@qwen-code/qwen-code 0.21.2 → 0.21.3-nightly.20260803.e1e5b42ce

package/chunks/{run-qwen-serve-RN2R6QQW.js → run-qwen-serve-MTVRAUBR.js}

@@ -1,9 +1,8 @@
 // Force strict mode and setup for ESM
 "use strict";
 import {
-  DaemonMetricsRing,
   computeCpuPercent
-} from "./chunk-HGT6JR3U.js";
+} from "./chunk-LBOVL47Y.js";
 import {
   RUNTIME_STARTUP_CANCELLED_MESSAGE
 } from "./chunk-QUPXZXLV.js";
@@ -34,11 +33,16 @@ import {
   prepareManagedScratchRoot,
   require_express,
   resolveWebShellDir,
-  setDeferredRuntimeRequestTiming
-} from "./chunk-MB32MS4U.js";
+  setDeferredRuntimeRequestTiming,
+  toDaemonStatusMemoryLimits
+} from "./chunk-ZLJQIGCH.js";
 import {
   isLoopbackBind
 } from "./chunk-RXFQM6FQ.js";
+import {
+  formatMemoryBudgetStderr,
+  resolveDaemonMemoryBudget
+} from "./chunk-HGCQ5SE7.js";
 import {
   DEFAULT_COMPACTED_REPLAY_MAX_BYTES,
   DEFAULT_MAX_JOURNAL_BYTES,
@@ -47,10 +51,12 @@ import {
   normalizeMaxJournalBytes,
   normalizeMaxJournalEvents
 } from "./chunk-XCGIMOLN.js";
+import "./chunk-K6KS3LTN.js";
 import "./chunk-FC4KQFCS.js";
+import "./chunk-T3CHYWJZ.js";
 import {
   resolveChannelCwd
-} from "./chunk-FANRZEAJ.js";
+} from "./chunk-VWOVHOTT.js";
 import {
   loadServeFastPathSettings,
   preResolveServeFastPathHomeEnvOverrides
@@ -59,39 +65,37 @@ import {
   MAX_REGISTERED_WORKSPACES,
   resolveWorkspaceInputs,
   workspaceRegistrationId
-} from "./chunk-LYI2PZDK.js";
-import "./chunk-E2FN6SM7.js";
+} from "./chunk-A5LNZGJB.js";
+import "./chunk-CIUUYB23.js";
 import {
   LOAD_REPLAY_META_KEY
-} from "./chunk-CPZAH673.js";
-import "./chunk-B4AJNNOL.js";
+} from "./chunk-UKZEXS6K.js";
+import "./chunk-J26HHFI5.js";
 import {
   canonicalizeWorkspace,
   translateAndCheckAbsoluteWorkspacePath
-} from "./chunk-2RFKHEXL.js";
-import "./chunk-FCYYAB4S.js";
+} from "./chunk-6PLDPT2C.js";
+import "./chunk-FHPKHXHT.js";
 import "./chunk-FYJ7W2PI.js";
 import {
   ChannelDeliveryError,
   ChannelWebhookEnqueueError,
   QWEN_SERVER_TOKEN_ENV,
   isChannelDeliveryError
-} from "./chunk-U4VTUMMK.js";
+} from "./chunk-QONSRAEV.js";
 import {
   normalizeWorkerDiagnostic,
   sanitizeWorkerDiagnostic
-} from "./chunk-6MPDY6ZG.js";
+} from "./chunk-24ER7UV4.js";
 import {
   channelSelectionNames
 } from "./chunk-ALMR3E6Q.js";
 import "./chunk-BK2UNYWD.js";
 import "./chunk-M5GB774H.js";
 import "./chunk-5HBA2Z7V.js";
-import "./chunk-K6KS3LTN.js";
-import "./chunk-T3CHYWJZ.js";
 import {
   sanitizeLogText
-} from "./chunk-JCJ7LKNA.js";
+} from "./chunk-DIEFUBV6.js";
 import "./chunk-UDXSXLER.js";
 import "./chunk-GOXKNSDZ.js";
 import {
@@ -99,18 +103,15 @@ import {
   resolveConfigPathLite
 } from "./chunk-IDS7MSUP.js";
 import "./chunk-6PEXNQTX.js";
-import "./chunk-ALGGS7UH.js";
-import {
-  isWithinRoot
-} from "./chunk-MR3PXB6E.js";
 import {
   getCliVersion
-} from "./chunk-2KKCCLPT.js";
+} from "./chunk-C4IUDGPI.js";
