diff --git a/apps/desktop/e2e/settings-row-focus-ring.spec.ts b/apps/desktop/e2e/settings-row-focus-ring.spec.ts
new file mode 100644
index 0000000000..060f02f143
--- /dev/null
+++ b/apps/desktop/e2e/settings-row-focus-ring.spec.ts
@@ -0,0 +1,175 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+import { expect, test } from './fixtures';
+import type { Locator, Page } from '@playwright/test';
+
+/**
+ * A settings row rings its OWN tab stop and nothing else.
+ *
+ * Astryx's Item draws `outline: 2px solid accent` at `:has(:focus-visible)`
+ * unconditionally. That is right only for a row whose click target is the
+ * invisible `<button>` Item renders for `onClick`/`href`. Every other row holds
+ * a control that rings itself, so the row outline was a second ring around the
+ * whole label + description + control band.
+ *
+ * Opening the 默认模型 picker showed it at its worst: a Selector popup is a
+ * native `popover`, so the top layer moves where the popup PAINTS but not where
+ * it sits in the DOM. Focus lands on the popup's search input, `:has()` walks
+ * up to the row, and the row drew a full-width ring while the trigger drew
+ * none — reported as "点开下拉框时外面整行都会出现一个蓝色的框".
+ *
+ * `packages/ui/src/styles.css` narrows the ring to the row's own tab stop.
+ * Three things have to hold together, and a fix that over-reaches passes the
+ * first two while failing the third:
+ *
+ *   1. an open picker leaves its row unringed,
+ *   2. keyboard focus on a row control rings the CONTROL and not the row,
+ *   3. a clickable row still rings, because there the ring IS the indicator.
+ *
+ * Computed style rather than a screenshot: the ring is one declaration, and
+ * reading it back needs no tolerance for anti-aliasing or theme drift.
+ */
+
+async function openSettingsPage(page: Page, section: string) {
+  await page.getByRole('button', { name: '设置' }).click();
+  await page.getByRole('button', { name: section, exact: true }).click();
+}
+
+/** Tab until `target` owns focus, so the browser treats it as keyboard focus. */
+async function tabTo(page: Page, target: Locator, limit = 40) {
+  for (let index = 0; index < limit; index++) {
+    await page.keyboard.press('Tab');
+    if (await target.evaluate((element) => element === document.activeElement)) return;
+  }
+  throw new Error('Tab order never reached the target control');
+}
+
+/** The row around whatever currently has focus, and whether it draws a ring. */
+function focusedRowOutline() {
+  const active = document.activeElement as HTMLElement | null;
+  const row = active?.closest('.astryx-item') as HTMLElement | null;
+  if (!row) return null;
+  const style = getComputedStyle(row);
+  return { outlineStyle: style.outlineStyle, outlineWidth: style.outlineWidth };
+}
+
+/** The trigger's own focus affordance: Astryx fields state it in the border. */
+const fieldChrome = (element: SVGElement | HTMLElement) => {
+  const field = (element as HTMLElement).parentElement as HTMLElement;
+  const style = getComputedStyle(field);
+  return `${style.borderColor} | ${style.boxShadow}`;
+};
+
+test('an open picker does not ring the settings row that hosts it', async ({
+  window: page,
+}) => {
+  await openSettingsPage(page, '通用');
+  await expect(page.getByRole('textbox', { name: '助手语气偏好' })).toBeEnabled();
+  await page.getByRole('button', { name: '默认模型' }).click();
+
+  // Wait for the state under test to actually exist before reading it. Astryx
+  // hands the popup's search input focus inside a requestAnimationFrame
+  // (Selector.tsx `onOpen`), so a read that wins that race would find focus
+  // still on the mouse-clicked trigger — which is not `:focus-visible`, so
+  // NOTHING rings and the assertion below passes against the bug too. Waiting
+  // on the condition rather than on a timeout is what keeps this test honest.
+  await page.waitForFunction(() => {
+    const active = document.activeElement;
+    return (
+      document.querySelector('[popover]:popover-open') != null &&
+      active instanceof HTMLElement &&
+      active.closest('[popover]:popover-open') != null
+    );
+  });
+
+  const state = await page.evaluate(() => {
+    const active = document.activeElement as HTMLElement;
+    const row = active.closest('.astryx-item') as HTMLElement | null;
+    return {
+      // The mechanism, asserted so the ring check can never pass vacuously:
+      // the focused popup input is still a DOM descendant of the row, because
+      // the top layer moves where a popover PAINTS and not where it sits. If
+      // that ever stops being true this test goes quiet instead of green.
+      focusInsideRow: row != null,
+      outlineStyle: row ? getComputedStyle(row).outlineStyle : null,
+    };
+  });
+
+  expect(state.focusInsideRow).toBe(true);
+  expect(state.outlineStyle).toBe('none');
+});
+
+test('keyboard focus rings the row control, never the row', async ({ window: page }) => {
+  await openSettingsPage(page, '通用');
+  await expect(page.getByRole('textbox', { name: '助手语气偏好' })).toBeEnabled();
+
+  const trigger = page.getByRole('button', { name: '默认模型' });
+  const resting = await trigger.evaluate(fieldChrome);
+
+  await page.getByRole('textbox', { name: '助手语气偏好' }).focus();
+  await tabTo(page, trigger);
+
+  expect((await page.evaluate(focusedRowOutline))?.outlineStyle).toBe('none');
+  // …and the control still says it has focus. Astryx's field convention is an
+  // accent border, not an outline, which is why dropping the row's outline
+  // costs no keyboard visibility. Polled because border-color transitions.
