test: repair session-restore, CLI, shell, and Cloud placement app-host tests

[teamleaderleo]

App-host repairs for the terminal/session/CLI/shell group, into fix/app-host-green (#13643). Failures are from #13403's CI run 35688876556.

Fixed

• CloudPlacementCoordinator aDetachedTerminalDropsItsStaleTabBeforeMoving: test bug. c402b9c (Repair main's full macOS suite: package test compile and app-host regressions #13403) made the test install a daemon graph. When the placement lane drains, it reconciles against that graph, which predates the projected tab, and clears tab_projected. The real projection reply always carries a mutation cursor (CmuxTuiSnapshotParser.placedTab requires one), and that cursor fences this reconcile. The fake provider returned no cursor. It now takes a projectCursor, and the test sets it ahead of the installed graph.
• TabManagerSessionSnapshotTests.testSessionSnapshotPersistsRemoteSurfaceProjectionsAndRestoreRelinksThem: stale test. After Fix Cloud display ownership and connection readiness #13196, validateOwnership refuses a workspace the app can't resolve. The test now registers the restored TabManager with LiveWorkspaceFixture.withAppRegistration, the same fix test: give Cloud catalog tests live destination workspaces #13651 uses.
• ZshShellIntegrationHandoffTests.testShellIntegrationRelayPromptRefreshUsesRefreshReasonInZsh: test bug. Since Add first-class Mosh transport for remote workspaces #8442, precmd also sends report_shell_state, from a separate background child. The test waited only until the log was non-empty, so it could see that line alone. It now waits, with a deadline, for the ports_kick line itself. The bash sibling test gets the same fix.
• CLINotifyProcessIntegrationRegressionTests.testDefaultFreestyleSSHAttachRelaysAfterDelayedSuccessfulCredential: product bug. For the first 2 s after sending the password, the expect wrapper watches for a rejection with log_user 0. A session that authenticates and exits inside that window hits the eof branch, and its output was dropped. The CLI now flushes expect_out(buffer) before it exits. Fix CLI notify integration regression tests #13207 replaced the test's sleep 9 with a FIFO release, which is what exposed the bug.
• CLINotifyProcessIntegrationRegressionTests.testVMDevRejectsBadInputBeforeAnyRequest: test regression. Fix CLI notify integration regression tests #13207 dropped 4120e35's expectsSocketConnection: false path. Each rejected run then waited 60 s on a mock expectation that only the listener closing fulfills, and the listener closes after the wait. That path is restored: the listener stays open and its backlog is checked with a non-blocking accept after the CLI exits.
• Cloud SCP with OpenSSH transferUsesRealSFTPAndChecksTheHostKey: test bug, not a host dependency. The test starts its own /usr/sbin/sshd, and every other scenario passed on the runner. The terminal case (Fix idle Cloud SCP watches and report local transfer failures #12759) read the pty master only after the CLI exited. A pty holds about 1 KiB, the host-key failure report is longer, and the CLI blocked on stderr until the 30 s timeout. A thread now drains the master while the CLI runs.

Already fixed by #13651: "Restoring a Cloud terminal reserves a manual mirror instead of a local shell".

Not failing in the latest run: CloudManualMirrorPresentation, CloudManualMirrorTransport, CloudNativeLayoutProjection, CmuxTuiSurfaceProviderRegistryPolling.

Left open: these need an app-host run to diagnose, and there's no evidence yet for a fix.

• offPlanGeometryWithUnchangedSizingInputsReconverges: fails every run (%1 plan=320x450 view=511x450, rearms=3). All three recovery passes re-impose the extent, but the bonsplit divider stays at the perturbed 0.8. The mirror code hasn't changed since July. The last bonsplit bump was Prevent terminal flicker and delayed nested layout on workspace switches #12310 (faf8418, new resizeSubviewsWithOldSize).
• workspaceRevealKeepsTerminalSizeUntilAnUnchangedGeometryPass: flaky. It passed in 35666501009 and failed in 35688876556: the final resize to 360 never published a smaller native size within 5 s.
• WorkspaceTerminalFocusRecoveryTests.testTerminalFirstResponderFeedbackPreservesActiveFocusTransaction: fails every run. The surface never becomes first responder. Other focus suites show the same symptom: WorkspaceTerminalFocusRecoverySwiftTests and TerminalNotificationDirectInteractionTests.

swiftc -parse passes on every changed Swift file. check-test-determinism.py --strict and lint-pbxproj-test-wiring.sh are clean. The expect eof flush was reproduced and verified locally with /usr/bin/expect, and the pty drain with a 20 KB writer.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

---

Summary by cubic

Repairs app-host test failures across session-restore, CLI, shell, and Cloud placement, and fixes a CLI bug where an SSH session that authenticated and exited right away dropped its output.

Bug Fixes

• The SSH expect wrapper now flushes expect_out(buffer) on eof before exiting, so a session that authenticates and ends inside the two-second rejection watch still prints its output.

Test Fixes

• The Cloud placement test sets a mutation cursor on the fake projection reply so the reconcile can't clear the projected tab.
• The session-restore test registers the restored window with the app before relinking projections.
• Shell relay tests wait for the surface.ports_kick line itself instead of any relay line.
• The invalid vm dev input test keeps the listener open and checks its backlog after the CLI exits, instead of waiting on a mock expectation that only closes after the timeout.
• The SCP pty test drains the terminal master while the CLI runs, preventing stderr from blocking past the timeout.

^{Written for commit bafd6c3. Summary will update on new commits.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository: manaflow-ai/cmux/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 4560d4ba-0efa-422d-b097-05d544cfc533

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}