test: wire the three orphaned suites into CI

.github/workflows/e2e.yml

@@ -158,13 +158,18 @@ jobs:
         with:
           path: ~/.cache/ms-playwright
           key: playwright-${{ runner.os }}-${{ hashFiles('packages/e2e/package.json') }}
+          # On a key rotation (any packages/e2e dep bump) the fallback restores
+          # the previous browsers and `playwright install` downloads only the
+          # missing version instead of a full cold install.
+          restore-keys: |
+            playwright-${{ runner.os }}-
 
-      - name: Install Playwright browsers
+      - name: Install Playwright chromium
         if: steps.cache-playwright.outputs.cache-hit != 'true'
         working-directory: packages/e2e
         run: npx playwright install chromium --with-deps
 
-      - name: Install Playwright system deps
+      - name: Install Playwright chromium system deps
         if: steps.cache-playwright.outputs.cache-hit == 'true'
         working-directory: packages/e2e
         run: npx playwright install-deps chromium

.github/workflows/storage-integration.yml

@@ -135,4 +135,4 @@ jobs:
             echo "::group::$f"
             bun test "$f"
             echo "::endgroup::"
-          done < <(find apps/api -name '*.integration.test.ts' | sort)
+          done < <(find apps/api packages -name '*.integration.test.ts' -not -path '*/node_modules/*' | sort)

.github/workflows/test.yml

@@ -38,6 +38,7 @@ jobs:
     outputs:
       relevant: ${{ steps.filter.outputs.relevant }}
       sandbox: ${{ steps.filter.outputs.sandbox }}
+      web: ${{ steps.filter.outputs.web }}
     steps:
       - name: Detect relevant changes
         id: filter
@@ -53,10 +54,12 @@ jobs:
               '[release]:'*)
                 echo "relevant=false" >> "$GITHUB_OUTPUT"
                 echo "sandbox=false" >> "$GITHUB_OUTPUT"
+                echo "web=false" >> "$GITHUB_OUTPUT"
                 ;;
               *)
                 echo "relevant=true" >> "$GITHUB_OUTPUT"
                 echo "sandbox=true" >> "$GITHUB_OUTPUT"
+                echo "web=true" >> "$GITHUB_OUTPUT"
                 ;;
             esac
             exit 0
@@ -68,6 +71,7 @@ jobs:
             echo "Could not list PR files — running to be safe."
             echo "relevant=true" >> "$GITHUB_OUTPUT"
             echo "sandbox=true" >> "$GITHUB_OUTPUT"
+            echo "web=true" >> "$GITHUB_OUTPUT"
             exit 0
           fi
           echo "Changed files:"; echo "$FILES"
@@ -76,6 +80,7 @@ jobs:
           # change (the daemon bundle pulls in typegen, shared, and harness).
           relevant=false
           sandbox=false
+          web=false
           while IFS= read -r f; do
             [ -z "$f" ] && continue
             case "$f" in
@@ -85,9 +90,13 @@ jobs:
             case "$f" in
               packages/sandbox/*|packages/typegen/*|packages/shared/*|packages/harness/*) sandbox=true ;;
             esac
+            case "$f" in
+              apps/web/*|packages/ui/*) web=true ;;
+            esac
           done <<< "$FILES"
           echo "relevant=$relevant" >> "$GITHUB_OUTPUT"
           echo "sandbox=$sandbox" >> "$GITHUB_OUTPUT"
+          echo "web=$web" >> "$GITHUB_OUTPUT"
 