+  await expect.poll(() => trigger.evaluate(fieldChrome)).not.toBe(resting);
+});
+
+test('a row whose own tab stop is the click target keeps its ring', async ({
+  window: page,
+}) => {
+  await openSettingsPage(page, '远程接入');
+
+  // A catalog row is a clickable Item, so Item renders an invisible <button>
+  // as a direct child and the row outline is that button's only indicator.
+  const row = page.locator('.settingsRemoteAccessCatalogRow').first();
+  await expect(row).toBeVisible();
+  const rowButton = row.locator(':scope > button').first();
+
+  await page.getByRole('button', { name: '远程接入', exact: true }).focus();
+  await tabTo(page, rowButton);
+
+  expect(await page.evaluate(focusedRowOutline)).toEqual({
+    outlineStyle: 'solid',
+    outlineWidth: '2px',
+  });
+});
+
+test('the row keeps its ring under forced colors, where the control loses its own', async ({
+  window: page,
+}) => {
+  await openSettingsPage(page, '通用');
+  await expect(page.getByRole('textbox', { name: '助手语气偏好' })).toBeEnabled();
+  await page.emulateMedia({ forcedColors: 'active' });
+
+  const trigger = page.getByRole('button', { name: '默认模型' });
+  const resting = await trigger.evaluate(fieldChrome);
+
+  await page.getByRole('textbox', { name: '助手语气偏好' }).focus();
+  await tabTo(page, trigger);
+
+  // Windows High Contrast drops box-shadow and repaints every border in one
+  // system color, so the field's own affordance reads the same focused as
+  // resting — measured here rather than assumed, because it is the entire
+  // reason the row must keep its ring.
+  expect(await trigger.evaluate(fieldChrome)).toBe(resting);
+  expect((await page.evaluate(focusedRowOutline))?.outlineStyle).toBe('solid');
+});
diff --git a/packages/ui/src/styles.css b/packages/ui/src/styles.css
index 8b5b7a3ea4..fe9719eeab 100644
--- a/packages/ui/src/styles.css
+++ b/packages/ui/src/styles.css
@@ -20,6 +20,62 @@
 /* Maka product compositions shared across desktop renderer surfaces. Astryx
    owns generic component chrome; these selectors own only product semantics. */
 
+/* An Item row draws a focus ring for ITS OWN tab stop, never for a control
+   that already draws one.
+
+   Astryx's Item (and everything built on it — ListItem, and so every settings
+   row) applies `outline: 2px solid accent` at `:has(:focus-visible)`,
+   unconditionally. That is right for an interactive row: `onClick`/`href`
+   makes Item render an INVISIBLE `<button>`/`<a>` as a direct child, and the
+   row's ring is that button's only visible focus indicator. It is wrong for
+   every other row, where the focusable thing is a real control in the end slot
+   that rings itself — the row ring is then a second outline around a whole
+   label + description + control band.
+
+   Two ways this showed up, both measured in a real window:
+
+   1. Tabbing onto any settings-row control — a Selector trigger, a Switch, a
+      SegmentedControl, a ghost Button — lit the control AND the full row.
+   2. Opening the 默认模型 picker outlined the entire row and nothing else.
+      A Selector popup is a native `popover` element: the top layer moves where
+      it PAINTS, not where it sits in the DOM, so it is still a descendant of
+      the row. Focus goes to the popup's search input on open, `:has()` walks
+      up to the row, and the row rings — while the trigger, whose own wrapper
+      is a sibling of the popup, does not. Reported as "点开下拉框时外面整行
+      出现一个蓝色的框".
+
+   `:not(:has(> :is(a, button):focus-visible))` is what separates the two: the
+   only `<a>`/`<button>` that is a DIRECT child of an Item root is that
+   invisible tab stop, since start/end/label content each sit inside their own
+   span. So an interactive row keeps its ring when its own tab stop has focus,
+   and loses it only when focus is on a control that owns its indicator — which
+   is also the case for a row carrying both.
+
+   Product CSS is the last cascade layer (`components`, see
+   cascade-layers.css), so this wins over the compiled StyleX atom regardless
+   of that atom's `:not(#\#)` specificity padding.
+
+   Gated on `forced-colors: none` because the premise — "the control already
+   draws one" — is only true when the page controls its own colors. Under
+   Windows High Contrast the control's affordance is gone: forced-colors drops
+   `box-shadow` entirely and repaints every border in the same system color, so
+   an Astryx field looks identical resting and focused, and its trigger button
+   sets `outline: none` on itself (Selector.tsx: the wrapper is supposed to
+   carry the ring). The row outline is then the ONLY indicator that survives —
+   outline is the one focus property forced-colors preserves — and taking it
+   away leaves a keyboard user with no visible focus at all. So there, the
+   doubled ring is the lesser problem and the row keeps it.
+
+   Fixing it at the control instead was the other option and is not better:
+   it would mean product CSS reaching past the design system into the internals
+   of Selector, Switch, SegmentedControl and Button, one forced-colors rule per
+   control, to restore what each was supposed to draw itself. */
+@media (forced-colors: none) {
+  .astryx-item:has(:focus-visible):not(:has(> :is(a, button):focus-visible)) {
+    outline: none;
+  }
+}
+
 .maka-quote-chip-remove { display: inline-flex; flex: 0 0 auto; align-items: center; justify-content: center; border: 0; color: var(--muted-foreground); background: transparent; transition: color var(--duration-quick) var(--ease-out-strong), background var(--duration-quick) var(--ease-out-strong); }
 .maka-quote-chip-remove:hover { color: var(--foreground); background: var(--state-hover-bg); }
 .maka-quote-chip-remove:focus-visible { color: var(--foreground); background: var(--state-hover-bg); outline: var(--focus-ring-width) solid var(--focus-ring); outline-offset: 1px; }