feat(dashboards): an Overview board as the folder's entry point

[QuentinBisson]

Problem

The six alerts in agent-platform.rules.yml page with a runbook and no graph. The platform has no board of its own health, and Postgres and Valkey have no board at all — the one board that exists reads agentgateway_gen_ai_*, so it is an LLM cost and latency view, not a platform view.

Change

dashboards in both charts renders one ConfigMap per JSON under the connectivity chart's dashboards/ directory, discovered by observability-operator through the app.giantswarm.io/kind: dashboard label and placed by the observability.giantswarm.io/organization and .../folder annotations. Both values are display names carrying a space, so neither can be a label.

It ships one board: Overview, uid agent-platform-overview, in Shared Org / Agent Platform.

Row	What it shows
What this is	What the platform is, and a table of which board covers what, with a dashboard link
Component health	HelmReleases not ready, kagent-controller replicas, MCP servers failed and connected, Postgres clusters unhealthy, containers restarting, muster session store, Valkey memory — plus readiness and restarts over time
What it is doing	Tokens/s, cost/hour, tool calls/s, agents calling, gateway error ratio, MCP servers; tokens by agent and tool calls by MCP server
Postgres	Cluster phase, backends waiting, WAL archive lag
Valkey	Memory against the limit, connected clients, keys evicted

Every health panel reads the same series as the alert that pages for it — gotk_resource_info, kube_deployment_status_replicas_available, muster_mcpserver_state, kube_cnpg_cluster_info, kube_pod_container_status_restarts_total, muster_session_store_backend, redis_memory_used_bytes — and each panel description names that alert. A panel and a page cannot disagree.

Postgres gets rows here rather than a board: giantswarm/cloudnative-pg-app (Team Shield) already ships the full CloudNativePG board fleet-wide, and duplicating its JSON into our folder would be the worst of both. Its board is in the staff-only Giant Swarm organization, so for a customer these rows are the whole Postgres story.

The ConfigMaps render under the resolved global.observability.metrics.serviceMonitor.enabled: the operator that picks them up is part of the observability platform the monitors write to, and a board over series nobody collects is worse than no board. A component's own board still comes from that component's chart.

A row only renders where its data exists

A default installation runs neither CloudNativePG nor the agentgateway data plane, and several run no Valkey. The board shipped verbatim, so those rows opened as tiles of "No data" for the customers who reach this folder.

Each component-owned row now repeats over a hidden variable, label_values(<metric>, __name__), which yields exactly one value while the metric exists and none while it does not:

Gate	Probe metric	What it hides
postgres_present	kube_cnpg_cluster_info	the Postgres row
valkey_present	redis_memory_used_bytes	the Valkey row
agentgateway_present	agentgateway_requests_total	the six agentgateway panels of "what it is doing"

The agentgateway gate is per panel, not per row: MCP servers and tool calls by MCP server in the same row read muster's series, which every installation has.

This gates on data rather than on components.<name>.enabled. A component that is enabled but not yet scraped is hidden the same way, and the JSON stays a file that can be pasted into Grafana.

Checks

• dashboard-linter --strict with giantswarm/dashboards' own lint config, at the version its install-tools.sh pins: clean, exit 0.
• Classic v1 schema (schemaVersion: 41, no stripped v2 envelope), editable: false, uid and owner:team-bumblebee tag present — what check-dashboard-schema.sh and validate-dashboards.sh assert.
• All 54 verify-* targets pass. Re-run after the rebase on feat(dashboards): load the muster board into Shared Org / Agent Platform #615 and the gating: verify-target, verify-modes, verify-wiring, and dashboard-linter --strict (exit 0).

Note on the golden comparisons

verify-modes, verify-wiring and verify-target drop the board by name, not with --set: GOLDEN_REF's schema has no dashboards key, so a values hold fails its render outright and then every document reads as added. While fixing that I found the verify-modes stripper loses the leading --- whenever the dropped document is the first one; it now keeps it.

Third of three PRs putting every agent platform board in one folder, after #615 (merged) and #616. Rebased on #615. Part of giantswarm/giantswarm#36711.

[circleci-architect]

Published Helm chart agent-platform

4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28

Chart	agent-platform
Version	4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28
OCI reference	oci://gsoci.azurecr.io/charts/giantswarm/agent-platform:4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28
Digest	sha256:74105b3341f3fdc3d606361f1510adfff872f8d2f0b0722172d30ea707aae465
Registry	public — gsoci.azurecr.io
Git catalog	giantswarm-test-catalog (index)

Pull this chart

helm pull oci://gsoci.azurecr.io/charts/giantswarm/agent-platform --version 4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28

_{Posted by architect-orb · build 11027 · commit 8505e28 · updated in place on every push}

[circleci-architect]

Published Helm chart agent-platform-connectivity

4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28

Chart	agent-platform-connectivity
Version	4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28
OCI reference	oci://gsoci.azurecr.io/charts/giantswarm/agent-platform-connectivity:4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28
Digest	sha256:53897754b2e382b4c2e779f5d4de066c16308a869f42d45e21dfa34766741dd1
Registry	public — gsoci.azurecr.io
Git catalog	giantswarm-test-catalog (index)

Pull this chart

helm pull oci://gsoci.azurecr.io/charts/giantswarm/agent-platform-connectivity --version 4.51.1-dev.feat-overview-dashboard.2026-09-22.08-58-35.h8505e28

_{Posted by architect-orb · build 11026 · commit 8505e28 · updated in place on every push}