   format:
     needs: changes
@@ -233,3 +242,43 @@ jobs:
       - name: Tear down
         if: always()
         run: docker rm -f sandbox-smoke || true
+
+  # Playwright component tests for the sections-editor field widgets
+  # (apps/web/ct). Self-contained — real chromium, no server/DB — so it lives
+  # here rather than e2e.yml, gated to web changes only.
+  component-tests:
+    needs: changes
+    if: needs.changes.outputs.web == 'true'
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Setup Bun and install dependencies
+        uses: ./.github/actions/setup-bun
+
+      - name: Cache Playwright browsers
+        id: cache-playwright
+        uses: actions/cache@v4
+        with:
+          path: ~/.cache/ms-playwright
+          key: playwright-ct-${{ runner.os }}-${{ hashFiles('apps/web/package.json') }}
+          # On a key rotation (playwright version bump) the fallback restores
+          # the previous browsers and `playwright install` downloads only the
+          # missing version instead of everything.
+          restore-keys: |
+            playwright-ct-${{ runner.os }}-
+
+      - name: Install Playwright chromium
+        if: steps.cache-playwright.outputs.cache-hit != 'true'
+        working-directory: apps/web
+        run: npx playwright install chromium --with-deps
+
+      - name: Install Playwright chromium system deps
+        if: steps.cache-playwright.outputs.cache-hit == 'true'
+        working-directory: apps/web
+        run: npx playwright install-deps chromium
+
+      - name: Run component tests
+        run: bun run --cwd=apps/web test:ct

.gitignore

@@ -90,6 +90,8 @@ studio-demo/
 # Playwright
 apps/web/test-results/
 apps/web/playwright-report/
+# Playwright component-testing's vite build cache (test:ct)
+apps/web/ct/playwright/.cache/
 apps/web/screenshots/
 packages/e2e/test-results/
 packages/e2e/playwright-report/

apps/api/src/harnesses/decopilot/built-in-tools/user-ask.e2e.test.ts → apps/api/src/harnesses/decopilot/built-in-tools/user-ask.test.ts

@@ -1,14 +1,13 @@
 /**
- * user_ask Built-in Tool E2E Tests
+ * user_ask built-in tool registration/schema tests
  *
- * End-to-end tests that verify the complete integration of user_ask tool:
+ * Verifies the user_ask tool wiring in getBuiltInTools():
  * - Tool is registered in getBuiltInTools()
  * - Tool has correct metadata (description, schemas)
  * - Tool has no execute function (client-side only)
  * - Input schema validates correctly
  * - Output schema is defined correctly
  *
- * Run with: bun test user-ask.e2e.test.ts
  */
 
 import { describe, expect, test } from "bun:test";

apps/web/ct/tests/array-object-drilldown.ct.tsx

@@ -1,8 +1,19 @@
 import { expect, test } from "@playwright/experimental-ct-react";
+import type { Locator } from "@playwright/test";
 import { SchemaFormHarness } from "../harness/schema-form-harness";
 import { sectionWithProps, TEST_RESOLVE_TYPE } from "../harness/fixtures";
 import { readBreadcrumb, readFormValue } from "../harness/ct-utils";
 
+/**
+ * The sortable row wrapper is a role=button whose accessible NAME concatenates
+ * the label with its inline action buttons ("Alpha Hide item Open actions for
+ * Alpha"), so an exact-name query no longer matches. Match rows by the visible
+ * label text instead (the inline action buttons are icon-only, and the array
+ * header label lives outside any button).
+ */
+const itemRow = (component: Locator, label: string) =>
+  component.getByRole("button").filter({ hasText: label });
+
 /**
  * ArrayField with OBJECT items + breadcrumb drill-down, exercised through the
  * full SchemaFormHarness (raw JSON Schema -> resolveSchema -> SchemaForm).
@@ -100,12 +111,8 @@ test("item rows are labelled by titleBy with a count badge", async ({
 
   // Empty breadcrumb -> list view with both rows.
   await expect.poll(() => readBreadcrumb(component)).toEqual([]);
-  await expect(
-    component.getByRole("button", { name: "Alpha", exact: true }),
-  ).toBeVisible();
-  await expect(
-    component.getByRole("button", { name: "Beta", exact: true }),
-  ).toBeVisible();
+  await expect(itemRow(component, "Alpha")).toBeVisible();
+  await expect(itemRow(component, "Beta")).toBeVisible();
 
   // Count badge reflects the number of items.
   await expect(component.getByText("2", { exact: true })).toBeVisible();
@@ -123,7 +130,7 @@ test("clicking a row drills into that item and shows its Name value", async ({
     />,
   );
 