 import "./chunk-VJNBRRZG.js";
+import "./chunk-ALGGS7UH.js";
 import {
-  writeStderrLine,
-  writeStdoutLine
-} from "./chunk-OWAK7LMW.js";
+  isWithinRoot
+} from "./chunk-MR3PXB6E.js";
+import "./chunk-GOFAQQZA.js";
 import "./chunk-5M6IDOMF.js";
 import "./chunk-TWPJO254.js";
 import "./chunk-H6XPXXMH.js";
@@ -124,6 +125,10 @@ import {
   profileCheckpoint
 } from "./chunk-QSWDGAIC.js";
 import "./chunk-Q7YDKOGY.js";
+import {
+  writeStderrLine,
+  writeStdoutLine
+} from "./chunk-OWAK7LMW.js";
 import "./chunk-DMTGGOSA.js";
 import "./chunk-YQ3U5MUC.js";
 import {
@@ -144,6 +149,10 @@ import * as path from "node:path";
 import * as os from "node:os";
 import { monitorEventLoopDelay, performance as performance2 } from "node:perf_hooks";
 
+// packages/core/dist/src/memory/scopes.js
+init_esbuild_shims();
+var MEMORY_PROJECT_SCOPES = ["git-root", "workspace"];
+
 // packages/cli/src/serve/permission-audit.ts
 init_esbuild_shims();
 var DEFAULT_PERMISSION_AUDIT_RING_SIZE = 512;
@@ -1622,6 +1631,156 @@ ${capUtf8Record(
 }
 __name(initDaemonLogger, "initDaemonLogger");
 
+// packages/cli/src/serve/daemon-metrics-ring.ts
+init_esbuild_shims();
+var MAX_SAMPLES_PER_WINDOW = 4096;
+var DaemonMetricsRing = class {
+  static {
+    __name(this, "DaemonMetricsRing");
+  }
+  capacity;
+  buckets = [];
+  curRequests = 0;
+  curErrors = 0;
+  curDurations = [];
+  curPromptsCompleted = 0;
+  curQueueWaits = [];
+  curPromptDurations = [];
+  curLlmDurations = [];
+  curLlmApiErrors = 0;
+  curLlmApiRetries = 0;
+  curPipeInBytes = 0;
+  curPipeOutBytes = 0;
+  curTokensIn = 0;
+  curTokensOut = 0;
+  constructor(options) {
+    this.capacity = Math.max(1, Math.floor(options.capacity));
+  }
+  /** Fold one completed HTTP request into the open window. */
+  recordRequest(durationMs, statusCode) {
+    this.curRequests += 1;
+    if (statusCode >= 400) this.curErrors += 1;
+    pushCapped(this.curDurations, durationMs);
+  }
+  /** Fold one prompt's queue wait (dispatched from the per-session FIFO). */
+  recordPromptQueueWait(durationMs) {
+    pushCapped(this.curQueueWaits, durationMs);
+  }
+  /** Fold one finished prompt's end-to-end duration (also counts throughput). */
+  recordPromptDuration(durationMs) {
+    this.curPromptsCompleted += 1;
+    pushCapped(this.curPromptDurations, durationMs);
+  }
+  /** Fold one model round's LLM API round-trip time (from the token frame). */
+  recordLlmDuration(durationMs) {
+    pushCapped(this.curLlmDurations, durationMs);
+  }
+  /**
+   * Fold one model round's per-round model-API-error and automatic-retry
+   * increments (from the same token frame) into the open window. Both are
+   * plain counters — a burst in one window accrues exactly, past the sample
+   * cap that only bounds the percentile arrays. Non-finite / negative inputs
+   * are ignored so a malformed frame cannot poison the totals.
