22 Min to 90 Sec: I Recorded My Home Server Ritual as a
Every Sunday I burned 22 minutes running the same ten terminal commands to keep my home server alive. I recorded it once as a Claude skill. Now it's 90 seconds and I never touch the terminal. If you run any kind of server, container stack, or side-project infrastructure, you know this exact ritual — and you know nobody documents it. Most Record a Skill demos show a clean Notion click-through. I'll show you the messy version: WSL, docker, four repos, log tails, sudo prompts. By the end you'll have a copy-paste blueprint for your own maintenance skill. I'm Lazar, I run this stack daily. Here's the part that almost killed it.
So here's the setup you need to picture. I run a small home server — a mini PC sitting on a shelf — that hosts WSL Ubuntu, a docker stack, an n8n instance, and four separate agent repositories that power different client workflows. Nothing fancy. Just infrastructure that quietly runs my business while I sleep. And every Sunday morning, I had this ritual. Open the terminal. SSH into the box. Run git pull on four repos, one at a time. Check docker ps to make sure containers are alive. Tail the last hundred lines of three different log files. Ping the n8n health endpoint. Check disk usage. Restart the invoicing cron because it leaks memory after seven days. Then write a two-line note to myself confirming everything was green. Ten commands. Twenty-two minutes. Every single Sunday. And here's what nobody talks about — the reason this kind of maintenance never gets automated. It's not that it's hard. It's that it's boring, it's stateful, and every server is slightly different. Writing a shell script feels like overkill for something you do once a week. Writing a proper Ansible playbook is a whole afternoon. So you just keep doing it manually, week after week, telling yourself you'll automate it eventually. Enter Record a Skill. If you haven't seen this feature yet — inside Claude, you hit the plus menu, choose Record a Skill, and Claude watches what you do. Every command, every response, every decision point. When you're done, it saves the whole ritual as a reusable skill you can trigger by name. Most demos show someone recording a Notion workflow or a spreadsheet task. I want to show you the sysadmin version, because that's where this actually shines. Here's exactly how I recorded mine. Step one — I opened Claude, hit the plus menu, clicked Record a Skill, and named it Sunday Server Check. Step two — I started running my ritual, but this time I narrated. Not out loud. I typed a one-line intent before each command. Something like, pull latest on the invoicing repo. Then I ran the git pull. Then, check container health. Then docker ps. This narration matters. It's what teaches Claude the why, not just the what. Without it, the skill becomes a brittle replay. With it, Claude understands the intent and can adapt when a container name changes or a repo moves. Step three — the log checks. This is where most people would give up on recording. Because log output is noisy, it changes every week, and you don't want Claude to memorize specific lines. So I told Claude explicitly — tail the last hundred lines and summarize anything that looks like an error, warning, or unusual restart. That instruction got baked into the skill. Now every time it runs, Claude doesn't just dump logs at me. It reads them and tells me what actually matters. Step four — the cron restart. This is the part that leaks memory, and it needs sudo. I ran the command, hit the sudo prompt, typed my password. And here's the honest caveat — Claude cannot store your sudo password, and you don't want it to. So when I trigger the skill now, it pauses at that step and asks me to authenticate. Five seconds. Fine. Step five — the health report. At the end of my ritual, I always wrote a two-line summary. Green, everything up. Or, yellow, container three restarted twice this week, investigate. I told Claude to generate this summary automatically from everything it observed during the run, and to save it to a dated markdown file in my notes folder. Recorded. Saved. Done. The whole recording took me about twenty-eight minutes because I was narrating carefully and explaining edge cases. But that was a one-time cost. Next Sunday, I opened Claude, typed run Sunday Server Check, and watched it execute. Git pulls, four repos, parallel. Docker ps, parsed. Logs tailed and summarized. Cron restarted after I typed my sudo password. Health report written to my notes with today's date. Ninety seconds. From twenty-two minutes to ninety seconds. And here's the part I wasn't expecting. Because the health report is now consistent and dated, I have four weeks of maintenance history I never had before. I can actually see which container restarts most often, which repo has the most churn, which log file grows fastest. Recording the skill didn't just save time. It gave me observability I was too lazy to build manually. Now — what breaks. Two things, honest. First, if a command changes — say you rename a docker container or move a repo — the skill will try the old name, fail, and ask you what to do. It's not magic. You update the recording or you tell Claude the new name mid-run and it adapts. Second, permission prompts. Anything that needs interactive input — sudo, SSH key passphrases, two-factor codes — will pause the skill. That's actually the correct behavior. You don't want an automation that types passwords. But it means this isn't fully hands-off. It's ninety seconds of attention instead of twenty-two minutes of tedium. That's the trade.
Want more like this?
I publish practical AI automation, GenAI engineering, and faceless content workflows on YouTube every week.
Subscribe to bizflowai.io on YouTube — never miss a new tutorial.
Planning an AI automation project or need a second opinion on your architecture?
Connect with me on LinkedIn — Lazar Milicevic, GenAI Engineer & bizflowai.io Founder.
Visit bizflowai.io for our services, case studies, and AI consulting.