-  await component.getByRole("button", { name: "Alpha", exact: true }).click();
+  await itemRow(component, "Alpha").click();
 
   // Breadcrumb drilled into the Alpha row.
   await expect.poll(() => readBreadcrumb(component)).toContain("Alpha");
@@ -148,7 +155,7 @@ test("editing a drilled-in item updates the correct index", async ({
   );
 
   // Drill into the second row (index 1).
-  await component.getByRole("button", { name: "Beta", exact: true }).click();
+  await itemRow(component, "Beta").click();
   await expect.poll(() => readBreadcrumb(component)).toContain("Beta");
 
   const nameInput = component.getByLabel("Name");
@@ -187,7 +194,7 @@ test("editing a label field whose value equals the array label keeps the editor
     />,
   );
 
-  await component.getByRole("button", { name: "Banner", exact: true }).click();
+  await itemRow(component, "Banner").click();
   await expect
     .poll(() => readBreadcrumb(component))
     .toEqual(["Banner", "Banner"]);
@@ -225,7 +232,7 @@ test("editing a label to collide with an earlier sibling keeps the opened row",
     />,
   );
 
-  await component.getByRole("button", { name: "Beta", exact: true }).click();
+  await itemRow(component, "Beta").click();
   const nameInput = component.getByLabel("Name");
   await expect(nameInput).toHaveAttribute("id", "cards.1.name");
 
@@ -254,9 +261,11 @@ test("deleting a row removes the right item from the array", async ({
     />,
   );
 
-  // Open the actions menu for the Alpha row (button is in the DOM though hidden).
+  // Open the actions menu for the Alpha row (button is in the DOM though
+  // hidden). exact: true — the row wrapper is also a role=button whose
+  // accessible name CONTAINS "Open actions for Alpha".
   await component
-    .getByRole("button", { name: "Open actions for Alpha" })
+    .getByRole("button", { name: "Open actions for Alpha", exact: true })
     .click();
   // DropdownMenu content is portaled onto document.body -> query via page.
   await page.getByRole("menuitem", { name: "Delete" }).click();
@@ -292,7 +301,5 @@ test("fallback label 'Item 1' when itemSchema has no titleBy/title and item has
   );
 
   await expect.poll(() => readBreadcrumb(component)).toEqual([]);
-  await expect(
-    component.getByRole("button", { name: "Item 1", exact: true }),
-  ).toBeVisible();
+  await expect(itemRow(component, "Item 1")).toBeVisible();
 });

apps/web/ct/tests/string-field.ct.tsx

@@ -74,7 +74,9 @@ test("textarea format renders a <textarea> and round-trips typed value", async (
     .toEqual({ body: "multi\nline\ntext" });
 });
 
-test("rich-text format renders a <textarea> and round-trips typed value", async ({
+// rich-text / rich-text-inline now render a TipTap rich editor (toolbar +
+// contenteditable ProseMirror) instead of a <textarea>; the model value is HTML.
+test("rich-text format renders a rich editor and round-trips typed value as HTML", async ({
   mount,
 }) => {
   const meta = sectionWithProps({
@@ -84,17 +86,18 @@ test("rich-text format renders a <textarea> and round-trips typed value", async
     <SchemaFormHarness meta={meta} resolveType={TEST_RESOLVE_TYPE} />,
   );
 
-  const field = component.getByLabel("Body");
-  await expect(field).toBeVisible();
-  await expect(component.locator("textarea#body")).toBeVisible();
-  await field.fill("rich content");
+  await expect(component.getByText("Body")).toBeVisible();
+  await expect(component.getByRole("button", { name: "Bold" })).toBeVisible();
+  const editor = component.locator('[contenteditable="true"]');
+  await expect(editor).toBeVisible();
+  await editor.fill("rich content");
 
   await expect
     .poll(() => readFormValue(component))
-    .toEqual({ body: "rich content" });
+    .toEqual({ body: "<p>rich content</p>" });
 });
 
