fix(agents): reconcile stale restart-aborted subagent runs instead of resurrecting them

ly-wang19 · claude · ly-wang19 · commit 7cf5cc80c728 · 2026-06-09T12:25:04.000+08:00
restore reconciliation only treated a run as stale-unended when abortedLastRun was not true, so restart-aborted runs were exempted from the shared liveness policy and auto-recovered no matter how old. After long gateway downtime this resurrected hours-old aborted subagent runs. Drop the exemption so the existing isStaleUnendedSubagentRun window (2h or run timeout+grace) applies to aborted runs too; fresh restart-aborted runs are not yet stale and still recover. Fixes #90766 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
diff --git a/src/agents/subagent-registry-helpers.ts b/src/agents/subagent-registry-helpers.ts
@@ -190,9 +190,11 @@ export function resolveSubagentRunOrphanReason(params: {
     }
     if (
       params.includeStaleUnended === true &&
-      sessionEntry.abortedLastRun !== true &&
       isStaleUnendedSubagentRun(params.entry, params.now)
     ) {
+      // Stale-unended pruning also covers restart-aborted runs: the shared liveness window
+      // keeps fresh restart recovery (recent aborts are not yet stale) while a long-downtime
+      // aborted run is reconciled instead of auto-resurrected after the gateway comes back (#90766).
       return "stale-unended-run";
     }
     return null;
diff --git a/src/agents/subagent-registry.persistence.test.ts b/src/agents/subagent-registry.persistence.test.ts
@@ -724,21 +724,63 @@ describe("subagent registry persistence", () => {
     expect(listSubagentRunsForRequester("agent:main:main")).toHaveLength(0);
   });
 
-  it("keeps stale unended restored runs with abortedLastRun for restart recovery", async () => {
+  it("reconciles stale unended restored runs even when restart-aborted (#90766)", async () => {
+    const now = Date.now();
+    const runId = "run-stale-aborted-restore";
+    const childSessionKey = "agent:main:subagent:stale-aborted-restore";
+    const registryPath = await writePersistedRegistry(
+      {
+        version: 2,
+        runs: {
+          [runId]: {
+            runId,
+            childSessionKey,
+            requesterSessionKey: "agent:main:main",
+            requesterDisplayKey: "main",
+            task: "stale restart-aborted work",
+            cleanup: "keep",
+            createdAt: now - 3 * 60 * 60 * 1_000,
+            startedAt: now - 3 * 60 * 60 * 1_000,
+          },
+        },
+      },
+      { seedChildSessions: false },
+    );
+    await writeChildSessionEntry({
+      sessionKey: childSessionKey,
+      sessionId: "sess-stale-aborted-restore",
+      updatedAt: now,
+      abortedLastRun: true,
+    });
+
+    restartRegistry();
+    await waitForRegistryWork(async () => {
+      const after = JSON.parse(await fs.readFile(registryPath, "utf8")) as {
+        runs?: Record<string, unknown>;
+      };
+      return after.runs?.[runId] === undefined;
+    });
+
+    // Long-downtime aborted run is reconciled, not auto-resumed.
+    expect(callGateway).not.toHaveBeenCalled();
+    expect(listSubagentRunsForRequester("agent:main:main")).toHaveLength(0);
+  });
+
+  it("recovers fresh restart-aborted runs that are not yet stale (#90766)", async () => {
     vi.mocked(callGateway).mockImplementationOnce(async (request) => {
       expectFields(request, {
         method: "agent.wait",
       });
       expectFields((request as { params?: unknown }).params, {
-        runId: "run-stale-aborted-restore",
+        runId: "run-fresh-aborted-restore",
       });
       return {
         status: "pending",
       };
     });
     const now = Date.now();
-    const runId = "run-stale-aborted-restore";
-    const childSessionKey = "agent:main:subagent:stale-aborted-restore";
+    const runId = "run-fresh-aborted-restore";
+    const childSessionKey = "agent:main:subagent:fresh-aborted-restore";
     await writePersistedRegistry(
       {
         version: 2,
@@ -748,18 +790,18 @@ describe("subagent registry persistence", () => {
             childSessionKey,
             requesterSessionKey: "agent:main:main",
             requesterDisplayKey: "main",
-            task: "stale restart-recoverable work",
+            task: "fresh restart-recoverable work",
             cleanup: "keep",
-            createdAt: now - 3 * 60 * 60 * 1_000,
-            startedAt: now - 3 * 60 * 60 * 1_000,
+            createdAt: now - 60_000,
+            startedAt: now - 60_000,
           },
         },
       },
       { seedChildSessions: false },
     );
     await writeChildSessionEntry({
       sessionKey: childSessionKey,
-      sessionId: "sess-stale-aborted-restore",
+      sessionId: "sess-fresh-aborted-restore",
       updatedAt: now,
       abortedLastRun: true,
     });

Original file line number	Diff line number	Diff line change
`@@ -190,9 +190,11 @@ export function resolveSubagentRunOrphanReason(params: {`
`190`	`190`	`}`
`191`	`191`	`if (`
`192`	`192`	`params.includeStaleUnended === true &&`
`193`		`- sessionEntry.abortedLastRun !== true &&`
`194`	`193`	`isStaleUnendedSubagentRun(params.entry, params.now)`
`195`	`194`	`) {`
	`195`	`+ // Stale-unended pruning also covers restart-aborted runs: the shared liveness window`
	`196`	`+ // keeps fresh restart recovery (recent aborts are not yet stale) while a long-downtime`
	`197`	`+ // aborted run is reconciled instead of auto-resurrected after the gateway comes back (#90766).`
`196`	`198`	`return "stale-unended-run";`
`197`	`199`	`}`
`198`	`200`	`return null;`