prometheus-mimir-grafanalisted

# Prometheus, Mimir, and Grafana — for agents Target audience: an AI agent (or a human working through one) that has to *do things with metrics* — query, triage, alert, build and fix dashboards, and pick the right KPIs — against a stack that runs Prometheus, Grafana Mimir, and/or Grafana. Works whether the agent is given curl access to a Mimir gateway, an MCP server wrapper, or just a Grafana URL and a service-account token. ## Why this matters Metrics lie in three directions: (1) the agent queries the wrong metric or wrong label, (2) the query is syntactically fine but semantically broken (`rate` after aggregation, `histogram_quantile` of the mean, default histogram buckets sized for the wrong service), (3) the dashboard *looks* correct but the datasource variable is empty or the unit is off by 1000×. Each failure mode has an easy check. This skill is those checks, organized so the agent reaches for them before issuing the first query. ## The one-paragraph rubric Prometheus stores samples identified by `metric_name{label=value, ...}`. Mimir is a horizontally-scalable multi-tenant store that speaks Prometheus's wire protocol and API under a `/prometheus/api/v1/…` prefix, gated by `X-Scope-OrgID`. Grafana is the UI and the dashboards-as-JSON store. PromQL returns instant vectors, range vectors, scalars, or strings; `rate()` always wraps a counter before any `sum`; `histogram_quantile()` always consumes aggregated `_bucket` rates; and `$__rate_interval` is the only interva