@openclaw/diagnostics-prometheus 2026.5.24-beta.1 → 2026.5.25-beta.1

package/dist/api.js

@@ -1,3 +1,4 @@
 import { emptyPluginConfigSchema } from "openclaw/plugin-sdk/plugin-entry";
+import { isInternalDiagnosticEventMetadata } from "openclaw/plugin-sdk/diagnostic-runtime";
 import { redactSensitiveText } from "openclaw/plugin-sdk/security-runtime";
-export { emptyPluginConfigSchema, redactSensitiveText };
+export { emptyPluginConfigSchema, isInternalDiagnosticEventMetadata, redactSensitiveText };

package/dist/index.js

@@ -1,4 +1,4 @@
-import { redactSensitiveText } from "./api.js";
+import { isInternalDiagnosticEventMetadata, redactSensitiveText } from "./api.js";
 import { definePluginEntry } from "openclaw/plugin-sdk/plugin-entry";
 //#region extensions/diagnostics-prometheus/src/service.ts
 const DURATION_BUCKETS_SECONDS = [
@@ -47,6 +47,19 @@ const BYTE_BUCKETS = [
 	4294967296,
 	17179869184
 ];
+const RATIO_BUCKETS = [
+	.01,
+	.05,
+	.1,
+	.25,
+	.5,
+	.75,
+	1,
+	2,
+	4,
+	8,
+	16
+];
 const LOW_CARDINALITY_VALUE_RE = /^[A-Za-z0-9_.:-]{1,120}$/u;
 const MAX_PROMETHEUS_SERIES = 2048;
 const DROPPED_SERIES_COUNTER_NAME = "openclaw_prometheus_series_dropped_total";
@@ -183,6 +196,9 @@ function createPrometheusMetricStore() {
 function safeErrorMessage(err) {
 	return redactSensitiveText(err instanceof Error ? err.message ?? err.name : String(err)).replaceAll("\0", " ").replace(/[\r\n\t\u2028\u2029]/gu, " ").slice(0, 500);
 }
+function shouldRecordDiagnosticEvent(metadata) {
+	return metadata.trusted || isInternalDiagnosticEventMetadata(metadata);
+}
 function renderPrometheusMetrics(store) {
 	const snapshot = store.snapshot();
 	const lines = [];
@@ -247,6 +263,17 @@ function modelCallLabels(evt) {
 		transport: lowCardinalityLabel(evt.transport)
 	};
 }
+function modelFailoverLabels(evt) {
+	return {
+		from_model: lowCardinalityLabel(evt.fromModel),
+		from_provider: lowCardinalityLabel(evt.fromProvider),
+		lane: lowCardinalityQueueLaneLabel(evt.lane),
+		reason: lowCardinalityLabel(evt.reason, "other"),
+		suspended: evt.suspended === void 0 ? "unknown" : String(evt.suspended),
+		to_model: lowCardinalityLabel(evt.toModel),
+		to_provider: lowCardinalityLabel(evt.toProvider)
+	};
+}
 function toolExecutionLabels(evt) {
 	return {
 		error_category: evt.type === "tool.execution.error" ? lowCardinalityLabel(evt.errorCategory, "other") : "none",
@@ -257,6 +284,15 @@ function toolExecutionLabels(evt) {
 		tool_source: lowCardinalityLabel(evt.toolSource, "core")
 	};
 }
+function toolExecutionBlockedLabels(evt) {
+	return {
+		denied_reason: lowCardinalityLabel(evt.deniedReason, "other"),
+		params_kind: lowCardinalityLabel(evt.paramsSummary?.kind),
+		tool: lowCardinalityLabel(evt.toolName, "tool"),
+		tool_owner: lowCardinalityLabel(evt.toolOwner, "none"),
+		tool_source: lowCardinalityLabel(evt.toolSource, "core")
+	};
+}
 function skillLabels(evt) {
 	return {
 		activation: lowCardinalityLabel(evt.activation, "unknown"),
@@ -277,6 +313,18 @@ function harnessLabels(evt) {
 		provider: lowCardinalityLabel(evt.provider)
 	};
 }
+function webhookLabels(evt) {
+	return {
+		channel: lowCardinalityLabel(evt.channel),
+		webhook: lowCardinalityLabel(evt.updateType)
+	};
+}
+function sessionStuckLabels(evt) {
+	return {
+		reason: lowCardinalityLabel(evt.reason, "none"),
+		state: evt.state
+	};
+}
 function sessionRecoveryLabels(evt) {
 	return {
 		action: evt.type === "session.recovery.completed" ? lowCardinalityLabel(evt.action, "unknown") : evt.allowActiveAbort ? "abort" : "recover",
@@ -285,6 +333,18 @@ function sessionRecoveryLabels(evt) {
 		status: evt.type === "session.recovery.completed" ? evt.status : "requested"
 	};
 }
+function livenessLabels(evt) {
+	return { reason: lowCardinalityLabel(evt.reasons.join(":"), "unknown") };
+}
+function payloadLargeLabels(evt) {
+	return {
+		action: evt.action,
+		channel: lowCardinalityLabel(evt.channel, "none"),
+		plugin: lowCardinalityLabel(evt.pluginId, "none"),
