Merge upstream cmux main @ 4e9d779888 (2026-10-01)

[rajinsyed]

Merges 259 first-parent upstream cmux commits (upstream main 2026-09-29 → 2026-09-30, through 4e9d779888). The headline upstream change is shared terminal sizing: a terminal viewed from the Mac, an iPhone, or another Mac through Devices now has one grid, sized by default to the smallest viewer, with bounds drawn in the divider color and Connected Devices… on iOS. User-facing notes and the one open decision are in SUPERMUX-UPGRADES.md (2026-10-01 section).

Conflicts

25 files conflicted, resolved per SUPERMUX.md: upstream's code plus the re-applied SUPERMUX fences. Most were mechanical (a fork line next to a new upstream argument, dependency, enum case, or CI if:). The ones that needed judgment:

• Notification history encoding. Upstream added isAgentEvent to NotificationFeedHistoryRecord, whose fork fence carries explicit CodingKeys/init(from:). Without adding the key there, the flag would be silently dropped from persisted history; it is now encoded and decoded.
• Workspace close (--force). Upstream added a running-process confirmation to the socket, control and mobile close paths. It stays upstream's and runs before the fork's empty-home close; in TerminalController+MobileWorkspaceList.swift the fork fence is split in two around it.
• iOS native scroll. The fork's enqueueScrollMechanicsDelta keeps its structure and now applies upstream's gridDisplayScale, matching upstream's own delta path for grids scaled to fit.
• iOS Feed vs. Notifications. Upstream's new agent Feed hides the Notifications tab by default, which would hide the fork's project-aware notification rows. A new fence (#601) keeps the Notifications tab visible beside the Feed; recorded as an open decision.
• New upstream switch. TerminalCopyAction.swift (new upstream file) switches exhaustively over the built-in actions; a fence (#600) adds .newClaudeHarness.
• SidebarWidthPolicyTests.swift moved to Swift Testing upstream; the right-sidebar minimum-width fences are re-expressed as #expect. RemoteDaemonRPCClientTimeoutIsolationTests.swift: upstream rewrote the second test without a PTY attach, so only the first test keeps the fork's queue fence.
• Localizable.xcstrings and project.pbxproj: the clone's installed merge drivers are older copies and bailed, so both were merged manually (per-key union with upstream's current driver; union of added entries, then normalize-pbxproj.py + check-pbxproj.sh). All 172 fork-added and 254 upstream-added pbxproj ids are present.

Registry

• Updated rows for moved code: feat(ios): port the usage-limits tracker to the phone #27, Main full-suite CI is red #59, Remote Macs: remote simulators keep working when simctl is slow #70, Remote Macs: size choices stay on their own Mac; agent E2E isolation; CoreSimulator crash guard #71, Fix Mac Release build: stop calling DEBUG-only helpers from fork code #98, #178, #228, #259, #322, #352–354, #362, #398, #433, #434, #447, #463, #500, #504 (path moved under Debug/).
• Added #600 (TerminalCopyAction.swift) and #601 (MobileDisplaySettings.swift). Numbers 517–599 are left free for the in-flight remote-workspaces branch, which already registers rows in that range.
• No row retired. scripts/supermux-check-touchpoints.sh passes, and the fence inventory matches pre-merge except the intended changes above.

Follow-up commits

• fix(supermux): restore upstream catalog entries the previous merge left stale: the 2026-09-30 merge kept older wording for 18 upstream-owned strings (for example "Couldn't reach Stack" on sign-in errors) and re-added 21 keys upstream had deleted. Every non-supermux. key now equals upstream except the iOS: redial at once after the Mac's idle timeout instead of probing the dead session #84 exception.
• docs(supermux): upgrade notes for the 2026-10-01 upstream merge.

