itself, its folders, devices and its disk space error, use syncthing
  • The router, its DHCP DNS setting, its log and WAN drops, use asus-router
  • Writing a Dockerfile, choosing a base image or designing a Compose stack in general, use docker-patterns
  • Judging an exposed port, a token or a password policy, use the security-auditor agent
  • Checking the health of the owner's own computer, use machine-health

Hard rules

The reason behind each rule is in conventions.md.

  1. Ask before every change on the VM, a service, the Proxmox host or the router: a file edit, a restart, a stop, a docker compose up -d, a mount, an fstab line, a package. Say what changes, the worst case and how to undo it, then wait for a yes. One yes covers one change. Reading is free.
  2. The git clone is the source of truth: change it first, copy to the VM with scp. Never change only the VM copy, never fetch from a GitHub raw link.
  3. Every service carries a json-file log limit, max-size and max-file, and older containers are recreated to pick it up.
  4. Create every mounted folder as the login user before the first start. A data disk mount uses create_host_path: false and an empty mount point with chattr +i.
  5. Back up a configuration file or database before editing it, in the topic file's order: AdGuard Home edit then restart, Uptime Kuma stop then edit.
  6. Lowering the AdGuard Home query log interval deletes old queries at the next start or hourly check, for good. Say so, offer a copy of both log files first, and wait for a yes.
  7. Restart Uptime Kuma after creating a tag.
  8. Secrets live in .env beside the Compose file, mode 600, entered hidden, checked by length, never printed or committed. The agent's own credentials live in an env file outside every repository, mode 600, read one key at a time with grep and cut, never sourced or printed.
  9. Never mount NTFS with force. Clear a dirty volume with ntfsfix -d after the owner says yes, or check it in Windows.
  10. Probes and monitors target an address and port, never a .internal name.
  11. The local CA is name-constrained to DNS:internal, its key only on the owner's computer and in the VM's .env. Once its include serves https, every proxy host's SSL tab stays on None.

Related skills

  • proxmox owns the host, attaching a disk to the VM, and growing the VM's disk, partition and LVM volume
  • syncthing owns Syncthing in its container, including the folder that carries the network watch files
  • asus-router owns the router's DHCP DNS server setting and reading WAN drops from its log
  • docker-patterns owns Compose and image design in general
  • The security-auditor agent owns judging exposed ports, tokens and passwords
  • bash owns scripts that wrap these commands

Checklist

The commands behind each item are in conventions.md.

  • Changed in the clone, copied with scp, then docker compose pull, up -d and ps
  • A container whose mounted file changed was restarted, Telegraf above all
  • Every service has max-size and max-file, containers recreated once after
  • Every mounted folder existed before the first start, owned by the login user
  • Config or database backed up before the edit, value read back after
  • A lower query log interval approved after the owner heard old queries go for good
  • Uptime Kuma restarted after a new tag, and its log shows no PROMETHEUS lines
  • .env is mode 600, values checked by length, nothing secret printed
  • Data disk mounted by UUID with nofail, mount point chattr +i, findmnt shows it
  • df -h / read after any change that writes data, largest container logs checked
  • Local CA read back with CA:TRUE, pathlen:0, Permitted: DNS:internal, both exclusions, and ca.key in no repository
  • Every proxied .internal name in INTERNAL_DNS_NAMES, every SSL tab on None, proxy restarted after a new certificate
  • The certificate notAfter= date is more than 30 days away, or the job ran again

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.