Example Prompts

What you can ask depends on which backends you configured. The server advertises its tools to the LLM, so plain language is enough: you describe what you want, the model picks the tool. You never name a tool yourself.

A good habit is to open a session with a health check; it confirms, without touching the network, which instances and backends are wired up.

Run a health check.

Plugin catalog

These work as soon as monitoring_plugins.catalog_path points at a local checkout of the monitoring-plugins repository (the check-plugins/ directory). No Icinga connection is required, so they are the easiest first test.

List all Linuxfabrik monitoring plugins.

Which plugins run on Windows?

Explain the disk-usage plugin: what does it check, which arguments does it take, and what perfdata does it emit?

Which plugin backs the check command cmd-check-disk-usage?

Show me the source of the main function of disk-usage.

Read the full source of the disk-usage plugin and explain how it decides between WARNING and CRITICAL.

Live host and service state

These require an instance with a reachable icinga2_core backend. Name the instance in the prompt, or omit it when only one instance is configured and the server selects it automatically.

What is currently not OK on prod-zh?

Show all DOWN hosts on prod-zh.

Which services are CRITICAL on prod-zh, and what did their last check output?

Show the full state of host db01 on prod-zh, including its last check output and perfdata.

Investigate the problems on prod-zh and prod-fr.

Root-cause bridge

The point of pairing live state with the plugin catalog is to explain a problem, not just report it. In one conversation the server can read the failing service, resolve its check command to a plugin, and explain what that plugin measures.

Why is the disk service on db01 critical, and what does that check actually measure?

Go through every problem on prod-zh and, for each failing service, explain what its check does and which thresholds it uses.

Acting on problems

These require an instance with write credentials (write_username / write_password). Without them the action tools are not offered at all, so a read-only deployment stays read-only. Every action is attributed to the configured write user.

Acknowledge the disk problem on db01 with the comment "cleanup in progress".

Schedule a two-hour downtime for host db01 and all its services, comment "kernel upgrade".

Remove the downtime from host db01.

Recheck the http service on web01 now.

Where to go next

• Configuration: wire up instances, backends and secrets so the prompts above have something to talk to.
• Quickstart with Claude: register the server in Claude Desktop or Claude Code.