fix: hide start agent button during preparation and fix auto-start race condition on step move (#516)

2026-04-01 23:00:05 +01:00
parent 348ee25e6f
commit 6fea6277a8
4 changed files with 246 additions and 0 deletions
@@ -548,6 +548,19 @@ func (s *Service) autoStartStepPrompt(
 	// Record a user message so the auto-start prompt is visible in chat history.
 	s.recordAutoStartMessage(ctx, taskID, sessionID, prompt, planMode)

+	// If the session is in CREATED state, the agent was never started (e.g. workspace-only
+	// preparation from a blocked auto-start). PromptTask will reject CREATED sessions,
+	// so use StartCreatedSession which properly launches the agent on the prepared workspace.
+	// Pass skipMessageRecord=true since recordAutoStartMessage above already recorded it.
+	if session.State == models.TaskSessionStateCreated {
+		s.logger.Info("auto-start: session is CREATED, launching agent via StartCreatedSession",
+			zap.String("task_id", taskID),
+			zap.String("session_id", sessionID),
+			zap.String("step_name", stepName))
+		_, err := s.StartCreatedSession(ctx, taskID, sessionID, session.AgentProfileID, prompt, true, planMode, nil)
+		return err
+	}
+
 	const maxRetryAttempts = 5
 	for attempt := 1; attempt <= maxRetryAttempts; attempt++ {
 		_, err := s.PromptTask(ctx, taskID, sessionID, prompt, "", planMode, nil)
@@ -7,6 +7,7 @@ import type { Message, TaskSessionState, TaskState } from "@/lib/types/http";
 import type { ToolCallMetadata } from "@/components/task/chat/types";
 import { launchSession } from "@/lib/services/session-launch-service";
 import { buildStartCreatedRequest } from "@/lib/services/session-launch-helpers";
+import { useAppStore } from "@/components/state-provider";
 import { ChatMessage } from "@/components/task/chat/messages/chat-message";
 import { PermissionRequestMessage } from "@/components/task/chat/messages/permission-request-message";
 import { StatusMessage } from "@/components/task/chat/messages/status-message";
@@ -39,6 +40,10 @@ type AdapterContext = {

 function TaskDescriptionStartButton({ taskId, sessionId }: { taskId: string; sessionId: string }) {
  const [isStarting, setIsStarting] = useState(false);
+  const prepareStatus = useAppStore(
+    (state) => state.prepareProgress.bySessionId[sessionId]?.status ?? null,
+  );
+
  const handleStart = useCallback(async () => {
    setIsStarting(true);
    try {
@@ -51,6 +56,9 @@ function TaskDescriptionStartButton({ taskId, sessionId }: { taskId: string; ses
    }
  }, [taskId, sessionId]);

+  // Hide while environment is being prepared
+  if (prepareStatus === "preparing") return null;
+
  return (
    <div className="flex justify-end mt-1.5">
      <Button
@@ -59,6 +67,7 @@ function TaskDescriptionStartButton({ taskId, sessionId }: { taskId: string; ses
        className="cursor-pointer gap-1.5"
        onClick={handleStart}
        disabled={isStarting}
+        data-testid="task-description-start-button"
      >
        <IconPlayerPlay className="h-3.5 w-3.5" />
        {isStarting ? "Starting…" : "Start agent"}
@@ -0,0 +1,124 @@
+import { test, expect } from "../../fixtures/test-base";
+import { KanbanPage } from "../../pages/kanban-page";
+import { SessionPage } from "../../pages/session-page";
+
+test.describe("PR watcher auto-start on move", () => {
+  /**
+   * Race condition: When a task created from a PR watcher sits on a step
+   * WITHOUT auto_start_agent, the user opens the task (creating a CREATED
+   * session with workspace-only preparation), then moves the task to a step
+   * WITH auto_start_agent on the kanban board.
+   *
+   * BUG: The on_enter auto_start_agent action calls PromptTask which rejects
+   * CREATED sessions, causing the prompt to be queued instead of starting
+   * the agent. The queued prompt is never consumed because no agent is running.
+   *
+   * EXPECTED: The agent starts immediately on the prepared workspace.
