v0.10.23.

The node telemetry release: Monitoring finally sees every node (r467) — the last purely additive item on the audit backlog, closed with a composite agent op and a card row that stops lying about whose numbers they are.

Added

• Per-node telemetry on the Monitoring page (r467). Clicking a remote node's card now switches the overview cards and the workload grid to THAT node's live numbers — CPU cores and load, memory, disk, and per-service CPU/memory — instead of the panel host's figures under a "Node" label (the old honest note said so; now it is simply true). A new composite agent.stats handled op runs on the node: host stats from /proc (the host's real values even from inside the agent container — no lxcfs virtualization on a standard node), per-container docker stats --no-stream, and the node's disk via df. Everything rides the sealed, nonce-bound agent protocol; the panel joins container names to the service rows pinned to that node exactly like /v1/stats does for the panel host. The cards refresh every 8 seconds; an unreachable agent degrades to placeholders with an explicit "agent unreachable" sub-label rather than silently showing the master's numbers. New endpoint GET /v1/servers/:id/stats (admin-only) and SDK client.servers.stats(id); the panel-host view is byte-identical when "Local Host" is selected. Node agents must be ≥0.10.23 — an older agent answers unknown_op, which the route surfaces as "agent unreachable" instead of guessing.