feat(codex): harden runtime targeting and dashboard editing

ui/tests/unit/components/compatible-cli/codex-top-level-controls-card.test.tsx

@@ -0,0 +1,37 @@
+import { describe, expect, it, vi } from 'vitest';
+import { render, screen, userEvent } from '@tests/setup/test-utils';
+import { CodexTopLevelControlsCard } from '@/components/compatible-cli/codex-top-level-controls-card';
+
+describe('CodexTopLevelControlsCard', () => {
+  it('submits only changed fields so untouched unsupported values are preserved upstream', async () => {
+    const onSave = vi.fn();
+
+    render(
+      <CodexTopLevelControlsCard
+        values={{
+          model: null,
+          modelReasoningEffort: null,
+          modelProvider: null,
+          approvalPolicy: null,
+          sandboxMode: null,
+          webSearch: null,
+          toolOutputTokenLimit: null,
+          personality: null,
+        }}
+        providerNames={[]}
+        onSave={onSave}
+      />
+    );
+
+    const saveButton = screen.getByRole('button', { name: 'Save top-level settings' });
+    expect(saveButton).toBeDisabled();
+
+    await userEvent.type(screen.getByPlaceholderText('gpt-5.4'), 'gpt-5.4-mini');
+    expect(saveButton).toBeEnabled();
+
+    await userEvent.click(saveButton);
+
+    expect(onSave).toHaveBeenCalledTimes(1);
+    expect(onSave).toHaveBeenCalledWith({ model: 'gpt-5.4-mini' });
+  });
+});