+   *
+   * Flow:
+   *   1. Create workflow: Waiting (no auto_start) → Review (auto_start) → Done
+   *   2. Create task as PR watcher would (metadata.agent_profile_id)
+   *   3. Open task → session auto-prepares (CREATED, workspace-only)
+   *   4. Go back to kanban
+   *   5. Move task from Waiting to Review via API
+   *   6. Open the task
+   *   7. Assert: agent starts and completes (NOT queued)
+   */
+  test("starts agent immediately when moving task with CREATED session to auto-start step", async ({
+    testPage,
+    apiClient,
+    seedData,
+  }) => {
+    test.setTimeout(120_000);
+
+    // --- Create workflow: Waiting → Review (auto_start) → Done ---
+    const workflow = await apiClient.createWorkflow(seedData.workspaceId, "PR Move Workflow");
+
+    const waitingStep = await apiClient.createWorkflowStep(workflow.id, "Waiting", 0);
+    const reviewStep = await apiClient.createWorkflowStep(workflow.id, "Review", 1);
+    const doneStep = await apiClient.createWorkflowStep(workflow.id, "Done", 2);
+
+    // Configure Review step: auto_start + move to Done on turn complete
+    await apiClient.updateWorkflowStep(reviewStep.id, {
+      prompt: '/e2e:message("review complete")\n{{task_prompt}}',
+      events: {
+        on_enter: [{ type: "auto_start_agent" }],
+        on_turn_complete: [{ type: "move_to_step", config: { step_id: doneStep.id } }],
+      },
+    });
+
+    await apiClient.saveUserSettings({
+      workspace_id: seedData.workspaceId,
+      workflow_filter_id: workflow.id,
+      enable_preview_on_click: false,
+    });
+
+    // --- Create task as PR watcher would ---
+    const task = await apiClient.createTask(seedData.workspaceId, "PR #55: Refactor auth", {
+      description: "Review the auth refactoring changes",
+      workflow_id: workflow.id,
+      workflow_step_id: waitingStep.id,
+      repositories: [
+        {
+          repository_id: seedData.repositoryId,
+          base_branch: "main",
+        },
+      ],
+      metadata: {
+        agent_profile_id: seedData.agentProfileId,
+        pr_number: 55,
+        pr_branch: "refactor/auth",
+        pr_repo: "testorg/testrepo",
+        pr_author: "contributor",
+      },
+    });
+
+    // --- Navigate to kanban and verify task in Waiting ---
+    const kanban = new KanbanPage(testPage);
+    await kanban.goto();
+
+    const card = kanban.taskCardInColumn("PR #55: Refactor auth", waitingStep.id);
+    await expect(card).toBeVisible({ timeout: 15_000 });
+
+    // --- Open task → creates CREATED session (workspace-only preparation) ---
+    await card.click();
+    await expect(testPage).toHaveURL(/\/t\//, { timeout: 15_000 });
+
+    const session = new SessionPage(testPage);
+    await session.waitForLoad();
+
+    // Wait for the session to be created (auto-prepare fires from useAutoStartSession)
+    await expect
+      .poll(
+        async () => {
+          const { sessions } = await apiClient.listTaskSessions(task.id);
+          return sessions.length > 0 && sessions[0]?.state === "CREATED";
+        },
+        { timeout: 30_000, message: "Waiting for session to be created in CREATED state" },
+      )
+      .toBe(true);
+
+    // --- Go back to kanban ---
+    await kanban.goto();
+    await expect(kanban.taskCardByTitle("PR #55: Refactor auth")).toBeVisible({ timeout: 15_000 });
+
+    // --- Move task from Waiting to Review (triggers on_enter: auto_start_agent) ---
+    await apiClient.moveTask(task.id, workflow.id, reviewStep.id);
+
+    // --- Task should move through Review to Done after agent completes ---
+    // If the bug is present, the prompt would be queued and the agent never starts,
+    // so the task would stay in Review indefinitely.