+   */
+  recordApiActivity(errors, retries) {
+    if (Number.isFinite(errors) && errors > 0) this.curLlmApiErrors += errors;
+    if (Number.isFinite(retries) && retries > 0) {
+      this.curLlmApiRetries += retries;
+    }
+  }
+  /** Fold one ACP-child pipe message's payload size into the open window. */
+  recordPipe(direction, bytes) {
+    if (!Number.isFinite(bytes) || bytes < 0) return;
+    if (direction === "inbound") this.curPipeInBytes += bytes;
+    else this.curPipeOutBytes += bytes;
+  }
+  /** Fold one model turn's token usage into the open window. */
+  recordTokens(inputTokens, outputTokens) {
+    if (Number.isFinite(inputTokens) && inputTokens > 0) {
+      this.curTokensIn += inputTokens;
+    }
+    if (Number.isFinite(outputTokens) && outputTokens > 0) {
+      this.curTokensOut += outputTokens;
+    }
+  }
+  /**
+   * Seal the open window into a bucket stamped at `now` (folding in the
+   * gauges), append it, evict the oldest past capacity, then reset the
+   * accumulators for the next window.
+   */
+  sample(now, gauges) {
+    this.buckets.push({
+      t: now,
+      activeSessions: finiteGauge(gauges.activeSessions),
+      activePrompts: finiteGauge(gauges.activePrompts),
+      queuedPrompts: finiteGauge(gauges.queuedPrompts),
+      requests: this.curRequests,
+      errors: this.curErrors,
+      latencyP50Ms: percentile(this.curDurations, 0.5),
+      latencyP95Ms: percentile(this.curDurations, 0.95),
+      promptsCompleted: this.curPromptsCompleted,
+      promptQueueWaitP95Ms: percentile(this.curQueueWaits, 0.95),
+      promptDurationP95Ms: percentile(this.curPromptDurations, 0.95),
+      llmApiP50Ms: percentile(this.curLlmDurations, 0.5),
+      llmApiP95Ms: percentile(this.curLlmDurations, 0.95),
+      llmApiErrors: this.curLlmApiErrors,
+      llmApiRetries: this.curLlmApiRetries,
+      cpuPercent: finiteGauge(gauges.cpuPercent),
+      rssBytes: finiteGauge(gauges.rssBytes),
+      heapUsedBytes: finiteGauge(gauges.heapUsedBytes),
+      eventLoopLagP99Ms: finiteGauge(gauges.eventLoopLagP99Ms),
+      pipeInBytes: this.curPipeInBytes,
+      pipeOutBytes: this.curPipeOutBytes,
+      sseConnections: finiteGauge(gauges.sseConnections),
+      wsConnections: finiteGauge(gauges.wsConnections),
+      acpConnections: finiteGauge(gauges.acpConnections),
+      rateLimitRejected: finiteGauge(gauges.rateLimitRejected),
+      tokensIn: this.curTokensIn,
+      tokensOut: this.curTokensOut,
+      childCpuPercent: finiteGauge(gauges.childCpuPercent),
+      childRssBytes: finiteGauge(gauges.childRssBytes)
+    });
+    if (this.buckets.length > this.capacity) {
+      this.buckets.splice(0, this.buckets.length - this.capacity);
+    }
+    this.curRequests = 0;
+    this.curErrors = 0;
+    this.curDurations.length = 0;
+    this.curPromptsCompleted = 0;
+    this.curQueueWaits.length = 0;
+    this.curPromptDurations.length = 0;
+    this.curLlmDurations.length = 0;
+    this.curLlmApiErrors = 0;
+    this.curLlmApiRetries = 0;
+    this.curPipeInBytes = 0;
+    this.curPipeOutBytes = 0;
+    this.curTokensIn = 0;
+    this.curTokensOut = 0;
+  }
+  /** Oldest→newest copy of the retained buckets (safe to serialize). */
+  snapshot() {
+    return this.buckets.slice();
+  }
+};
+function finiteGauge(value) {
+  return Number.isFinite(value) ? value : 0;
