[yaacovcorcos]

Problem

Two main browser tests fail on Linux CI and drop PRs from the merge queue. Neither failure is related to the PRs they drop.

1. MessagesTimeline "renders video attachments with the shared video player". The test points the video at https://environment.test/api/assets/demo.mp4, a host that doesn't exist. On CI the browser's load error can arrive during the frame the test waits before reading the markup. By then MediaVideoPlayer has already swapped in "Video preview unavailable", so the <video> assertion fails.
2. RightPanelTabs "uses a custom binding from a text field and closes on Escape". After Escape, focus stayed on <body> for the whole 5 s wait. This failed twice in a row on this PR's own CI and has dropped many queue entries recently. Even with no running animations, Base UI defers close completion through its shared animation-frame scheduler. That makes focus restoration depend on frame timing inside the CI iframe.

Change

Both are test-only changes, and product code is unchanged.

• Video: for this one test, a capture-phase error listener on document stops error events whose target is a <video>. The listener is removed in onTestFinished. This test covers rendering, and the next test still covers the unavailable state. It sits inside the existing Scient-owned SCIENT-FORK section, and upstream T3 doesn't have it.
• RightPanelTabs: the change is made in paired SCIENT-FORK markers inside the upstream-owned test.
  • It sets Base UI's own BASE_UI_ANIMATIONS_DISABLED global (declared in @base-ui/react/global.d.ts and read by useAnimationsFinished). The file's unstubAllGlobals resets it.
  • It establishes real window and input focus before opening the menu, and checks that focus moved into the menu before Escape.
  • It then asserts that the menu unmounted and focus is back on the trigger.
  • Blocking the trigger's focus() still makes the test fail.

Verification

• Video tests: 2 passed (Windows, headless Chromium).
• RightPanelTabs.keyboard.browser.test.tsx: 10 of 10 full-file runs passed (90 tests), with no React act() warnings.
• vp fmt and vp lint on both files: 0 errors.
• What's proven: that Base UI picks the trigger, and that it defers close completion through its frame scheduler even with no animations.
• What's not proven: the exact Linux mechanism, because the failure never reproduced locally. This PR's CI on Linux is the confirmation.

Documentation impact: None — test-only changes.

Model: Claude Opus 5.5 in Claude Code (Scient). The RightPanelTabs investigation and fix are by Codex gpt-6.1-sol.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.6 KiB	5.6 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	4.3 KiB	4.3 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	26.2 KiB	26.2 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.6 KiB	5.6 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	4.3 KiB	4.3 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	26.6 KiB	26.6 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 8eeb303 · PR result: e675953 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Full compact HTTP snapshot with live cursor.

• Codex decoded thread snapshot: 133.1 KiB
• Claude decoded thread snapshot: 133.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.