[jlongster]

Fixes the app e2e specs that fail on dev since subagents always run in the background (#63382, #63391). Seven cases fail on both Linux and Windows e2e (…, app-2), e.g. run 43205730927. Test-only change.

Spec	Was	Now
timeline-geometry.spec.ts "scrolls the timeline behind the composer region…" (pointer, touch)	Used a blocking subagent to get the background hint row, which subagents no longer have	A blocking shell supplies the hint row. A busy child session supplies the 1 subagent pill. Same geometry checks.
timeline-stream.spec.ts "counts a backgrounded subagent…"	Expected the removed (background) subtitle and sent background: true	Checks the exact task description. The launch keeps status: "running" while the child's idle → busy → idle drives the count.
timeline-stream.spec.ts "separates blocking and already-backgrounded work…"	Used a blocking subagent as blocking work	A blocking shell, a background shell and a running child. Checks the hint, the 1 subagent · 1 shell menu, interrupt, shell output and kill.
timeline-stream.spec.ts "holds a revealed call … under its stuck group header" (3 variants)	Revealed a starting subagent's call from its chip, which #63391 removed	Reveal from a finished shell's notice, a live path (session-timeline-row.tsx → virtualizer.tsx revealPart). The standalone "jumps from a finished shell's notice to its call" case is folded into these, keeping its assertions.

The describe block is renamed "background work". No product code changes.

🚩 Flag: none (test).

📊 Proof (production build, 5 workers, fully parallel, no retries)

• Both full specs: 78/78 pass.
• The seven affected cases with --repeat-each=10: 70/70 pass.
• typecheck:tsc passes in app. typecheck:e2e has 1 fewer error than base and no new ones. Prettier and oxlint are clean.

[opencodejr]

This test-only change looks sound. On the base, the seven affected app e2e cases fail as described. With this pull request, all of them pass in a local Playwright run (76 of 78 in the two specs). The two failures are in static projection tests that this pull request doesn't touch: one also fails on the base here, and the other passed on the base, so it looks environment-dependent rather than caused by this change. typecheck:e2e has one fewer error than the base and no new ones in these specs.

Public API: unchanged. Only e2e specs are modified.

The new fixtures are simpler than before. Using shellCommands replaces the hand-sent shell.created event, and status("idle", …) replaces the as never transport cast. I have no concerns.