fix(agent-jobs): stop heartbeat from resurrecting terminated jobs

packages/backend/convex/agentJobs.ts

@@ -1223,6 +1223,7 @@ export const recoverStaleJobs = internalMutation({
         const now = Date.now();
         await ctx.db.patch(job._id, {
           status: 'timed_out',
+          workspaceStatus: 'expired',
           error:
             job.error ??
             'The worker stopped reporting progress; the job timed out.',
@@ -1512,12 +1513,15 @@ export const heartbeatWorkspace = mutation({
     if (job?.claimedBy !== args.workerId) {
       throw new ConvexError('Agent job not claimed by this worker.');
     }
+    if (isTerminalStatus(job.status)) {
+      return { success: true, cancelRequested: true };
+    }
     await ctx.db.patch(args.jobId, {
       workspaceStatus: 'active',
       lastHeartbeatAt: Date.now(),
       updatedAt: Date.now(),
     });
-    return { success: true, cancelRequested: job.status === 'cancelled' };
+    return { success: true, cancelRequested: false };
   },
 });
 

packages/backend/tests/unit/recover-stale-jobs.test.ts

@@ -119,6 +119,7 @@ describe('recoverStaleJobs', () => {
 
     expect(result).toEqual({ recovered: 1 });
     expect(state.job?.status).toBe('timed_out');
+    expect(state.job?.workspaceStatus).toBe('expired');
     expect(state.job?.completedAt).toBeTypeOf('number');
     expect(state.request?.status).toBe('failed');
     expect(state.thread?.status).toBe('failed');

packages/backend/tests/unit/update-status.test.ts

@@ -24,7 +24,7 @@ const spoonInput = {
 
 const seedJob = async (
   t: ReturnType<typeof convexTest>,
-  status: 'cancelled' | 'running',
+  status: 'cancelled' | 'running' | 'timed_out',
 ) =>
   await t.mutation(async (ctx) => {
     const now = Date.now();
@@ -109,7 +109,7 @@ describe('updateStatus terminal guard', () => {
 });
 
 describe('heartbeatWorkspace cancellation propagation', () => {
-  test('reports cancellation while refreshing the workspace heartbeat', async () => {
+  test('requests teardown for a cancelled job without resurrecting it', async () => {
     const t = convexTest(schema, modules);
     const jobId = await seedJob(t, 'cancelled');
 
@@ -122,8 +122,25 @@ describe('heartbeatWorkspace cancellation propagation', () => {
     expect(result).toEqual({ success: true, cancelRequested: true });
     const job = await t.run(async (ctx) => await ctx.db.get(jobId));
     expect(job?.status).toBe('cancelled');
-    expect(job?.workspaceStatus).toBe('active');
+    expect(job?.workspaceStatus).not.toBe('active');
-    expect(job?.lastHeartbeatAt).toBeTypeOf('number');
+    expect(job?.lastHeartbeatAt).toBeUndefined();
+  });
+
+  test('requests teardown for a timed_out job without resurrecting it', async () => {
+    const t = convexTest(schema, modules);
+    const jobId = await seedJob(t, 'timed_out');
+
+    const result = await t.mutation(api.agentJobs.heartbeatWorkspace, {
+      workerToken: 'test-worker-token',
+      workerId: 'worker-1',
+      jobId,
+    });
+
+    expect(result).toEqual({ success: true, cancelRequested: true });
+    const job = await t.run(async (ctx) => await ctx.db.get(jobId));
+    expect(job?.status).toBe('timed_out');
+    expect(job?.workspaceStatus).not.toBe('active');
+    expect(job?.lastHeartbeatAt).toBeUndefined();
   });
 
   test('reports no cancellation for an active job', async () => {