automaker/.automaker/memory.md

Agent Memory - Lessons Learned

This file documents issues encountered by previous agents and their solutions. Read this before starting work to avoid repeating mistakes.

Testing Issues

Issue: Mock project setup not navigating to board view

Problem: Setting currentProject in localStorage didn't automatically show the board view - app stayed on welcome view. Fix: The currentView state is not persisted in localStorage. Instead of trying to set it, have tests click on the recent project from the welcome view to trigger setCurrentProject() which handles the view transition properly.

// Don't do this:
await setupMockProject(page); // Sets localStorage
await page.goto("/");
await waitForElement(page, "board-view"); // ❌ Fails - still on welcome view

// Do this instead:
await setupMockProject(page);
await page.goto("/");
await waitForElement(page, "welcome-view");
const recentProject = page.locator('[data-testid="recent-project-test-project-1"]');
await recentProject.click(); // ✅ Triggers proper view transition
await waitForElement(page, "board-view");

Issue: View output button test IDs are conditional

Problem: Tests failed looking for view-output-inprogress-${featureId} when the actual button had view-output-${featureId}. Fix: The button test ID depends on whether the feature is actively running:

• view-output-${featureId} - shown when feature is in runningAutoTasks (actively running)
• view-output-inprogress-${featureId} - shown when status is "in_progress" but NOT actively running

After dragging a feature to in_progress, wait for the auto_mode_feature_start event to fire before looking for the button:

// Wait for feature to start running
const viewOutputButton = page.locator(
  `[data-testid="view-output-${featureId}"], [data-testid="view-output-inprogress-${featureId}"]`
).first();
await expect(viewOutputButton).toBeVisible({ timeout: 8000 });

Issue: Elements not appearing due to async event timing

Problem: Tests checked for UI elements before async events (like auto_mode_feature_start) had fired and updated the UI. Fix: Add appropriate timeouts when waiting for elements that depend on async events. The mock auto mode takes ~2.4 seconds to complete, so allow sufficient time:

// Mock auto mode timing: ~2.4s + 1.5s delay = ~4s total
await waitForAgentOutputModalHidden(page, { timeout: 10000 });

Issue: Slider interaction testing

Problem: Clicking on slider track didn't reliably set specific values. Fix: Use the slider's keyboard interaction or calculate the exact click position on the track. For max value, click on the rightmost edge of the track.

Issue: Port binding blocked in sandbox mode

Problem: Playwright tests couldn't bind to port in sandbox mode. Fix: Tests don't need sandbox disabled - the issue was TEST_REUSE_SERVER environment variable. Make sure to start the dev server separately or let Playwright's webServer config handle it.

Code Architecture

Issue: Understanding store state persistence

Problem: Not all store state is persisted to localStorage. Fix: Check the partialize function in app-store.ts to see which state is persisted:

partialize: (state) => ({
  projects: state.projects,
  currentProject: state.currentProject,
  theme: state.theme,
  sidebarOpen: state.sidebarOpen,
  apiKeys: state.apiKeys,
  chatSessions: state.chatSessions,
  chatHistoryOpen: state.chatHistoryOpen,
  maxConcurrency: state.maxConcurrency, // Added for concurrency feature
})

Note: currentView is NOT persisted - it's managed through actions.

Issue: Auto mode task lifecycle

Problem: Confusion about when features are considered "running" vs "in_progress". Fix: Understand the task lifecycle:

1. Feature dragged to "in_progress" column → status becomes "in_progress"
2. auto_mode_feature_start event fires → feature added to runningAutoTasks
3. Agent works on feature → periodic events sent
4. auto_mode_feature_complete event fires → feature removed from runningAutoTasks
5. If passes: true → status becomes "verified", if passes: false → stays "in_progress"

Best Practices Discovered

Testing utilities are critical

Create comprehensive testing utilities in tests/utils.ts to avoid repeating selector logic:

• waitForElement - waits for elements to appear
• waitForElementHidden - waits for elements to disappear
• setupMockProject - sets up mock localStorage state
• navigateToBoard - handles navigation from welcome to board view

Always add data-testid attributes

When implementing features, immediately add data-testid attributes to key UI elements. This makes tests more reliable and easier to write.

Test timeouts should be generous but not excessive

• Default timeout: 30s (set in playwright.config.ts)
• Element waits: 5-15s for critical elements
• Auto mode completion: 10s (accounts for ~4s mock duration)
• Don't increase timeouts past 10s for individual operations

Mock auto mode timing

The mock auto mode in electron.ts has predictable timing:

• Total duration: ~2.4 seconds (300+500+300+300+500+500ms)
• Plus 1.5s delay before auto-closing modals
• Total: ~4 seconds from start to completion