zabbix-investigatelisted
Install: claude install-skill stufently/zabbix-ai-cli-mcp
# Zabbix: investigate a host
Collect the facts about one host, then reason over them. The tool gathers; you interpret.
## Workflow
1. **One call gathers the context.** Do not assemble it by hand.
```bash
zabbix-ai-cli-mcp host investigate web01 --json
```
This returns host state, agent availability, active problems, recent events,
silent and unsupported items, and any maintenance window covering the host.
2. **Follow the strongest signal**, in this order:
- `status.agent_available: "unavailable"` → Zabbix cannot reach the host at all.
The interface `error` field usually says why. Everything else is downstream of this.
- `status.in_maintenance: true` → alerts are suppressed. Check whether the window
should still be open before treating silence as health.
- `no_data_items` non-empty → the host answers but specific checks stopped
reporting. Compare their keys: one subsystem or all of them?
- `unsupported_items` non-empty → the check itself is failing, and `error` says how.
This is a monitoring fault, not necessarily a host fault.
- `active_problems` → read `severity` and `age` together; an old average-severity
problem is often more telling than a fresh warning.
3. **Values over time**, when a problem is about a threshold:
```bash
zabbix-ai-cli-mcp metrics history web01 "cpu util" --last 24h --json
```
The `summary` block gives min, average and max without reading every point.
4. **Recent history**: `recent_