Testing

• Tagged Mac Debug build: CMUX_DEV_BACKEND_MODE=local CMUX_RELOAD_NO_GLOBAL_CLI_LINKS=1 CARGO_PROFILE_RELEASE_BUILD_OVERRIDE_STRIP=false ./scripts/reload.sh --tag sync-up succeeded (merge commit; not launched).
• iOS simulator build: xcodebuild -workspace ios/cmux.xcworkspace -scheme cmux-ios -configuration Debug -destination 'platform=iOS Simulator,name=iPhone 18 Pro' build succeeded.
• ./scripts/test-unit.sh build-for-testing (compile only, no tests run): the app target, including the restored catalog, compiles, and all 1,199 cmuxTests files were compiled with no errors in fork-touched files. The test target still fails on five files that are byte-identical to upstream main (CLIVMTransferTests, CloudWorkspaceLiveProjectionTests, LastSurfaceClosePreferenceTests, PaneResizeShortcutTests, WorkspaceCloseTabsContextMenuTests). That is upstream's own break at 4e9d779888, with open fixes upstream (test: restore cmuxTests compile on main manaflow-ai/cmux#16285, #16245, #16306); the next merge should pick it up.
• swift test: SupermuxKit 1057/1058 in the full run (the timing-sensitive closeRetainsEscalationUntilAnIgnoringProcessIsKilled failed while the Mac build loaded the machine; its suite passes 16/16 alone), SupermuxMobileCore 88/88, SupermuxMobileKit 225/225, SupermuxMobileUI 205/205.
• CI guard scripts that read the edited workflows pass: test_ci_guard_workflow_structure.py, test_ci_workflow_guards_are_wired.py, test_ci_reusable_workflow_permissions.py, test_ci_workflow_path_filter_parity.py. test_ci_change_areas.py fails the same way on a pristine upstream tree here (environmental).

Not verified: app-hosted suites (not run locally by policy), any live behavior in the tagged Mac app or on a phone. Worth a dogfood pass: iPhone scrollback on a shared grid larger than the phone (the fork's bounded scroll geometry still uses the unscaled cell height), the Feed and Notifications tabs together, and a Devices mirror resizing its source terminal.

Changelog

Changed: Supermux now includes upstream cmux through 2026-09-30, including shared terminal sizing across Mac, iPhone and Devices mirrors, Connected Devices on iOS, and the iOS Feed (shown beside Notifications)

[cubic-dev-ai]

10 issues found across 1976 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name=".github/workflows/ci-owned-pool-rescue.yml">

<violation number="1" location=".github/workflows/ci-owned-pool-rescue.yml:40">
P1: The periodic sweeper cannot discover these side-lane runs: it adopts marker artifacts and CI reruns, while these no-picker workflows upload no marker. Removing their `workflow_run` subscriptions leaves attempt-1 owned jobs without rescue; restore the subscriptions or add a marker the sweeper consumes.</violation>
</file>

<file name=".github/workflows/nightly.yml">

<violation number="1" location=".github/workflows/nightly.yml:1309">
P1: The `release` environment excludes `rc/**`, even though this workflow publishes every RC-branch push; RC runs will be blocked from signing and publishing. Add `rc/**` to the environment’s allowed deployment branches, or use an environment that permits RC refs.</violation>
</file>

<file name="Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+ExplicitTerminalInput.swift">

<violation number="1" location="Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+ExplicitTerminalInput.swift:56">
P1: The exactly-once route runs before this guard, so detached terminals can still receive and acknowledge input through `deliverExactlyOnce`. Check `terminalAllowsTraffic` before that call, as the paste paths do.</violation>
</file>

<file name="Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+TerminalScrollDelivery.swift">

<violation number="1" location="Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+TerminalScrollDelivery.swift:37">
P2: This check only gates calls before enqueueing; a scroll already pending when a detach event arrives is still dispatched after the in-flight request completes. Recheck attachment in `performTerminalScroll` before sending the RPC so queued TUI wheel input is not delivered after detach.</violation>
</file>

<file name="Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/WorktreeSeed/WorktreeSeedRepository.swift">

<violation number="1" location="Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/WorktreeSeed/WorktreeSeedRepository.swift:100">
P2: This marks `outer -> inner` as in-repository even when `inner` points to `/outside/secret`, because the final target component is never resolved. Resolve target symlink chains while retaining the dangling-target fallback.</violation>
</file>

<file name="Packages/macOS/CmuxCloudTui/Sources/CmuxCloudTui/CloudTuiFrontendProjectionRequests.swift">

<violation number="1" location="Packages/macOS/CmuxCloudTui/Sources/CmuxCloudTui/CloudTuiFrontendProjectionRequests.swift:21">
P2: When no projection exists, this omits the revision precondition, so concurrent first writes can both succeed and the later write can erase the earlier membership. Send revision `0` for `nil`; the caller can then retry the revision conflict against the latest projection.</violation>
</file>

<file name="CLI/CMUXCLI+AgentHibernation.swift">

<violation number="1" location="CLI/CMUXCLI+AgentHibernation.swift:21">
P2: This filter silently discards an incomplete `--workspace` option, so hibernate/wake proceeds without the requested workspace scope. Reject unparsed option tokens before sending the request.</violation>
</file>

<file name="Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/Workstream/WorkstreamItem.swift">

