agent-relay-server 0.91.2 → 0.91.3

package/docs/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "Agent Relay API",
-    "version": "0.91.2",
+    "version": "0.91.3",
     "description": "Real-time message bus for inter-agent communication. Agent-first: this spec is designed for machine consumption — agents can self-discover the full API surface via GET /api/spec.",
     "license": {
       "name": "MIT",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-relay-server",
-  "version": "0.91.2",
+  "version": "0.91.3",
   "description": "Lightweight HTTP message relay for inter-agent communication across machines",
   "module": "src/index.ts",
   "type": "module",

package/runner/src/config.ts

@@ -99,6 +99,11 @@ export function registrationTimeoutMsFromEnv(): number {
   return Number.isFinite(parsed) && parsed > 0 ? parsed : 60_000;
 }
 
+export function nativeSelfResumeTimeoutMsFromEnv(): number {
+  const parsed = Number(process.env.AGENT_RELAY_NATIVE_SELF_RESUME_TIMEOUT_MS);
+  return Number.isFinite(parsed) && parsed > 0 ? parsed : 120_000;
+}
+
 export function agentProfileNameFromEnv(): string | undefined {
   return process.env.AGENT_RELAY_AGENT_PROFILE;
 }

package/src/agent-death-diagnosis.ts

@@ -173,6 +173,10 @@ export function classifyAgentDeath(ev: DeathEvidence): AgentDeathDiagnosis {
     pushUnique(evidence, "log/error text matched a crash pattern (panic/segfault/traceback)");
     return verdict("provider-crash", "medium", "Likely provider crash (log pattern)");
   }
+  if (/self-resume-(?:exit-before-resume|relaunch-failed)/i.test(text)) {
+    pushUnique(evidence, "native self-resume exited before Relay observed a resumed provider");
+    return verdict("provider-crash", "high", "Self-resume failed before relaunch completed");
+  }
 
   // 5) killed — external SIGKILL/SIGTERM that isn't OOM and isn't a parent-initiated shutdown.
   if (signal && /KILL|TERM|INT|HUP|QUIT/i.test(signal)) {

package/src/services/terminal-provider-exit.ts

@@ -8,6 +8,7 @@ import { isRecord } from "agent-relay-sdk";
 import { clearAgentMetaKeys } from "../db";
 import { notifyAgentOffline } from "../agent-lifecycle-events";
 import { diagnoseAgentDeath } from "../agent-death-diagnosis";
+import { routeNotification } from "../notification-router";
 import type { AgentCard } from "../types";
 
 /**
@@ -29,5 +30,27 @@ export function reconcileTerminalProviderExit(agentId: string, before: AgentCard
   if (!isRecord(marker)) return;
   const reasonText = typeof marker.reason === "string" ? marker.reason : "provider exited";
   const reason = `provider exited (${reasonText})`;
-  notifyAgentOffline(agentId, reason, diagnoseAgentDeath({ agent: after, reason }));
+  const diagnosis = diagnoseAgentDeath({ agent: after, reason });
+  notifyAgentOffline(agentId, reason, diagnosis);
+  if (!after.spawnedBy && reasonText.startsWith("self-resume-")) {
+    routeNotification({
+      type: "agent.exited",
+      scope: { agentId },
+      recipients: ["user"],
+      message: {
+        subject: "Self-resume failed",
+        body: `Agent \`${agentId}\` exited during self-resume before the resumed session was confirmed. Manual recovery may need the provider resume id from the runner log.`,
+        payload: {
+          kind: "agent.self_resume_failed",
+          agentId,
+          reason,
+          diagnosis,
+          ...(typeof marker.claudeResumeId === "string" ? { claudeResumeId: marker.claudeResumeId } : {}),
+          ...(typeof marker.selfResumeCommandId === "string" ? { commandId: marker.selfResumeCommandId } : {}),
+          ...(typeof marker.selfResumeGeneration === "number" ? { generation: marker.selfResumeGeneration } : {}),
+        },
+        replyExpected: false,
+      },
+    });
+  }
 }