-test("rich-text-inline format renders a <textarea> and round-trips typed value", async ({
+test("rich-text-inline format renders a rich editor and round-trips typed value as HTML", async ({
   mount,
 }) => {
   const meta = sectionWithProps({
@@ -104,14 +107,14 @@ test("rich-text-inline format renders a <textarea> and round-trips typed value",
     <SchemaFormHarness meta={meta} resolveType={TEST_RESOLVE_TYPE} />,
   );
 
-  const field = component.getByLabel("Body");
-  await expect(field).toBeVisible();
-  await expect(component.locator("textarea#body")).toBeVisible();
-  await field.fill("inline content");
+  await expect(component.getByText("Body")).toBeVisible();
+  const editor = component.locator('[contenteditable="true"]');
+  await expect(editor).toBeVisible();
+  await editor.fill("inline content");
 
   await expect
     .poll(() => readFormValue(component))
-    .toEqual({ body: "inline content" });
+    .toEqual({ body: "<p>inline content</p>" });
 });
 
 test("markdown format renders a <textarea> and round-trips typed value", async ({

apps/web/ct/tests/widget-dispatch.ct.tsx

@@ -22,7 +22,7 @@ test("string schema → text input", async ({ mount }) => {
   await expect(input).toHaveAttribute("type", "text");
 });
 
-test("number schema → number input", async ({ mount }) => {
+test("number schema → decimal text input", async ({ mount }) => {
   const meta = sectionWithProps({
     count: { type: "number", title: "Count" },
   });
@@ -32,10 +32,13 @@ test("number schema → number input", async ({ mount }) => {
 
   const input = component.getByLabel("Count");
   await expect(input).toBeVisible();
-  await expect(input).toHaveAttribute("type", "number");
+  // NumberField keeps the raw typed string in a text input (so "0." survives
+  // mid-typing) and signals numeric intent via inputmode.
+  await expect(input).toHaveAttribute("type", "text");
+  await expect(input).toHaveAttribute("inputmode", "decimal");
 });
 
-test("integer schema → number input", async ({ mount }) => {
+test("integer schema → numeric text input", async ({ mount }) => {
   const meta = sectionWithProps({
     quantity: { type: "integer", title: "Quantity" },
   });
@@ -45,7 +48,9 @@ test("integer schema → number input", async ({ mount }) => {
 
   const input = component.getByLabel("Quantity");
   await expect(input).toBeVisible();
-  await expect(input).toHaveAttribute("type", "number");
+  // Same raw-string NumberField as `number`, but integer → inputmode=numeric.
+  await expect(input).toHaveAttribute("type", "text");
+  await expect(input).toHaveAttribute("inputmode", "numeric");
 });
 
 test("boolean schema → switch", async ({ mount }) => {

packages/typegen/src/cli.e2e.test.ts → packages/typegen/src/cli.test.ts

@@ -85,16 +85,17 @@ const TOOLS = [
 ];
 
 let httpServer: ReturnType<typeof Bun.serve>;
+let mcpServer: Server;
 let baseUrl: string;
 let tmp: string;
 
 // A fresh server per test: one stateful transport accepts a single session,
 // and each CLI invocation is its own client session.
 beforeEach(async () => {
-  const server = new Server(
+  const server = (mcpServer = new Server(
     { name: "fake-crm", version: "1.0.0" },
     { capabilities: { tools: {} } },
-  );
+  ));
   server.setRequestHandler(ListToolsRequestSchema, async () => ({
     tools: TOOLS,
   }));
@@ -132,7 +133,15 @@ beforeEach(async () => {
   baseUrl = `http://localhost:${httpServer.port}`;
 });
 
-afterEach(() => httpServer?.stop(true));
+afterEach(async () => {
+  // Close the MCP server (which closes its transport and any session
+  // streams) BEFORE stopping the HTTP server: a lingering session stream
+  // keeps the worker's event loop alive after the suite finishes — under
+  // `bun test --parallel` that hangs the whole run at exit (the coordinator
+  // waits on worker exit; this cost a 10-minute timeout on CI once).
+  await mcpServer?.close().catch(() => {});
+  httpServer?.stop(true);
+});
 
 beforeAll(async () => {
   tmp = await mkdtemp(join(tmpdir(), "typegen-e2e-"));