+		reason: lowCardinalityLabel(evt.reason, "none"),
+		surface: lowCardinalityLabel(evt.surface, "unknown")
+	};
+}
 function talkLabels(evt) {
 	return {
 		brain: lowCardinalityLabel(evt.brain),
@@ -325,7 +385,7 @@ function recordModelUsage(store, evt) {
 	store.histogram("openclaw_model_usage_duration_seconds", "Model usage event duration in seconds.", labels, seconds(evt.durationMs));
 }
 function recordDiagnosticEvent(store, evt, metadata) {
-	if (!metadata.trusted) return;
+	if (!shouldRecordDiagnosticEvent(metadata)) return;
 	switch (evt.type) {
 		case "model.usage":
 			recordModelUsage(store, evt);
@@ -339,11 +399,17 @@ function recordDiagnosticEvent(store, evt, metadata) {
 			store.histogram("openclaw_model_call_duration_seconds", "Provider model call duration in seconds.", modelCallLabels(evt), seconds(evt.durationMs));
 			store.counter("openclaw_model_call_total", "Provider model calls completed by outcome.", modelCallLabels(evt));
 			return;
+		case "model.failover":
+			store.counter("openclaw_model_failover_total", "Model failovers by source, destination, lane, and reason.", modelFailoverLabels(evt));
+			return;
 		case "tool.execution.completed":
 		case "tool.execution.error":
 			store.histogram("openclaw_tool_execution_duration_seconds", "Tool execution duration in seconds.", toolExecutionLabels(evt), seconds(evt.durationMs));
 			store.counter("openclaw_tool_execution_total", "Tool executions completed by outcome.", toolExecutionLabels(evt));
 			return;
+		case "tool.execution.blocked":
+			store.counter("openclaw_tool_execution_blocked_total", "Tool executions blocked by policy or sandbox diagnostics.", toolExecutionBlockedLabels(evt));
+			return;
 		case "skill.used":
 			store.counter("openclaw_skill_used_total", "Skills used by agent runs.", skillLabels(evt));
 			return;
@@ -364,6 +430,15 @@ function recordDiagnosticEvent(store, evt, metadata) {
 				reason: lowCardinalityLabel(evt.reason, "none")
 			}, seconds(evt.durationMs));
 			return;
+		case "webhook.received":
+			store.counter("openclaw_webhook_received_total", "Webhook requests received by channel and update type.", webhookLabels(evt));
+			return;
+		case "webhook.processed":
+			store.histogram("openclaw_webhook_duration_seconds", "Webhook processing duration in seconds.", webhookLabels(evt), seconds(evt.durationMs));
+			return;
+		case "webhook.error":
+			store.counter("openclaw_webhook_error_total", "Webhook processing errors by channel and update type.", webhookLabels(evt));
+			return;
 		case "message.delivery.started":
 			store.counter("openclaw_message_delivery_started_total", "Outbound message delivery attempts started.", {
 				channel: lowCardinalityLabel(evt.channel),
@@ -433,6 +508,10 @@ function recordDiagnosticEvent(store, evt, metadata) {
 			});
 			if (evt.queueDepth !== void 0) store.gauge("openclaw_session_queue_depth", "Latest observed session queue depth.", { state: evt.state }, numericValue(evt.queueDepth));
 			return;
+		case "session.stuck":
+			store.counter("openclaw_session_stuck_total", "Stale session bookkeeping observations with no active work.", sessionStuckLabels(evt));
+			store.histogram("openclaw_session_stuck_age_seconds", "Age of stale session bookkeeping observations in seconds.", sessionStuckLabels(evt), seconds(evt.ageMs));
+			return;
 		case "session.turn.created":
 			store.counter("openclaw_session_turn_created_total", "Agent session turns created.", {
 				agent: lowCardinalityLabel(evt.agentId),
@@ -452,6 +531,16 @@ function recordDiagnosticEvent(store, evt, metadata) {
 				reason: evt.reason
 			});
 			return;
+		case "diagnostic.liveness.warning":
+			store.counter("openclaw_liveness_warning_total", "Diagnostic liveness warning events.", livenessLabels(evt));
+			store.gauge("openclaw_liveness_sessions", "Latest session counts reported with diagnostic liveness warnings.", { state: "active" }, numericValue(evt.active));
