zabbix-alert-analysislisted
Install: claude install-skill stufently/zabbix-ai-cli-mcp
# Zabbix: why was there no alert
A notification can be dropped at five different points, and most of them record no
error anywhere. Do not reconstruct the chain by hand.
## Workflow
1. **Get the event identifier.** If you have the notification text — pasted from
chat, email, or a ticket — hand it over whole:
```bash
zabbix-ai-cli-mcp resolve "Problem started at 08:17 ... Host: web01 ... Original problem ID: 757474" --json
```
It extracts the event, host and trigger. Without this step an instruction like
"why didn't this one alert?" cannot be acted on, because the identifiers exist
only inside the message.
2. **Explain the delivery**:
```bash
zabbix-ai-cli-mcp alert why 757474 --json
```
3. **Read `findings` first.** They are ordered facts, not conclusions. The supporting
detail sits in `suppressed_by`, `delivery_attempts`, `actions`, `media_types` and
`recipients`.
## What each link means
- **Suppressed** — the event fell inside a maintenance window. Nothing else matters;
`suppressed_by` names the window and its end.
- **Delivery attempts exist and failed** — Zabbix tried. `error` carries the
transport's own message: a rejected token, a refused connection, a bad address.
- **No delivery attempts at all** — nothing was ever queued, so the cause is
configuration. Work down: is any trigger action enabled; is the media type
enabled; does any recipient have media for it.
- **Recipients listed but `severity_accepted: false`** — the