fix: handle workflow recording failures and track maxSteps

Author: nicoalbanese

apps/web/app/workflows/chat-post-finish-usage.test.ts

@@ -154,6 +154,43 @@ describe("recordWorkflowUsage", () => {
     });
   });
 
+  test("continues recording usage when workflow run persistence fails", async () => {
+    spies.recordWorkflowRun.mockImplementationOnce(() =>
+      Promise.reject(new Error("workflow runs table missing")),
+    );
+
+    const usage = makeUsage({
+      inputTokens: 100,
+      outputTokens: 50,
+      totalTokens: 150,
+    });
+
+    await recordWorkflowUsage(
+      "user-1",
+      "gpt-4",
+      usage,
+      makeAssistantMessage(),
+      undefined,
+      {
+        workflowRunId: "wrun-1",
+        chatId: "chat-1",
+        sessionId: "session-1",
+        status: "completed",
+        startedAt: "2026-01-01T00:00:00.000Z",
+        finishedAt: "2026-01-01T00:00:05.000Z",
+        totalDurationMs: 5000,
+        stepTimings: [],
+      },
+    );
+
+    expect(spies.recordWorkflowRun).toHaveBeenCalledTimes(1);
+    expect(spies.recordUsage).toHaveBeenCalledTimes(1);
+    expect((spies.recordUsage.mock.calls as unknown[][])[0][1]).toMatchObject({
+      agentType: "main",
+      model: "gpt-4",
+    });
+  });
+
   test("skips main recording when totalUsage is undefined", async () => {
     await recordWorkflowUsage(
       "user-1",

apps/web/app/workflows/chat-post-finish.ts

@@ -252,18 +252,22 @@ export async function recordWorkflowUsage(
       await import("@open-harness/agent");
 
     if (workflowRun) {
-      await recordWorkflowRun({
-        id: workflowRun.workflowRunId,
-        chatId: workflowRun.chatId,
-        sessionId: workflowRun.sessionId,
-        userId,
-        modelId,
-        status: workflowRun.status,
-        startedAt: workflowRun.startedAt,
-        finishedAt: workflowRun.finishedAt,
-        totalDurationMs: workflowRun.totalDurationMs,
-        stepTimings: workflowRun.stepTimings,
-      });
+      try {
+        await recordWorkflowRun({
+          id: workflowRun.workflowRunId,
+          chatId: workflowRun.chatId,
+          sessionId: workflowRun.sessionId,
+          userId,
+          modelId,
+          status: workflowRun.status,
+          startedAt: workflowRun.startedAt,
+          finishedAt: workflowRun.finishedAt,
+          totalDurationMs: workflowRun.totalDurationMs,
+          stepTimings: workflowRun.stepTimings,
+        });
+      } catch (error) {
+        console.error("[workflow] Failed to record workflow run:", error);
+      }
     }
 
     // Record main agent usage

apps/web/app/workflows/chat.test.ts

@@ -308,7 +308,7 @@ describe("runAgentWorkflow", () => {
     expect(rwCalls[0][1]).toBe("gpt-4");
   });
 
-  test("passes workflow timing data to recordWorkflowUsage", async () => {
+  test("marks workflow run as failed when maxSteps is exhausted", async () => {
     agentFinishReason = "tool-calls";
     agentRawFinishReason = "provider_tool_use";
 
@@ -331,7 +331,7 @@ describe("runAgentWorkflow", () => {
     };
 
     expect(workflowRun.workflowRunId).toBe("wrun_test-123");
-    expect(workflowRun.status).toBe("completed");
+    expect(workflowRun.status).toBe("failed");
     expect(workflowRun.totalDurationMs).toBeGreaterThanOrEqual(0);
     expect(workflowRun.stepTimings).toHaveLength(2);
     expect(workflowRun.stepTimings).toEqual([

apps/web/app/workflows/chat.ts

@@ -477,6 +477,7 @@ export async function runAgentWorkflow(options: Options) {
     latestMessage.role === "assistant" ? latestMessage : undefined;
   const stepTimings: WorkflowRunStepTiming[] = [];
   let wasAborted = false;
+  let exhaustedMaxSteps = false;
   let totalUsage: LanguageModelUsage | undefined;
   let finalFinishReason: FinishReason | undefined;
   let streamClosed = false;
@@ -526,12 +527,18 @@ export async function runAgentWorkflow(options: Options) {
           : result.stepUsage;
       }
 
-      if (
-        result.finishReason !== "tool-calls" ||
-        shouldPauseForToolInteraction(
+      const shouldContinue =
+        result.finishReason === "tool-calls" &&
+        !shouldPauseForToolInteraction(
           result.responseMessage?.parts ?? pendingAssistantResponse.parts,
-        )
-      ) {
+        );
+
+      if (!shouldContinue) {
+        break;
+      }
+
+      if (options.maxSteps !== undefined && step + 1 >= options.maxSteps) {
+        exhaustedMaxSteps = true;
         break;
       }
     }
@@ -699,7 +706,11 @@ export async function runAgentWorkflow(options: Options) {
       await refreshDiffCache(options.sessionId, sandboxState);
     }
 
-    workflowStatus = wasAborted ? "aborted" : "completed";
+    workflowStatus = wasAborted
+      ? "aborted"
+      : exhaustedMaxSteps
+        ? "failed"
+        : "completed";
   } catch (error) {
     workflowStatus = wasAborted ? "aborted" : "failed";
     caughtError = error;