+			store.gauge("openclaw_liveness_sessions", "Latest session counts reported with diagnostic liveness warnings.", { state: "waiting" }, numericValue(evt.waiting));
+			store.gauge("openclaw_liveness_sessions", "Latest session counts reported with diagnostic liveness warnings.", { state: "queued" }, numericValue(evt.queued));
+			store.histogram("openclaw_liveness_event_loop_delay_p99_seconds", "P99 event-loop delay reported by diagnostic liveness warnings in seconds.", livenessLabels(evt), seconds(evt.eventLoopDelayP99Ms));
+			store.histogram("openclaw_liveness_event_loop_delay_max_seconds", "Maximum event-loop delay reported by diagnostic liveness warnings in seconds.", livenessLabels(evt), seconds(evt.eventLoopDelayMaxMs));
+			store.histogram("openclaw_liveness_event_loop_utilization_ratio", "Event-loop utilization reported by diagnostic liveness warnings.", livenessLabels(evt), numericValue(evt.eventLoopUtilization), RATIO_BUCKETS);
+			store.histogram("openclaw_liveness_cpu_core_ratio", "CPU core ratio reported by diagnostic liveness warnings.", livenessLabels(evt), numericValue(evt.cpuCoreRatio), RATIO_BUCKETS);
+			return;
 		case "diagnostic.async_queue.dropped":
 			store.counter("openclaw_diagnostic_async_queue_dropped_total", "Async diagnostic queue drops by dropped event class.", { drop_class: "total" }, numericValue(evt.droppedEvents));
 			if (evt.droppedTrustedEvents !== void 0) store.counter("openclaw_diagnostic_async_queue_dropped_total", "Async diagnostic queue drops by dropped event class.", { drop_class: "trusted" }, numericValue(evt.droppedTrustedEvents));
@@ -459,8 +548,7 @@ function recordDiagnosticEvent(store, evt, metadata) {
 			if (evt.droppedPriorityEvents !== void 0) store.counter("openclaw_diagnostic_async_queue_dropped_total", "Async diagnostic queue drops by dropped event class.", { drop_class: "priority" }, numericValue(evt.droppedPriorityEvents));
 			store.gauge("openclaw_diagnostic_async_queue_length", "Latest async diagnostic queue length after a drop summary.", {}, numericValue(evt.queueLength));
 			return;
-		case "diagnostic.heartbeat":
-		case "diagnostic.liveness.warning": return;
+		case "diagnostic.heartbeat": return;
 		case "telemetry.exporter":
 			store.counter("openclaw_telemetry_exporter_total", "Telemetry exporter lifecycle events.", {
 				exporter: lowCardinalityLabel(evt.exporter),
@@ -469,6 +557,10 @@ function recordDiagnosticEvent(store, evt, metadata) {
 				status: evt.status
 			});
 			return;
+		case "payload.large":
+			store.counter("openclaw_payload_large_total", "Oversized payload diagnostics by surface and action.", payloadLargeLabels(evt));
+			store.histogram("openclaw_payload_large_bytes", "Oversized payload byte sizes by surface and action.", payloadLargeLabels(evt), numericValue(evt.bytes), BYTE_BUCKETS);
+			return;
 		default: return;
 	}
 }

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "@openclaw/diagnostics-prometheus",
-  "version": "2026.5.24-beta.1",
+  "version": "2026.5.25-beta.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@openclaw/diagnostics-prometheus",
-      "version": "2026.5.24-beta.1"
+      "version": "2026.5.25-beta.1"
     }
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openclaw/diagnostics-prometheus",
-  "version": "2026.5.24-beta.1",
+  "version": "2026.5.25-beta.1",
   "description": "OpenClaw diagnostics Prometheus exporter",
   "repository": {
     "type": "git",
@@ -18,10 +18,10 @@
       "minHostVersion": ">=2026.4.25"
     },
     "compat": {
-      "pluginApi": ">=2026.5.24-beta.1"
+      "pluginApi": ">=2026.5.25-beta.1"
     },
     "build": {
-      "openclawVersion": "2026.5.24-beta.1"
+      "openclawVersion": "2026.5.25-beta.1"
     },
     "release": {
       "publishToClawHub": true,
@@ -37,7 +37,7 @@
     "npm-shrinkwrap.json"
   ],
   "peerDependencies": {
-    "openclaw": ">=2026.5.24-beta.1"
+    "openclaw": ">=2026.5.25-beta.1"
   },
   "peerDependenciesMeta": {
     "openclaw": {