+    const cardInDone = kanban.taskCardInColumn("PR #55: Refactor auth", doneStep.id);
+    await expect(cardInDone).toBeVisible({ timeout: 60_000 });
+
+    // --- Navigate to session and verify agent completed (not stuck with queued prompt) ---
+    await cardInDone.click();
+    await expect(testPage).toHaveURL(/\/t\//, { timeout: 15_000 });
+
+    const sessionAfter = new SessionPage(testPage);
+    await sessionAfter.waitForLoad();
+
+    // The agent should have completed — idle input visible
+    await expect(sessionAfter.idleInput()).toBeVisible({ timeout: 30_000 });
+  });
+});
@@ -0,0 +1,100 @@
+import { test, expect } from "../../fixtures/test-base";
+import { KanbanPage } from "../../pages/kanban-page";
+import { SessionPage } from "../../pages/session-page";
+
+test.describe("PR watcher start button timing", () => {
+  /**
+   * When a task is created from a PR watcher on a workflow step WITHOUT
+   * auto_start_agent, opening the task triggers environment preparation
+   * (workspace-only, no agent). The "Start agent" button should appear
+   * after preparation completes and clicking it should start the agent.
+   *
+   * Flow:
+   *   1. Create workflow step without auto_start_agent
+   *   2. Create a task as the PR watcher would (with metadata.agent_profile_id)
+   *   3. Navigate to task — auto-start fires, backend downgrades to prepare
+   *   4. Wait for CREATED session to exist (env preparing)
+   *   5. Reload page so SSR picks up the session and renders it
+   *   6. Assert: "Start agent" button becomes visible
+   *   7. Click the button — agent starts and completes
+   */
+  test("start button appears after environment preparation and starts agent", async ({
+    testPage,
+    apiClient,
+    seedData,
+  }) => {
+    test.setTimeout(120_000);
+
+    // --- Create a workflow step WITHOUT auto_start_agent ---
+    const reviewStep = await apiClient.createWorkflowStep(seedData.workflowId, "Review", 0);
+
+    await apiClient.saveUserSettings({
+      workspace_id: seedData.workspaceId,
+      workflow_filter_id: seedData.workflowId,
+      enable_preview_on_click: false,
+    });
+
+    // --- Create a task as the PR watcher would ---
+    const task = await apiClient.createTask(seedData.workspaceId, "PR #42: Add feature", {
+      description: "/e2e:simple-message",
+      workflow_id: seedData.workflowId,
+      workflow_step_id: reviewStep.id,
+      repositories: [
+        {
+          repository_id: seedData.repositoryId,
+          base_branch: "main",
+        },
+      ],
+      metadata: {
+        agent_profile_id: seedData.agentProfileId,
+        pr_number: 42,
+        pr_branch: "feature/add-feature",
+        pr_repo: "testorg/testrepo",
+        pr_author: "contributor",
+      },
+    });
+
+    // --- Navigate to kanban and click the task ---
+    const kanban = new KanbanPage(testPage);
+    await kanban.goto();
+
+    const card = kanban.taskCardByTitle("PR #42: Add feature");
+    await expect(card).toBeVisible({ timeout: 15_000 });
+    await card.click();
+
+    // Wait for navigation to session view
+    await expect(testPage).toHaveURL(/\/t\//, { timeout: 15_000 });
+
+    // --- Wait for auto-start to create a CREATED session ---
+    // useAutoStartSession fires with autoStart=true, backend downgrades to
+    // prepare (step lacks auto_start_agent). Session stays in CREATED state.
+    await expect
+      .poll(
+        async () => {
+          const { sessions } = await apiClient.listTaskSessions(task.id);
+          return sessions.length > 0;
+        },
+        { timeout: 30_000, message: "Waiting for session to be created" },
+      )
+      .toBe(true);
+
+    // --- Reload so SSR picks up the session and renders the chat ---
+    // After auto-start creates the CREATED session, the frontend doesn't
+    // automatically set it as the active session. Reloading ensures SSR
+    // resolves the session and the chat panel renders with the task description.
+    await testPage.reload();
+
+    const session = new SessionPage(testPage);
+    await session.waitForLoad();
+
+    // --- Assert: "Start agent" button becomes visible after prep completes ---
+    const startButton = session.chat.getByTestId("task-description-start-button");
+    await expect(startButton).toBeVisible({ timeout: 60_000 });
+
+    // --- Click the button to start the agent ---
+    await startButton.click();
+
+    // Agent should start and eventually complete (mock agent with simple message)
+    await expect(session.idleInput()).toBeVisible({ timeout: 30_000 });
+  });
+});