+}
+__name(finiteGauge, "finiteGauge");
+function pushCapped(target, value) {
+  if (Number.isFinite(value) && value >= 0 && target.length < MAX_SAMPLES_PER_WINDOW) {
+    target.push(value);
+  }
+}
+__name(pushCapped, "pushCapped");
+function percentile(samples, q) {
+  if (samples.length === 0) return 0;
+  const sorted = samples.slice().sort((a, b) => a - b);
+  const rank = Math.ceil(q * sorted.length);
+  const idx = Math.min(sorted.length - 1, Math.max(0, rank - 1));
+  return sorted[idx];
+}
+__name(percentile, "percentile");
+
 // packages/cli/src/serve/large-pipe-frame-observer.ts
 init_esbuild_shims();
 var LARGE_PIPE_FRAME_THRESHOLD_BYTES = 256 * 1024;
@@ -2371,7 +2530,7 @@ var FAST_PATH_RUNTIME_START_AFTER_HEALTH_MS = 50;
 var FAST_PATH_RUNTIME_START_FALLBACK_MS = 1e3;
 var RUNTIME_STARTUP_TIMEOUT_ENV = "QWEN_SERVE_RUNTIME_STARTUP_TIMEOUT_MS";
 var MAX_EVENT_RING_SIZE = 1e6;
-var DEFAULT_MAX_SESSIONS = 20;
+var DEFAULT_MAX_SESSIONS = 32;
 var DEFAULT_MAX_PENDING_PROMPTS_PER_SESSION = 5;
 var DEFAULT_EVENT_RING_SIZE = 8e3;
 var DEFAULT_SESSION_IDLE_TIMEOUT_MS = 30 * 6e4;
@@ -2537,6 +2696,31 @@ function sessionArtifactsPersistenceAvailableFromSettings(settings) {
   return settings?.general?.chatRecording !== false;
 }
 __name(sessionArtifactsPersistenceAvailableFromSettings, "sessionArtifactsPersistenceAvailableFromSettings");
+function subSessionConcurrencyCapsFromSettings(serve, onWarning = writeStderrLine) {
+  const asCap = /* @__PURE__ */ __name((key, value) => {
+    if (value === void 0) return void 0;
+    if (typeof value === "number" && Number.isInteger(value) && value >= 1) {
+      return value;
+    }
+    onWarning(
+      `qwen serve: ignoring invalid ${key} (${JSON.stringify(value)}); expected a positive integer, falling back to the built-in default.`
+    );
+    return void 0;
+  }, "asCap");
+  const maxConcurrentPerCaller = asCap(
+    "maxConcurrentSubSessionsPerCaller",
+    serve.maxConcurrentSubSessionsPerCaller
+  );
+  const maxConcurrentTotal = asCap(
+    "maxConcurrentSubSessionsTotal",
+    serve.maxConcurrentSubSessionsTotal
+  );
+  return {
+    ...maxConcurrentPerCaller !== void 0 ? { maxConcurrentPerCaller } : {},
+    ...maxConcurrentTotal !== void 0 ? { maxConcurrentTotal } : {}
+  };
+}
+__name(subSessionConcurrencyCapsFromSettings, "subSessionConcurrencyCapsFromSettings");
 var settingsWriteLocks = /* @__PURE__ */ new Map();
 function withSettingsLock(workspace, fn) {
   const prev = settingsWriteLocks.get(workspace) ?? Promise.resolve();
@@ -2549,12 +2733,12 @@ var retryableChannelWorkerShutdownErrors = /* @__PURE__ */ new WeakSet();
 var channelWorkerRuntimePromise;
 async function loadChannelWorkerRuntime() {
   channelWorkerRuntimePromise ??= Promise.all([
-    import("./channel-worker-supervisor-CABBXYHK.js"),
-    import("./pidfile-NL7MKDRU.js"),
-    import("./runtime-QFJ2MJF5.js"),
+    import("./channel-worker-supervisor-ATGFTZRN.js"),
+    import("./pidfile-PCADVJQW.js"),
+    import("./runtime-DQGE2DYR.js"),
     import("./cli-entry-path-4VJ3Y6T2.js"),
-    import("./channel-worker-group-OVWGYMZS.js"),
-    import("./channel-worker-manager-H55GKOQA.js")
+    import("./channel-worker-group-NDG4WOOZ.js"),
+    import("./channel-worker-manager-ZCW6UW34.js")
   ]).then(
     ([
       supervisor,
