fix: update chat history tests for reversed order and improve plan mode tool filtering (#2506)

AGENTS.md

@@ -241,6 +241,18 @@ When running GitHub Actions with `pull_request_target` on cross-repo PRs (from f
 - Wrapping `ToggleGroupItem` in `TooltipTrigger` without `render` also breaks `:first-child`/`:last-child` CSS selectors for rounded corners on the group.
 - For drag handles and resize rails, prefer the native `title` attribute over `Tooltip` — tooltips appear immediately on hover and interfere with drag interactions, while `title` has a built-in delay.
 
+### Base UI Radio component selection in Playwright
+
+When testing Base UI Radio components in Playwright E2E tests, use `getByRole('radio', { name: 'Label' })` instead of `getByLabel('Label')`. The `getByLabel` selector may not work correctly with Base UI's Radio component structure.
+
+```ts
+// Correct: use role selector
+await page.getByRole("radio", { name: "React" }).click();
+
+// May not work with Base UI Radio
+await page.getByLabel("React").click();
+```
+
 ### Drizzle migration conflicts during rebase
 
 When rebasing a branch that has drizzle migrations conflicting with upstream (e.g., both have `0023_*.sql`):

e2e-tests/chat_history.spec.ts

@@ -23,32 +23,33 @@ test("should open, navigate, and select from history menu", async ({ po }) => {
   const historyMenu = po.page.locator('[data-mentions-menu="true"]');
   await expect(historyMenu).toBeVisible();
 
-  // Verify menu has items (oldest at top, newest at bottom)
+  // Verify menu has items (newest at top, oldest at bottom - array is reversed)
   const menuItems = po.page.locator('[data-mentions-menu="true"] li');
   await expect(menuItems).toHaveCount(2);
-  await expect(menuItems.nth(0)).toContainText("First test message");
-  await expect(menuItems.nth(1)).toContainText("Second test message");
+  await expect(menuItems.nth(0)).toContainText("Second test message");
+  await expect(menuItems.nth(1)).toContainText("First test message");
 
-  // Verify default selection is the last (most recent) item
+  // Verify default selection is the last visible item (oldest message, at bottom)
+  // After opening, a synthetic ArrowUp is dispatched which wraps to the bottom item
   const lastItem = menuItems.nth(1);
   await expect(lastItem).toHaveClass(/bg-accent/, { timeout: 500 });
 
-  // Navigate up to first item
+  // Navigate up to first item (newest message)
   await po.page.keyboard.press("ArrowUp");
   const firstItem = menuItems.nth(0);
   await expect(firstItem).toHaveClass(/bg-accent/);
-  await expect(firstItem).toContainText("First test message");
+  await expect(firstItem).toContainText("Second test message");
 
-  // Navigate up again to wrap to last item
+  // Navigate up again to wrap to last item (oldest message)
   await po.page.keyboard.press("ArrowUp");
   await expect(lastItem).toHaveClass(/bg-accent/);
 
-  // Select with Enter
+  // Select with Enter (selects oldest message)
   await po.page.keyboard.press("Enter");
 
   // Menu should close and text should be inserted
   await expect(historyMenu).not.toBeVisible({ timeout: Timeout.MEDIUM });
-  await expect(chatInput).toContainText("Second test message", {
+  await expect(chatInput).toContainText("First test message", {
     timeout: Timeout.MEDIUM,
   });
 
@@ -57,12 +58,12 @@ test("should open, navigate, and select from history menu", async ({ po }) => {
   await po.page.keyboard.press("ArrowUp");
   await expect(historyMenu).toBeVisible();
 
-  // Click the first item (oldest message)
+  // Click the first item (newest message, at top)
   await menuItems.nth(0).click();
 
-  // Verify menu closed and first message was inserted
+  // Verify menu closed and newest message was inserted
   await expect(historyMenu).not.toBeVisible({ timeout: Timeout.MEDIUM });
-  await expect(chatInput).toContainText("First test message", {
+  await expect(chatInput).toContainText("Second test message", {
     timeout: Timeout.MEDIUM,
   });
 });

e2e-tests/plan_mode.spec.ts

@@ -66,8 +66,8 @@ test("plan mode - questionnaire flow", async ({ po }) => {
     timeout: Timeout.MEDIUM,
   });
 
-  // Select "React" radio option
-  await po.page.getByLabel("React").click();
+  // Select "React" radio option (using role selector for base-ui Radio component)
+  await po.page.getByRole("radio", { name: "React" }).click();
 
   // Click Submit (single question → Submit button shown)
   await po.page.getByRole("button", { name: /Submit/ }).click();

src/pro/main/ipc/handlers/local_agent/tool_definitions.ts

@@ -338,6 +338,11 @@ export function buildAgentToolSet(
       continue;
     }
 
+    // Skip planning-specific tools when NOT in plan mode
+    if (!options.planModeOnly && PLANNING_SPECIFIC_TOOLS.has(tool.name)) {
+      continue;
+    }
+
     // In read-only mode, skip tools that modify state
     if (options.readOnly && tool.modifiesState) {
       continue;

testing/fake-llm-server/chatCompletionHandler.ts

@@ -33,45 +33,39 @@ export const createChatCompletionHandler =
       .slice()
       .reverse()
       .find((m: any) => m.role === "user");
+
+    // Extract text content from last user message (handles both string and array content)
+    let userTextContent = "";
     if (lastUserMessage) {
-      // Handle both string content and array content (AI SDK format)
-      let textContent = "";
       if (typeof lastUserMessage.content === "string") {
-        textContent = lastUserMessage.content;
+        userTextContent = lastUserMessage.content;
       } else if (Array.isArray(lastUserMessage.content)) {
         const textPart = lastUserMessage.content.find(
           (p: any) => p.type === "text",
         );
         if (textPart) {
-          textContent = textPart.text;
+          userTextContent = textPart.text;
         }
       }
 
-      const localAgentFixture = extractLocalAgentFixture(textContent);
+      const localAgentFixture = extractLocalAgentFixture(userTextContent);
       console.error(
-        `[local-agent] Checking message: "${textContent.slice(0, 50)}", fixture: ${localAgentFixture}`,
+        `[local-agent] Checking message: "${userTextContent.slice(0, 50)}", fixture: ${localAgentFixture}`,
       );
       if (localAgentFixture) {
         return handleLocalAgentFixture(req, res, localAgentFixture);
       }
 
       // Route plan acceptance message to exit-plan fixture
-      if (textContent.includes("I accept this plan")) {
+      if (userTextContent.includes("I accept this plan")) {
         return handleLocalAgentFixture(req, res, "exit-plan");
       }
     }
 
     let messageContent = CANNED_MESSAGE;
 
-    if (
-      lastMessage &&
-      Array.isArray(lastMessage.content) &&
-      lastMessage.content.some(
-        (part: { type: string; text: string }) =>
-          part.type === "text" &&
-          part.text.includes("[[UPLOAD_IMAGE_TO_CODEBASE]]"),
-      )
-    ) {
+    // Check for upload image to codebase using lastUserMessage (which already handles both string and array content)
+    if (userTextContent.includes("[[UPLOAD_IMAGE_TO_CODEBASE]]")) {
       messageContent = `Uploading image to codebase
 <dyad-write path="new/image/file.png" description="Uploaded image to codebase">
 DYAD_ATTACHMENT_0