[modacker]

Change

The test:webui-browser gate on webui was red with 14 failures. None of them is a product regression. The red had three causes, all in the test fixtures rather than in the WebUI client, and this PR fixes all three.

1. Twelve visual baselines were generated in the wrong environment

The twelve failing toHaveScreenshot cases reported pixel differences of 0.01 to 0.03 of the image pixels, and every one of the fourteen baseline images kept its exact pixel dimensions — shell-and-rail-linux.png is 1434x900 in both the committed and the freshly rendered version. A layout change would have moved the dimensions; nothing moved.

Rendering the committed baseline next to a freshly rendered one shows what the pixels are. The committed images draw the Chinese interface text with real CJK glyphs (新建任务, 导入会话, 插件, 输入消息…（输入 / 唤起命令）). GitHub's ubuntu-latest runner image ships no CJK font, so Chromium draws the very same labels as tofu boxes. The baselines were produced on a Linux box that had a CJK font installed, not on the runner the gate uses.

The baselines are therefore regenerated on the Linux Chromium the gate runs on, following the procedure the repository already documents in packages/webui/docs/verification/browser-suite.md:

To deliberately regenerate only these baselines, use pnpm exec playwright test packages/webui/test/browser/visual-baselines.spec.mjs --update-snapshots on Linux CI Chromium, inspect every resulting image, then run the same command without --update-snapshots to compare.

Nothing else changes: the file names, the file count and the pixel dimensions are all identical, so the repository still keeps the fourteen representative screenshots it documents. The job that produced them was temporary and is not part of this PR.

2. Two rail-pin expectations were left behind by a fixture rename

de314e05 changed the shared browser fixture's two synthetic session titles to English — { id: "A", title: "Synthetic session A" } and { id: "B", title: "Synthetic session B" } — for cross-environment portability. Two expectations in rail-pin.spec.mjs still asserted the old Chinese titles:

Expected: "取消置顶：绿川椒 Demo 订货小程序"
Received: "取消置顶：Synthetic session A"

Expected: "置顶：灵动岛卡住问题"
Received: "置顶：Synthetic session B"

The rendered label is correct; only the expectation is stale, so the two expectations now read the title the fixture actually publishes. The glyph assertions next to them are untouched, because they check the drawing rather than the title. The two old Chinese titles still in settings-account-tab.spec.mjs are deliberately left alone: those cases inject their own session array and do not read the shared fixture.

3. The settings dialog baseline recorded a frame of an entrance animation

After the regeneration, thirteen of the fourteen images came back byte for byte identical across independent runs on different runner instances, and one did not: settings-dialog-linux.png. It moved by a few hundred pixels, the text ink ran one pixel wider, and the theme preview images were re-rasterised a fraction of a pixel apart.

The cause is that the settings overlay fades in over the app and the panel behind it slides in (webui-overlay-enter, webui-modal-enter, webui-settings-content-in in packages/webui/src/client/features/settings/ui/settings.css). The screenshot was taken while those were still running, so the baseline held a frame of the entrance — and the entrance advances on the clock, so the next run caught a different frame. The browser panel case in the same file already waits for its animations to finish before it screenshots; the settings dialog case did not, and now does, over the whole subtree.

Validation

• Checks run and results: the Source verification workflow on this branch, verify (ubuntu-latest, 24), node 24, pnpm verify --profile full — green, including the test:webui-browser gate that was red on webui at 97e33fd8.
• Reproducibility check: two regenerations on the Linux runner with the settle wait in place produce all fourteen images byte for byte identical. Before the wait, the same comparison across three independent runs (twice with the plain job shape, once after the full verify pipeline) returned thirteen identical images and a differing settings dialog, so the job shape and the environment are not the cause and the animation frame was.
• Local run: packages/webui/test/browser/rail-pin.spec.mjs — 11 passed, including the pin button offers the unpin glyph on a pinned row and the pin button offers the pin glyph on a row that is not pinned. packages/webui/test/browser/settings-account-tab.spec.mjs — 9 passed, on a file this PR does not touch.
• NOT RUN: a local run of the two screenshot specs. A macOS run cannot compare against -linux.png baselines, so it proves nothing here; the Linux runner is the only meaningful check, which is why the gate was verified there.

Two trade-offs worth a maintainer decision

The baselines are now full of tofu boxes. They record what ubuntu-latest renders, which is the Chinese interface text as empty squares. The baselines therefore no longer show — and so no longer prove — that the Chinese text renders. The alternative is to install a CJK font in the browser job before the gate runs and regenerate the baselines with it, which would give baselines that show the real text, at the cost of a permanent CI change and an apt-get step on every Linux run. This PR takes the minimal route and leaves that decision open.

The baselines sit on a shared runner image. If that image ever gains a CJK font — or drops one — the gate goes red again for the same reason as before. The repository's own documentation already treats the runner as the pinned reference, so this is the status quo rather than something this PR introduces.

Publication and contribution checks

• I have permission to contribute these changes under the existing licenses applicable to the changed files/packages; imported material and its provenance are identified and existing notices are preserved.
• No credentials, account data, real user content, internal source history or private review material is included.
• Added/removed source files were reviewed before regenerating release/public-source.json; new tests are declared in test/vitest-suites.json where applicable. — No source file is added or removed: the diff is 14 modified baseline PNGs and two modified specs, so both manifests are unchanged.
• Shared English/Chinese documentation and capability/verification records are updated where applicable. — The procedure this PR follows is already documented in packages/webui/docs/verification/browser-suite.md; no document describes the old baselines.

Maintainer handoff

Publication scope or license changes (if any): none.

Shared-source port: not needed.