<violation number="1" location="Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/Workstream/WorkstreamItem.swift:65">
P2: The identity migration drops `reply` when it rebuilds an item with a canonicalized workstream ID, so remote terminal replies disappear from loaded history after restart. Preserve `item.reply` in `normalizedWorkstreamItem`.</violation>
</file>

<file name=".github/contributor/welcome.md">

<violation number="1" location=".github/contributor/welcome.md:9">
P2: This promises CI will run touched UI tests, but the UI-test lane is currently disabled unless `CI_UI_TESTS_ENABLED=1`, and it does not dispatch tests for fork PRs. Clarify that app-host suites are selected automatically while UI-test coverage depends on the lane being enabled and the PR being eligible.</violation>
</file>

<file name="Packages/Shared/CmuxAuthRuntime/Sources/CmuxAuthRuntime/Coordinator/AuthCoordinator+TeamSelection.swift">

<violation number="1" location="Packages/Shared/CmuxAuthRuntime/Sources/CmuxAuthRuntime/Coordinator/AuthCoordinator+TeamSelection.swift:92">
P2: This refresh can overwrite a team switch that completes while its fetch is in flight: the refresh applies its server selection without checking for a newer team mutation. Fence refresh results against team mutations or serialize refresh with selection so a stale response cannot restore the previous active scope.</violation>
</file>

<sub>Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.

Re-trigger cubic</sub>

[cubic-dev-ai]

P1: The periodic sweeper cannot discover these side-lane runs: it adopts marker artifacts and CI reruns, while these no-picker workflows upload no marker. Removing their workflow_run subscriptions leaves attempt-1 owned jobs without rescue; restore the subscriptions or add a marker the sweeper consumes.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At .github/workflows/ci-owned-pool-rescue.yml, line 40:

<comment>The periodic sweeper cannot discover these side-lane runs: it adopts marker artifacts and CI reruns, while these no-picker workflows upload no marker. Removing their `workflow_run` subscriptions leaves attempt-1 owned jobs without rescue; restore the subscriptions or add a marker the sweeper consumes.</comment>

<file context>
@@ -32,19 +32,14 @@ run-name: ${{ (github.event_name == 'schedule' || github.event_name == 'workflow
 #   vars.CI_NIGHTLY_TRUSTED_RUNNER a glaeda-runner-* label, attempt 1 of every
 #   push or schedule run on main asks for that one trusted mini, so every such
 #   run is watched (owned_pool_rescue.NIGHTLY_WORKFLOW_PATH).
+# The side lanes have no picker and are already covered by the periodic
+# sweeper. They stay in the script's target rules for explicit dispatches, but
+# no longer wake a fresh rescue through workflow_run.
</file context>

[cubic-dev-ai]

P1: The release environment excludes rc/**, even though this workflow publishes every RC-branch push; RC runs will be blocked from signing and publishing. Add rc/** to the environment’s allowed deployment branches, or use an environment that permits RC refs.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At .github/workflows/nightly.yml, line 1309:

<comment>The `release` environment excludes `rc/**`, even though this workflow publishes every RC-branch push; RC runs will be blocked from signing and publishing. Add `rc/**` to the environment’s allowed deployment branches, or use an environment that permits RC refs.</comment>

<file context>
@@ -1304,6 +1304,9 @@ jobs:
   build-sign-notarize-nightly:
+    # Production secrets: GitHub releases them only to protected refs (the
+    # environment's deployment branch policy: main, tags v*).
+    environment: release
     needs: [decide, build-nightly-ghostty-cli-helper, build-nightly-app, resolve-nightly-cmux-tui-client]
     if: needs.decide.outputs.should_build == 'true' && (github.event_name != 'schedule' || github.event.schedule == '47 8 * * *') && needs.decide.outputs.build_only != 'true'
</file context>

[cubic-dev-ai]

P1: The exactly-once route runs before this guard, so detached terminals can still receive and acknowledge input through deliverExactlyOnce. Check terminalAllowsTraffic before that call, as the paste paths do.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+ExplicitTerminalInput.swift, line 56:

<comment>The exactly-once route runs before this guard, so detached terminals can still receive and acknowledge input through `deliverExactlyOnce`. Check `terminalAllowsTraffic` before that call, as the paste paths do.</comment>

<file context>
@@ -53,6 +53,7 @@ extension MobileShellComposite {
         ) {
             return settlement == .delivered
         }
+        guard terminalAllowsTraffic(surfaceID: terminalID.rawValue) else { return false }
         let target = workspaceMutationTarget(for: workspaceID)
         guard let client = target.client else { return false }
</file context>