@@ -2659,7 +2843,7 @@ function shouldPreheatBridge(deps) {
 __name(shouldPreheatBridge, "shouldPreheatBridge");
 var coreRuntimePromise;
 function loadCoreRuntime() {
-  coreRuntimePromise ??= import("./core-runtime-6N4EFC2Q.js");
+  coreRuntimePromise ??= import("./core-runtime-2VCZN5QF.js");
   return coreRuntimePromise;
 }
 __name(loadCoreRuntime, "loadCoreRuntime");
@@ -2689,10 +2873,10 @@ __name(resolveDaemonLogBaseDirForRun, "resolveDaemonLogBaseDirForRun");
 var settingsRuntimePromise;
 function loadSettingsRuntimeModules() {
   settingsRuntimePromise ??= Promise.all([
-    import("./settings-LN7RNKBS.js"),
-    import("./environment-SWL3CMB3.js"),
-    import("./loadedSettingsAdapter-ZQAV5NDO.js"),
-    import("./trustedFolders-KOPSP44S.js")
+    import("./settings-77MDDUSX.js"),
+    import("./environment-FE5AX2U5.js"),
+    import("./loadedSettingsAdapter-EZYPDX2J.js"),
+    import("./trustedFolders-3M4IQ6K6.js")
   ]).then(([settings, environment, loadedSettingsAdapter, trustedFolders]) => ({
     settings,
     environment,
@@ -2705,8 +2889,8 @@ __name(loadSettingsRuntimeModules, "loadSettingsRuntimeModules");
 var channelWebhookConfigRuntimePromise;
 function loadChannelWebhookConfigRuntime() {
   channelWebhookConfigRuntimePromise ??= Promise.all([
-    import("./runtime-QFJ2MJF5.js"),
-    import("./config-utils-WADIDWGG.js")
+    import("./runtime-DQGE2DYR.js"),
+    import("./config-utils-E3ZQTHGM.js")
   ]).then(([channelRuntime, configUtils]) => ({
     loadChannelsConfig: channelRuntime.loadChannelsConfig,
     parseChannelWebhookConfig: configUtils.parseChannelWebhookConfig
@@ -2731,17 +2915,17 @@ async function loadServeRuntimeModules() {
     totalSessionAdmissionModule,
     workspaceRegistryModule
   ] = await Promise.all([
-    import("./server-EP2LLAPD.js"),
-    import("./bridge-W7AZSRZQ.js"),
-    import("./spawnChannel-YPCU62DI.js"),
+    import("./server-S22MLWTJ.js"),
+    import("./bridge-XNAWEDAN.js"),
+    import("./spawnChannel-TOWYOVDZ.js"),
     import("./process-registry-2CHQBXH5.js"),
-    import("./workspace-service-25ML3T47.js"),
+    import("./workspace-service-TEV5Q2AA.js"),
     import("./types-ML3TRJQ5.js"),
-    import("./daemon-status-provider-DDF7C5M7.js"),
-    import("./workspace-providers-status-YKLTLMJA.js"),
-    import("./workspace-skills-status-YKZWOYRX.js"),
-    import("./total-session-admission-5CALIVYO.js"),
-    import("./workspace-registry-PTYUSYXX.js")
+    import("./daemon-status-provider-F2YGDKHR.js"),
+    import("./workspace-providers-status-QKKXQHL7.js"),
+    import("./workspace-skills-status-VRL55OMU.js"),
+    import("./total-session-admission-LY3RMXXI.js"),
+    import("./workspace-registry-KC66O6DH.js")
   ]);
   return {
     createServeApp: serverModule.createServeApp,
@@ -3115,7 +3299,8 @@ function createBootstrapServeApp(input) {
         writerIdleTimeoutMs: positiveFiniteOrNull(opts.writerIdleTimeoutMs),
         channelIdleTimeoutMs: channelIdleTimeoutMs(opts.channelIdleTimeoutMs),
         sessionIdleTimeoutMs: sessionIdleTimeoutMs(opts.sessionIdleTimeoutMs),
-        acpConnectionCap: null
+        acpConnectionCap: null,
+        memory: toDaemonStatusMemoryLimits(opts.daemonMemoryBudget)
       },
       capabilities: {
         protocolVersions: getServeProtocolVersions(),
@@ -3403,9 +3588,19 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
