fix: refresh subagent idle timeout on tool-activity heartbeat

zeyuan.wang · zeyuan.wang · commit 619020dd23db · 2026-06-17T20:15:22.000+08:00
When a subagent performs long-running tool calls (e.g. web_fetch, read, exec), the LLM stream is idle. If no LLM token arrives within 120 seconds, the session is terminated with LLM idle timeout. This fix adds a tool-activity heartbeat mechanism: - A new module tool-activity-heartbeat.ts provides a subscription/ notification pair (onToolActivity / notifyToolActivity) modelled on the existing llm-request-activity pattern. - streamWithIdleTimeout (llm-idle-timeout.ts) now subscribes to both LLM-request activity AND tool-activity, resetting the idle timer via armTimer() on either signal. - attempt.ts calls notifyToolActivity(runAbortController.signal) after every completed tool call (both success and failure branches), so the idle watchdog sees activity even when the model is silent. The 30-second heartbeat window (TOOL_ACTIVITY_HEARTBEAT_MS = 30_000) is intentionally shorter than the default 120s idle timeout, giving slow tool calls multiple chances to refresh before the watchdog fires. Fixes #94124
diff --git a/src/agents/embedded-agent-runner/run/attempt.ts b/src/agents/embedded-agent-runner/run/attempt.ts
@@ -479,6 +479,7 @@ import {
   shouldTreatEmptyAssistantReplyAsSilent,
 } from "./incomplete-turn.js";
 import { resolveLlmIdleTimeoutMs, streamWithIdleTimeout } from "./llm-idle-timeout.js";
+import { notifyToolActivity } from "./tool-activity-heartbeat.js";
 import { resolveMessageMergeStrategy } from "./message-merge-strategy.js";
 import { installMessageToolOnlyTerminalHook } from "./message-tool-terminal.js";
 import { wrapStreamFnWithMessageTransform } from "./message-transform-stream-wrapper.js";
@@ -3677,6 +3678,7 @@ export async function runEmbeddedAttempt(
             result,
             timestamp: Date.now(),
           });
+          notifyToolActivity(runAbortController.signal);
           return result;
         } catch (error) {
           const message = formatErrorMessage(error);
@@ -3692,6 +3694,7 @@ export async function runEmbeddedAttempt(
             isError: true,
             timestamp: Date.now(),
           });
+          notifyToolActivity(runAbortController.signal);
           throw error;
         }
       };
diff --git a/src/agents/embedded-agent-runner/run/llm-idle-timeout.ts b/src/agents/embedded-agent-runner/run/llm-idle-timeout.ts
@@ -9,6 +9,7 @@ import {
 import { DEFAULT_LLM_IDLE_TIMEOUT_SECONDS } from "../../../config/agent-timeout-defaults.js";
 import type { OpenClawConfig } from "../../../config/types.openclaw.js";
 import { onLlmRequestActivity } from "../../../shared/llm-request-activity.js";
+import { onToolActivity } from "./tool-activity-heartbeat.js";
 import type { StreamFn } from "../../runtime/index.js";
 import type { MutableAssistantMessageEventStream } from "../../stream-compat.js";
 import { createStreamIteratorWrapper } from "../../stream-iterator-wrapper.js";
@@ -296,9 +297,11 @@ export function streamWithIdleTimeout(
             clearTimer();
           };
           const unsubscribeActivity = onLlmRequestActivity(streamAbortController.signal, armTimer);
+          const unsubscribeToolActivity = onToolActivity(streamAbortController.signal, armTimer);
           const cleanupIterator = () => {
             stopWaiting();
             unsubscribeActivity();
+            unsubscribeToolActivity();
             cleanupSourceSignal();
           };
 
diff --git a/src/agents/embedded-agent-runner/run/tool-activity-heartbeat.ts b/src/agents/embedded-agent-runner/run/tool-activity-heartbeat.ts
@@ -0,0 +1,55 @@
+/**
+ * Tool-activity heartbeat for subagent session idle timeout.
+ *
+ * When a subagent executes tool calls (web_fetch, read, exec, etc.), the LLM
+ * stream is idle.  The LLM idle-watchdog (llm-idle-timeout.ts) aborts the
+ * stream if no new tokens arrive within DEFAULT_LLM_IDLE_TIMEOUT_SECONDS (120s).
+ *
+ * This module lets the tool-execution path call `notifyToolActivity(signal)`
+ * after every completed tool-call so the watchdog sees activity and resets its
+ * 30-second sub-heartbeat — a tool that runs for minutes still keeps the
+ * session alive as long as it makes progress.
+ *
+ * The 30s window is intentionally shorter than the 120s default so a slow
+ * tool call has multiple chances to refresh the timer before the LLM watchdog
+ * fires.
+ */
+const TOOL_ACTIVITY_HEARTBEAT_MS = 30_000;
+
+const toolActivityListeners = new WeakMap<AbortSignal, Set<() => void>>();
+
+/**
+ * Subscribe to tool-activity heartbeats associated with a given AbortSignal.
+ * Returns an unsubscribe function.
+ */
+export function onToolActivity(
+  signal: AbortSignal,
+  listener: () => void,
+): () => void {
+  const listeners = toolActivityListeners.get(signal) ?? new Set<() => void>();
+  listeners.add(listener);
+  toolActivityListeners.set(signal, listeners);
+
+  return () => {
+    listeners.delete(listener);
+    if (listeners.size === 0) {
+      toolActivityListeners.delete(signal);
+    }
+  };
+}
+
+/**
+ * Notify all tool-activity listeners registered for the given AbortSignal.
+ */
+export function notifyToolActivity(
+  signal: AbortSignal | undefined,
+): void {
+  if (!signal) {
+    return;
+  }
+  for (const listener of toolActivityListeners.get(signal) ?? []) {
+    listener();
+  }
+}
+
+export { TOOL_ACTIVITY_HEARTBEAT_MS };
