fix(server): stop liveness pings on plugin sessions

Every DaemonClient pings its daemon every 10 s to notice a silently dead socket. A plugin's
session runs over its IPC channel (or an in-process queue for built-in plugins), whose loss is
reported directly, and the daemon already exempts plugin sockets from the application lease those
pings renew. The pings only woke both processes: about 2 wakeups a second in an idle daemon with
nine plugins and the built-ins.

DaemonClient gets a livenessHeartbeat option (default on); the plugin host turns it off.
Explicit verification (ensureConnected({ verify: true })) still pings.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

packages/client/src/daemon-client.test.ts

@@ -753,13 +753,14 @@ class DaemonClientSession {
   private readonly daemon = new FakeDaemon();
   private readonly client: DaemonClient;
 
-  constructor() {
+  constructor(options: { livenessHeartbeat?: boolean } = {}) {
     this.client = new DaemonClient({
       url: "ws://test",
       clientId: "clsk_heartbeat_test",
       logger: noopLogger,
       reconnect: { enabled: false },
       transportFactory: () => this.daemon.transport,
+      ...options,
     });
     clients.push(this.client);
   }
@@ -3257,6 +3258,18 @@ test("starts pinging one interval after connecting and holds the cadence", async
   expect(session.pingTimestamps()).toEqual(["10s", "20s", "30s"]);
 });
 
+test("a client with the liveness heartbeat off never pings on its own", async () => {
+  useHeartbeatClock();
+  const session = new DaemonClientSession({ livenessHeartbeat: false });
+  session.daemonAnswersPingsAfter("fast");
+
+  await session.connect();
+  await session.advance(60_000);
+
+  expect(session.state()).toEqual({ status: "connected" });
+  expect(session.pingTimestamps()).toEqual([]);
+});
+
 test("stops pinging once the connection is gone", async () => {
   useHeartbeatClock();
   const session = new DaemonClientSession();

packages/client/src/daemon-client.ts

@@ -469,6 +469,13 @@ export interface DaemonClientConfig {
   runtimeMetricsWindowMs?: number;
   trace?: DaemonClientTrace;
   capabilities?: Partial<Record<ClientCapability, unknown>>;
+  /**
+   * Ping the daemon every 10 s and reconnect after missed replies (default true). Turn it off
+   * only where the transport reports its own loss (a plugin's IPC channel, an in-process
+   * channel): there a ping cannot find a dead link the channel would not, and it costs both
+   * processes a wakeup per client every 10 s.
+   */
+  livenessHeartbeat?: boolean;
 }
 
 export interface DaemonClientTrace {
@@ -2720,7 +2727,11 @@ export class DaemonClient {
   }
 
   private scheduleNextLivenessHeartbeat(): void {
-    if (this.connectionState.status !== "connected" || this.livenessHeartbeatTimer) {
+    if (
+      this.config.livenessHeartbeat === false ||
+      this.connectionState.status !== "connected" ||
+      this.livenessHeartbeatTimer
+    ) {
       return;
     }
     this.livenessHeartbeatTimer = setTimeout(() => {

packages/server/src/server/plugins/plugin-process.ts

@@ -313,6 +313,9 @@ export function createPluginWorker(options: {
       // The runtime re-attaches a session when the daemon drops this socket.
       reconnect: { enabled: true },
       transportFactory,
+      // The channel is a pipe to the daemon (or an in-process queue for built-in plugins); its
+      // loss is reported directly, so a 10 s ping would only wake both sides.
+      livenessHeartbeat: false,
     });
     paseo = createPaseoApi(daemonClient);
     await daemonClient.connect();