Merge pull request #65 from bluk1020/fix/stuck-detector-workflow-timeout-fallback

fix: keep stuck detector active by default

.changeset/stuck-detector-workflow-timeout.md

@@ -0,0 +1,5 @@
+---
+"@runfusion/fusion": patch
+---
+
+Keep stuck task detection active by default with an explicit task-stuck timeout default, without coupling it to workflow step timeout settings.

packages/core/src/__tests__/settings-parity.test.ts

@@ -69,6 +69,11 @@ describe("settings key parity", () => {
     expect(DEFAULT_PROJECT_SETTINGS.completionDocumentationMode).toBe("off");
   });
 
+  it("keeps task stuck timeout active by default without coupling to workflow step timeout", () => {
+    expect(DEFAULT_PROJECT_SETTINGS.taskStuckTimeoutMs).toBe(600_000);
+    expect(DEFAULT_PROJECT_SETTINGS.workflowStepTimeoutMs).toBe(360_000);
+  });
+
   it("keeps github tracking keys in expected scopes with documented defaults", () => {
     expect(DEFAULT_PROJECT_SETTINGS.githubTrackingEnabledByDefault).toBe(false);
     expect(DEFAULT_PROJECT_SETTINGS.githubTrackingDefaultRepo).toBeUndefined();

packages/core/src/settings-schema.ts

@@ -214,7 +214,7 @@ export const DEFAULT_PROJECT_SETTINGS = {
   requirePlanApproval: false,
   specStalenessEnabled: false,
   specStalenessMaxAgeMs: 6 * 60 * 60 * 1000,
-  taskStuckTimeoutMs: undefined,
+  taskStuckTimeoutMs: 600_000,
   aiSessionTtlMs: 7 * 24 * 60 * 60 * 1000,
   aiSessionCleanupIntervalMs: 60 * 60 * 1000,
   autoUnpauseEnabled: true,

packages/core/src/types.ts

@@ -2180,7 +2180,7 @@ export interface ProjectSettings {
   /** Timeout in milliseconds for detecting stuck tasks. When a task's agent session
    *  shows no activity (no text deltas, tool calls, or progress updates) for longer
    *  than this duration, the task is considered stuck and will be terminated and retried.
-   *  Default: undefined (disabled). Suggested value: 600000 (10 minutes). */
+   *  Default: 600000 (10 minutes). Set to 0 to disable. */
   taskStuckTimeoutMs?: number;
   /** TTL in milliseconds for persisted AI planning/subtask/mission interview sessions.
    *  Sessions older than this cutoff are expired by the dashboard session cleanup loop.

packages/engine/src/__tests__/stuck-task-detector.test.ts

@@ -718,21 +718,52 @@ describe("StuckTaskDetector", () => {
       vi.useRealTimers();
     });
 
-    it("does nothing when timeout is disabled", async () => {
+    it("does not couple stuck detection to workflow step timeout", async () => {
       store = createMockStore({
-        getSettings: vi.fn().mockResolvedValue({ taskStuckTimeoutMs: undefined }),
+        getSettings: vi.fn().mockResolvedValue({
+          taskStuckTimeoutMs: undefined,
+          workflowStepTimeoutMs: 60_000,
+        }),
       });
-      const customDetector = new StuckTaskDetector(store);
+      const onStuck = vi.fn();
+      const customDetector = new StuckTaskDetector(store, { onStuck });
       const session = createMockSession();
 
       customDetector.trackTask("FN-001", session);
 
       vi.useFakeTimers({ shouldAdvanceTime: true });
-      vi.advanceTimersByTime(61000);
+      vi.advanceTimersByTime(61_000);
 
       await customDetector.checkNow();
 
-      expect(store.moveTask).not.toHaveBeenCalled();
+      expect(onStuck).not.toHaveBeenCalled();
+      expect(session.dispose).not.toHaveBeenCalled();
+
+      vi.useRealTimers();
+    });
+
+    it("kills stuck sessions when the project default stuck timeout is present", async () => {
+      store = createMockStore({
+        getSettings: vi.fn().mockResolvedValue({
+          taskStuckTimeoutMs: 600_000,
+          workflowStepTimeoutMs: 60_000,
+        }),
+      });
+      const onStuck = vi.fn();
+      const customDetector = new StuckTaskDetector(store, { onStuck });
+      const session = createMockSession();
+
+      customDetector.trackTask("FN-001", session);
+
+      vi.useFakeTimers({ shouldAdvanceTime: true });
+      vi.advanceTimersByTime(601_000);
+
+      await customDetector.checkNow();
+
+      expect(onStuck).toHaveBeenCalledWith(
+        expect.objectContaining({ taskId: "FN-001", reason: "inactivity" }),
+      );
+      expect(session.dispose).toHaveBeenCalled();
 
       vi.useRealTimers();
     });

packages/engine/src/stuck-task-detector.ts

@@ -11,7 +11,8 @@
  * - `recordProgress(taskId)` — step transitions (in-progress, done, skipped); resets counters
  *
  * The detector polls at a configurable interval and compares timestamps against
- * `taskStuckTimeoutMs` from settings.
+ * `taskStuckTimeoutMs` from settings. Project defaults keep this active by
+ * default while keeping workflow-step execution timeouts independent.
  */
 
 import type { TaskStore, Settings } from "@fusion/core";
@@ -471,7 +472,7 @@ export class StuckTaskDetector {
     if (settings.globalPause || settings.enginePaused) return;
 
     const timeoutMs = settings.taskStuckTimeoutMs;
-    if (!timeoutMs || timeoutMs <= 0) return; // Disabled
+    if (!timeoutMs || timeoutMs <= 0) return; // Disabled when task stuck timeout is explicitly unset/disabled
 
     const stuckTasks: string[] = [];