@11agents/cli 0.1.39 → 0.1.40

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@11agents/cli",
-  "version": "0.1.39",
+  "version": "0.1.40",
   "description": "11agents local runtime and telemetry CLI",
   "type": "module",
   "bin": {

package/src/commands/runtime.js

@@ -1653,12 +1653,31 @@ async function claimAndRunRuntimeTasks(registration, flags, deps, handlerModule,
 }
 
 async function drainRuntimeTasks(registration, flags, deps, handlerModule, retryState = createRetryState(), heartbeatIntervalMs = 15000, maxConcurrent = 1) {
+  if (!handlerModule) return 0
+  const runtimes = (registration?.runtimes || []).filter(r => r?.id)
+  if (!runtimes.length) return 0
+  const config = configFromFlags(flags)
+  const machineKey = registration?.machine?.machine_key || machineOverride(flags) || ''
+
+  const perRuntime = Math.max(1, (deps.cpuCount ?? os.cpus().length) - 1)
+  const effectiveConcurrent = Math.max(maxConcurrent, runtimes.length * perRuntime)
+
   let completed = 0
-  while (true) {
-    const claimed = await claimAndRunRuntimeTasks(registration, flags, deps, handlerModule, retryState, heartbeatIntervalMs, maxConcurrent)
-    completed += claimed
-    if (claimed === 0) return completed
-  }
+  // Worker pool: each worker loops claiming tasks until the queue is empty,
+  // so a fast worker immediately picks up the next task without waiting for
+  // slow workers to finish (unlike the old batched Promise.allSettled approach).
+  const workers = Array.from({ length: effectiveConcurrent }, (_, i) => {
+    const runtime = runtimes[i % runtimes.length]
+    return (async () => {
+      while (true) {
+        const ran = await runOneRuntimeTaskSlot(runtime, config, machineKey, registration, flags, deps, handlerModule, retryState, heartbeatIntervalMs)
+        if (!ran) break
+        completed++
+      }
+    })()
+  })
+  await Promise.allSettled(workers)
+  return completed
 }
 
 export async function startRuntimeDaemon(flags = {}, deps = {}) {