How I use Claude Code to manage sysadmin tasks (martinalderson.com)

An engineer shares a practical pattern for using Claude Code as an interactive sysadmin assistant: organize each set of maintenance tasks into its own git-backed folder with a CLAUDE.md that documents server connection info, hardware/OS inventory, common playbooks, known issues and pointers to helper files (e.g., benchmark-queries.md). The author versions these folders, keeps zero secrets in repo (use SSH keys and ~/.ssh/config aliases like "Host appserver1-lon-uk / ProxyJump bastion-host"), and lets Claude Code run "plan" steps, build inventories, and draft or verify commands. A concrete win: using Claude to research and implement clickhouse-backup with encryption to S3-compatible storage, test restores locally, and wire a dead-man webhook to healthchecks.io — all while iterating the CLAUDE.md so the agent and team have an up-to-date playbook. This pattern matters because it turns an LLM into a context-aware, auditable project memory and pair operator for small ops teams: faster onboarding, reproducible runbooks, and quicker incident resolution without embedding credentials. It maps cleanly to cloud workflows too (gcloud/azure/aws CLI instead of SSH). Caveats: it’s designed for small teams, not for large SRE orgs with strict compliance, and you must scrutinize every agent-issued command and maintain strict access controls to avoid risk.