[cubic-dev-ai]

P2: This check only gates calls before enqueueing; a scroll already pending when a detach event arrives is still dispatched after the in-flight request completes. Recheck attachment in performTerminalScroll before sending the RPC so queued TUI wheel input is not delivered after detach.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/iOS/CmuxMobileShell/Sources/CmuxMobileShell/MobileShellComposite+TerminalScrollDelivery.swift, line 37:

<comment>This check only gates calls before enqueueing; a scroll already pending when a detach event arrives is still dispatched after the in-flight request completes. Recheck attachment in `performTerminalScroll` before sending the RPC so queued TUI wheel input is not delivered after detach.</comment>

<file context>
@@ -34,6 +34,7 @@ extension MobileShellComposite {
     /// stale scroll packets.
     // SUPERMUX:end ios-terminal-alt-scroll-budget
     public func scrollTerminal(surfaceID: String, lines: Double, col: Int, row: Int) async {
+        guard terminalAllowsTraffic(surfaceID: surfaceID) else { return }
         // Screen-anchored sessions own primary-screen scrolling: the gesture
         // already moved the local mirror's viewport over locally accumulated
</file context>

[cubic-dev-ai]

P2: This marks outer -> inner as in-repository even when inner points to /outside/secret, because the final target component is never resolved. Resolve target symlink chains while retaining the dangling-target fallback.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/WorktreeSeed/WorktreeSeedRepository.swift, line 100:

<comment>This marks `outer -> inner` as in-repository even when `inner` points to `/outside/secret`, because the final target component is never resolved. Resolve target symlink chains while retaining the dangling-target fallback.</comment>

<file context>
@@ -74,12 +74,30 @@ public struct WorktreeSeedRepository: Sendable {
+        // The parent is resolved, not the target: the target may not exist, and
+        // every real component above it does.
+        let parent = target.deletingLastPathComponent().resolvingSymlinksInPath().standardizedFileURL
+        let resolved = parent.appendingPathComponent(target.lastPathComponent).standardizedFileURL.path
         if resolved == resolvedRootPath { return true }
         return resolved.hasPrefix(resolvedRootPath.hasSuffix("/") ? resolvedRootPath : resolvedRootPath + "/")
</file context>

[cubic-dev-ai]

P2: When no projection exists, this omits the revision precondition, so concurrent first writes can both succeed and the later write can erase the earlier membership. Send revision 0 for nil; the caller can then retry the revision conflict against the latest projection.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/macOS/CmuxCloudTui/Sources/CmuxCloudTui/CloudTuiFrontendProjectionRequests.swift, line 21:

<comment>When no projection exists, this omits the revision precondition, so concurrent first writes can both succeed and the later write can erase the earlier membership. Send revision `0` for `nil`; the caller can then retry the revision conflict against the latest projection.</comment>

<file context>
@@ -0,0 +1,26 @@
+            "generation": generation,
+            "projection": projection,
+        ]
+        if let expectedProjectionRevision {
+            fields["expected_projection_revision"] = String(expectedProjectionRevision)
+        }
</file context>

Suggested change

	if let expectedProjectionRevision {
	fields["expected_projection_revision"] = String(expectedProjectionRevision)
	fields["expected_projection_revision"] = String(expectedProjectionRevision ?? 0)

[cubic-dev-ai]

P2: This filter silently discards an incomplete --workspace option, so hibernate/wake proceeds without the requested workspace scope. Reject unparsed option tokens before sending the request.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At CLI/CMUXCLI+AgentHibernation.swift, line 21:

<comment>This filter silently discards an incomplete `--workspace` option, so hibernate/wake proceeds without the requested workspace scope. Reject unparsed option tokens before sending the request.</comment>

<file context>
@@ -0,0 +1,56 @@
+    ) throws {
+        let (workspaceRaw, afterWorkspace) = parseOption(args, name: "--workspace")
+        let (surfaceOption, remaining) = parseOption(afterWorkspace, name: "--surface")
+        let positional = remaining.filter { !$0.hasPrefix("-") }
+        guard let surfaceRaw = surfaceOption ?? positional.first,
+              !surfaceRaw.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty,
</file context>

[cubic-dev-ai]

P2: The identity migration drops reply when it rebuilds an item with a canonicalized workstream ID, so remote terminal replies disappear from loaded history after restart. Preserve item.reply in normalizedWorkstreamItem.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/macOS/CMUXAgentLaunch/Sources/CMUXAgentLaunch/Workstream/WorkstreamItem.swift, line 65:

<comment>The identity migration drops `reply` when it rebuilds an item with a canonicalized workstream ID, so remote terminal replies disappear from loaded history after restart. Preserve `item.reply` in `normalizedWorkstreamItem`.</comment>

<file context>
@@ -47,6 +60,9 @@ public struct WorkstreamItem: Identifiable, Codable, Sendable, Equatable {
     public var context: WorkstreamContext?
+    /// The terminal response associated with this exact event, when one was
+    /// submitted from mobile or another remote surface.
+    public var reply: WorkstreamReply?
     /// PID of the agent process that emitted the event (hook's parent
     /// pid). When non-nil, pending items get expired automatically as
</file context>

[cubic-dev-ai]

P2: This promises CI will run touched UI tests, but the UI-test lane is currently disabled unless CI_UI_TESTS_ENABLED=1, and it does not dispatch tests for fork PRs. Clarify that app-host suites are selected automatically while UI-test coverage depends on the lane being enabled and the PR being eligible.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At .github/contributor/welcome.md, line 9:

<comment>This promises CI will run touched UI tests, but the UI-test lane is currently disabled unless `CI_UI_TESTS_ENABLED=1`, and it does not dispatch tests for fork PRs. Clarify that app-host suites are selected automatically while UI-test coverage depends on the lane being enabled and the PR being eligible.</comment>

<file context>
@@ -6,5 +6,6 @@ A few things that help:
 
 - [Start here](https://github.com/manaflow-ai/cmux/blob/main/docs/start-here.md) covers what reviewers look for, what CI runs for you, and what happens next.
 - If the CLA check asks, reply with the sentence it gives you.
+- You don't need to run the app-host or UI tests locally. CI runs the ones your diff touches once a maintainer approves the first workflow run.
 - The [verification ladder](https://github.com/manaflow-ai/cmux/blob/main/docs/contributor-verification.md) shows which checks fit your change. Say in the description which ones you ran.
 - If we end up fixing the same problem another way, we'll credit you with a `Co-authored-by` trailer and link the fix here.
</file context>

Suggested change

	- You don't need to run the app-host or UI tests locally. CI runs the ones your diff touches once a maintainer approves the first workflow run.
	- CI selects touched app-host test suites automatically. UI tests run only when the UI-test lane is enabled and the PR is eligible; follow the [verification ladder](https://github.com/manaflow-ai/cmux/blob/main/docs/contributor-verification.md) for UI changes.

[cubic-dev-ai]

P2: This refresh can overwrite a team switch that completes while its fetch is in flight: the refresh applies its server selection without checking for a newer team mutation. Fence refresh results against team mutations or serialize refresh with selection so a stale response cannot restore the previous active scope.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. At Packages/Shared/CmuxAuthRuntime/Sources/CmuxAuthRuntime/Coordinator/AuthCoordinator+TeamSelection.swift, line 92:

<comment>This refresh can overwrite a team switch that completes while its fetch is in flight: the refresh applies its server selection without checking for a newer team mutation. Fence refresh results against team mutations or serialize refresh with selection so a stale response cannot restore the previous active scope.</comment>

<file context>
@@ -82,4 +82,13 @@ public extension AuthCoordinator {
+    /// team and falls back like sign-in does otherwise.
+    func refreshTeams() async {
+        guard isAuthenticated else { return }
+        await refreshTeams(generation: sessionGeneration)
+    }
 }
</file context>

[github-actions]

Merge receipt for 454545940f, merged 2026-10-03 07:49:47 UTC

• Not verified at merge: ci-status (failure), backend (failure), browser-skill (failure), CI fast guards (failure), Fast static checks (failure), guards (19) (failure), guest-install (failure), ios-tests (failure), linux-preflight (failure), package-conventions-lint (failure), suite-coverage (failure), system-keychain (failure), and 7 more
• Verified: agent-session-web-resources, auth-refresh-tests, catalog-structure, CI timing, client, command-regressions, contract, cpp consumer, cpp package, detect-ios-changes, diagnostic-presentation, diff-sidecar-check, and 39 more
• Skipped by policy: apply, Claude wrapper regressions, deploy-development, Dogfood build #​${{ github.event.pull_request.number }}, full-suite-coverage, ios-simulator, ios-simulator-build, macos, macOS admission gate, mobile-core-package, Notify IndexNow, remote-daemon, and 5 more
• Full suite: runs on main after merge.

Labeled merged-unverified: if main breaks near this merge, look here first.