prom-query_skill

Overview

This skill provides Prometheus-compatible metrics querying and alert interpretation. It lets you run instant and range PromQL queries, list alerts and targets, explore metric names, and inspect alerting/recording rules. Use it to triage incidents, interpret timeseries behavior, and find correlated signals across Prometheus, Thanos, Mimir, or VictoriaMetrics backends.

How this skill works

The skill translates natural-language intents into PromQL patterns and executes HTTP API queries against a configured Prometheus-compatible server. It supports instant queries, range queries with automatic downsampling, alerts listing by state, target health checks, metric name exploration, and rules inspection. Range results include summaries (min/max/avg/first/last/pointCount) and the skill warns when results are downsampled, offering zoom-in options.

When to use it

• Triage a firing alert and find the responsible metric(s) quickly
• Investigate trends or spikes using range queries over recent windows
• Check scrape target health and identify dropped or unhealthy exporters
• Explore available metrics to discover the correct metric name or labels
• Validate alerting rules or find why an alert fired

Best practices

• Start with alerts, then check targets before querying metrics to narrow root cause
• Use range queries with reasonable steps; zoom in if the result was downsampled
• Use label filters to limit cardinality and reduce query cost
• Translate intent to PromQL using provided patterns (error rate, latency, CPU, memory, disk, network)
• When a query times out, reduce time range, add filters, or use topk() to constrain results

Example use cases

• Show current service health: instant query 'up' across instances
• Find API error rate: rate(http_requests_total{code=~"5.."}[5m]) / rate(http_requests_total[5m])
• Investigate a spike: range query on request latency histogram and histogram_quantile for P99
• List firing alerts and group by severity with timestamps and summaries
• Discover metrics for a new service using explore 'service_name' and then run targeted queries