← ClaudeAtlas

observability-grafana-dashboardslisted

Use when a new OTel metric (or pprof profile type) is added to stapler-squad and needs a Grafana dashboard, when auditing whether existing metrics have dashboard coverage, or when a dashboard's panels feel wrong/noisy. Covers the RED/USE framework for deciding what a dashboard should answer, panel-type selection by metric shape, the per-project Grafana folder/provisioning convention, and the verification steps that catch silently-broken panels (wrong metric name, wrong profileTypeId) before calling a dashboard done.
tstapler/stapler-squad · ★ 1 · AI & Automation · score 64
Install: claude install-skill tstapler/stapler-squad
# Building and Maintaining stapler-squad's Grafana Dashboards Dashboard source lives in `~/dotfiles/stapler-scripts/observability/grafana/dashboards/stapler-squad/` (machine-shared docker-compose stack, not in this repo — see the `observability-victoriametrics-tempo` skill for the full architecture). This skill is about *authoring* dashboards; use that skill for *querying* the live backends first to find out what data actually exists. ## Step 1: Find every un-dashboarded metric before designing anything Don't design a panel from a guess at what's instrumented. Enumerate real instruments first: 1. Grep the Go source for meter registration: `Int64Counter`, `Int64ObservableGauge`, `Float64Histogram`, `Int64ObservableCounter`, etc. (see `telemetry/telemetry.go` and any `*_metrics.go` file for the pattern). 2. Cross-reference against the existing dashboard JSON's `targets[].expr` fields — anything instrumented but never queried in a dashboard is a gap. 3. For each gap, confirm the *live* metric name via VictoriaMetrics rather than guessing the OTel→Prometheus mangling yourself (dots→underscores, unit suffixes, `_total` on counters) — see the metric-name-mangling gotcha in `observability-victoriametrics-tempo`. A dashboard panel with a wrong metric name doesn't error, it just renders "No data" forever, which is worse than an error because nothing calls it out. ## Step 2: Decide what question the dashboard answers (RED vs. USE) Per Grafana's own dashboard