homelab-ai-skills

Runbooks my AI assistant executes against my homelab — published as working examples of the runbooks-as-skills pattern.

A skill is a small folder: a SKILL.md with the procedure, the safety rules and the interpretation guide, plus any helper scripts. The assistant loads it when a job matches ("replace disk", "is my DR current?") and follows it, instead of improvising against your storage at 11pm.

Why this beats a wiki:

• A wiki runbook is exercised when something breaks — exactly when you find out it's wrong. An agent-run skill is exercised constantly, so rot surfaces as a failed step within days, not during an outage.
• Safety rules become enforceable ("verify DR replication is green FIRST", "never receive with force toward the source"), not advisory.
• The skill records the gotchas the vendor docs don't ("ix-apps mounts at /mnt/.ix-apps, not /mnt/").

Skills in this repo

Skill	Job
truenas-health	One-shot fleet health: pools, capacity, scrubs, snapshot ages, stopped apps
truenas-replication-audit	Is the DR copy current? Per-dataset lag report + incremental sync procedure
truenas-disk-replacement	Failed-disk identification, replacement with redundancy intact, pool rebuild planning
truenas-app-migration	Moving TrueNAS 25.04 Docker app data between pools without losing anything

Adapting them

Placeholders like <TRUENAS_IP>, <DR_HOST_IP> and <USER> mark the values I stripped; pool names (BigDisk, FastPool, DR_pool) are mine — rename to yours. The "fleet facts" sections are the load-bearing part: keeping them current is what makes an agent's answers trustworthy. When a fact changes, the skill changes in the same commit.

These pair with the research and ops tooling on my other repos — see github.com/joanmarcriera and blog.riera.co.uk.

License

MIT