       status: deps.bridge ? "external_bridge" : shouldPreheat ? "scheduled" : "not_scheduled"
     }
   };
+  if (optsIn.memoryProjectScope !== void 0 && !MEMORY_PROJECT_SCOPES.includes(
+    optsIn.memoryProjectScope
+  )) {
+    throw new TypeError(
+      `Invalid memoryProjectScope: ${String(optsIn.memoryProjectScope)}. Must be "git-root" or "workspace".`
+    );
+  }
   preResolveServeFastPathHomeEnvOverrides();
   const daemonRuntimeBaseEnv = Object.freeze({
-    ...process.env
+    ...process.env,
+    ...optsIn.memoryProjectScope !== void 0 ? {
+      QWEN_CODE_MEMORY_PROJECT_SCOPE: optsIn.memoryProjectScope
+    } : {}
   });
   const rawToken = optsIn.token ?? process.env[QWEN_SERVER_TOKEN_ENV];
   const token = typeof rawToken === "string" && rawToken.trim().length > 0 ? rawToken.trim() : void 0;
@@ -3632,9 +3827,15 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
       `At most ${MAX_REGISTERED_WORKSPACES} --workspace values may be registered.`
     );
   }
+  opts.daemonMemoryBudget = resolveDaemonMemoryBudget({
+    budgetMb: opts.memoryBudgetMb
+  });
+  if (opts.daemonMemoryBudget.budgetSource === "flag" || opts.daemonMemoryBudget.insufficientMemory) {
+    writeStderrLine(formatMemoryBudgetStderr(opts.daemonMemoryBudget));
+  }
   let workspaceRegistrationStore = deps.workspaceRegistrationStore;
   if (workspaceRegistrationStore === void 0 && process.env["QWEN_SERVE_NO_PERSISTENT_REGISTRATION"] !== "1") {
-    const { WorkspaceRegistrationStore } = await import("./workspace-registration-store-5DMGKU64.js");
+    const { WorkspaceRegistrationStore } = await import("./workspace-registration-store-7J7KOXFD.js");
     workspaceRegistrationStore = new WorkspaceRegistrationStore(boundWorkspace);
   }
   if (workspaceRegistrationStore) {
@@ -4145,7 +4346,7 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
       loadCoreRuntime(),
       loadSettingsRuntimeModules(),
       cliVersionPromise,
-      import("./daemon-trust-policy-Q4KSOJTR.js")
+      import("./daemon-trust-policy-XCY7JETR.js")
     ]);
     cliVersion = resolvedCliVersion;
     settingsRuntime.environment.preResolveHomeEnvOverrides();
@@ -4510,7 +4711,7 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
         resolveSkillSettings,
         skillSettingStrings,
         updateWorkspaceSkillSettingLists
-      } = await import("./skill-settings-GMR3GMNY.js");
+      } = await import("./skill-settings-STR5NF4B.js");
       const fresh = loadSettingsForPersistence(workspace);
       const normalizedName = skillName.trim().toLowerCase();
       const resolved = resolveSkillSettings(fresh);
@@ -4609,10 +4810,13 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
         );
       }
     }), "persistSettingsFn");
-    const { createSubSessionLauncher } = await import("./create-sub-session-AEREHZBQ.js");
+    const { createSubSessionLauncher } = await import("./create-sub-session-52OUA2VU.js");
     const subSessionLauncher = createSubSessionLauncher({
       getBridge: /* @__PURE__ */ __name(() => bridgeRef, "getBridge"),
-      boundWorkspace
+      boundWorkspace,
+      ...subSessionConcurrencyCapsFromSettings(
+        runtimeBootSettings?.merged.serve ?? {}
+      )
     });
     const bridge = deps.bridge ?? runtime.createAcpSessionBridge({
       // Reverse tool channel: let `BridgeClient.extMethod` reach the WS
@@ -4911,7 +5115,10 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
       let secondaryBridgeRef;
       const secondarySubSessionLauncher = createSubSessionLauncher({
         getBridge: /* @__PURE__ */ __name(() => secondaryBridgeRef, "getBridge"),
-        boundWorkspace: workspaceInput.cwd
+        boundWorkspace: workspaceInput.cwd,
+        ...subSessionConcurrencyCapsFromSettings(
+          secondarySettings?.merged.serve ?? {}
+        )
       });
       const secondaryBridge = runtime.createAcpSessionBridge({
         clientMcpSender: secondaryClientMcpSenderRegistry.lookup,
@@ -5248,7 +5455,10 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
       let wsBridgeRef;
       const wsSubSessionLauncher = createSubSessionLauncher({
         getBridge: /* @__PURE__ */ __name(() => wsBridgeRef, "getBridge"),
-        boundWorkspace: cwd
+        boundWorkspace: cwd,
+        ...subSessionConcurrencyCapsFromSettings(
+          wsSettings?.merged.serve ?? {}
+        )
       });
       let wsBridge;
       try {
@@ -5710,8 +5920,8 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
           { createChannelManagementService },
           { WorkspaceChannelSettingsStore }
         ] = await Promise.all([
-          import("./channel-management-service-5S7YCWX5.js"),
-          import("./channel-settings-store-JFVRSNER.js")
+          import("./channel-management-service-UHX56NBO.js"),
+          import("./channel-settings-store-7P6RJSSF.js")
         ]);
         return createChannelManagementService({
           workspaceCwd: targetRuntime.workspaceCwd,
@@ -5907,8 +6117,8 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
         { createWorkspaceTrustReconciler },
         { createDaemonTrustPolicyMonitor }
       ] = await Promise.all([
-        import("./workspace-trust-reconciler-5FI2S3OS.js"),
-        import("./daemon-trust-policy-monitor-V6FTE3IY.js")
+        import("./workspace-trust-reconciler-COC5M7CY.js"),
+        import("./daemon-trust-policy-monitor-FBHB5GMQ.js")
       ]);
       const materializationFor = /* @__PURE__ */ __name((entry, snapshot, decision) => {
         const boundWorkspaces2 = entry.primary && decision.targetTrusted ? runtime.resolveBoundWorkspacesFromIdeEnv(
@@ -6115,7 +6325,7 @@ async function runQwenServeImpl(optsIn, deps, loggerLifecycle) {
     );
   }
   const channelValidationSettingsRuntime = opts.channelSelection ? await loadSettingsRuntimeModules() : void 0;
-  const channelValidationTrustPolicy = opts.channelSelection ? await import("./daemon-trust-policy-Q4KSOJTR.js") : void 0;
+  const channelValidationTrustPolicy = opts.channelSelection ? await import("./daemon-trust-policy-XCY7JETR.js") : void 0;
   const channelValidationTrustSnapshot = channelValidationTrustPolicy ? await channelValidationTrustPolicy.readDaemonTrustPolicySnapshot() : void 0;
   const resolveChannelWorkspaceGroupsAtListen = /* @__PURE__ */ __name(() => {
     if (!opts.channelSelection || !channelValidationSettingsRuntime || !channelRuntime) {
@@ -7026,6 +7236,7 @@ export {
   formatChannelWorkerDaemonUrl,
   resolveRuntimeStartupTimeoutMs,
   runQwenServe,
+  subSessionConcurrencyCapsFromSettings,
   validateAndCanonicalizeWorkspaceInput,
   validatePolicyConfig,
   waitForRuntimeStartingForShutdown