cmux-next: tools opened from the docked agent chat open in the strip

[teamleaderleo]

Summary

Section D of the 2026-10-06 call: the agent chat is a docked region on the left, and tools are tabs in the area to its right, like the Codex app. The chat dock is the left dock column with the agent_chat role (#18169), and it never gets tabs of its own.

What a person sees:

• From the docked chat: New Terminal (Ctrl-, menu, palette), New Browser and Cmd-T open a tab in the scrolling strip, never a split. The tab goes to the strip pane focused last and takes focus. A terminal gets only an explicit cwd and otherwise sends none, so the daemon's resolver picks it (decision NEW-TERMINAL-INHERITS-CWD, capability new-terminal-cwd-v1`). Cmd-T on a chat would make another chat, so it opens the New Tab page in the strip instead.
• A chat alone on its screen (one pane, one chat tab): New Terminal or New Browser moves the chat into a new left chat dock, and the new tab stays in the chat's old pane, focused. The move is one daemon commit (move-tab-to-column with a dock pin and respawn).
• While the screen is zoomed: nothing docks or moves; the tab opens in the zoomed pane.
• New Browser without a URL from the docked chat opens what the chat works on (cmux-next tabs: switch between terminal, browser and agent and keep the context #16620), in the strip.
• Chats outside the chat dock: a chat a person put in a plain dock, or that sits beside other columns, is a normal pane, so new tabs open in it as before. Only the role makes the chat dock.
• Unchanged: splits are still available when asked for. Scripts (CLI, MCP) keep opening tabs in the pane they name. Daemons without dock-column-role-v1 keep the old behavior.

Builds on #18169 (the role). Supersedes the inference-based next-chat-column-tools branch, which never got a PR.

Follow-ups: the dock refuses splits and drops, and layout operations skip it. A layout.dockAgentChat setting and Dock/Undock Chat actions come after that.

Testing

ChatColumnPlacementTests was committed red before the fix (b85c214), with a stub that always answers "here". Runs use scripts/measure/suite-runner.sh on a fleet runner.

Commit	Result
b85c214 (red)	the stub was missing import CmuxNextBridge, so it did not compile. Re-run with only that import added (scratch dc19226): 5 failed (theDockedChatSendsTheToolToTheStrip, theStripPaneFocusedLastGetsTheTab, theDockedChatSkipsOtherDocks, theDockedChatWithoutAStripGetsANewColumn, aChatAloneOnItsScreenDocks). The 4 tests that expect "here" passed.
90518c9 (green)	passed: ChatColumnPlacementTests (9 tests), NewTabKindTests, EmptyWorkspaceNewTabTests, StripNewTabTests, NewTerminalWorkspaceSettingsTests, AgentBrowserProfileTests, DockColumnCapabilityTests, DockColumnMappingTests, EncodingTests

5353e55 applies NEW-TERMINAL-INHERITS-CWD to the strip tab and to the terminal a lone chat leaves behind. No test covers that wiring yet; the docked-column agent on Lawrence's side is handing over tests for open-pane routing.

The first live check (build of 5353e55 on a capture mini, real key presses) found a bug. New Terminal from a lone chat opened a terminal tab beside it, and the chat did not dock. A new workspace's screen is one split tree, and ScreenLayout.column(containing:) only searches real columns, so the chat's screen looked like it had none. LayoutScreen.column(containing:) gives the implicit column.

Commit	Result
b998d19 (red)	aSplitTreeScreenIsOneImplicitColumn failed: no columns, so the placement was "here"
977bcf8 (green)	passed: ChatColumnPlacementTests, NewTabKindTests, StripNewTabTests, EmptyWorkspaceNewTabTests

Live check on a capture mini (fleet build of 977bcf8, job 2677fd2a). The routed steps used real foreground key presses; actions run from the CLI count as scripts and skip this routing. The setup was a new chat (Cmd-I) with the pane's other tabs closed via the CLI (setup only).

• New Terminal (Cmd-Shift-Ctrl-T) from the lone chat: the chat moved into a left dock, and the terminal stayed in the old pane on the right with focus. The terminal started in ~, which the daemon picked because no cwd was sent.
• New Browser (Cmd-Shift-L) from the docked chat: a browser tab opened in the right pane.
• Cmd-T from the docked chat: the New Tab page opened in the right pane, not another chat.

The chat's dock kept only the chat throughout.

Cursor review fixes (after the base merge that took this PR to Swift only):

Commit	Result
f5b6444 (red)	hosted swift test (run 37860865718) fails theDockedChatWithoutAStripStaysHere (got .newColumn) and aZoomedChatDocksNothing (got .dockChat)
0a6fdc6 (green), 39ea26e (empty retrigger: the first run was refused by a runner disk floor)	hosted swift test (run 37866107013) passes ChatColumnPlacementTests, PanePaintHoldTests, DebugPageTargetTests

• The strip-less chat dock case (.newColumn) is gone: the daemon never keeps a dock without a scrolling column (an all-docked screen undocks), and the path dropped an explicit cwd and sent browsers into the dock.
• A zoomed screen maps to its zoomed pane alone, so a zoomed chat looked lone and docked; zoom now keeps the tab here.
• A URL-less browser from the docked chat takes the chat's context, and a browser a lone chat leaves when it docks is recorded for Auto tabs.newTabKind. No unit test covers these two wirings (nothing in the suite builds a PaneController with an agent page).

Proof

Window-only captures, with host names covered. Each image is before and after, side by side.

Changelog

Changed: in cmux-next, a terminal, browser or New Tab opened from the docked agent chat opens as a tab in the area to its right.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes user-visible tab/pane routing and layout moves for agent chat; gated on daemon capabilities but affects focus, cwd, and dock commits when enabled.

Overview
Routes new terminals, browsers, and New Tab pages opened from the docked agent chat (agent_chat role) into the scrolling strip instead of the chat dock, using new ChatColumnPlacement logic (recent strip focus, zoom, lone-chat detection).

When a lone agent chat fills a screen, user-initiated New Terminal/Browser docks the chat in a new left column via TabMoves.toNewDockColumn with role: .agentChat and a single daemon commit (respawn for the tool left in the pane). TabMoves now forwards optional dock role to the daemon (dock-column-role-v1).

Tab lifecycle wiring: strip targets get focus; terminals from the dock use daemonResolvesCwd so cwd inheritance follows the daemon resolver; URL-less browsers from the dock use the chat pane’s working context while opening in the strip; lone-chat dock can respawn a browser in the vacated pane.

Adds ChatColumnPlacementTests for placement rules (implicit single-column screens, zoom, strip-less dock edge cases).

^{Reviewed by Cursor Bugbot for commit 39ea26e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[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: 1adf1b10-9e5d-4352-923e-311d997b0d45

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

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

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.}

[github-actions]

Not verified: macOS jobs did not run: compile and app tests were skipped on 39ea26e8ea; ci-status is green without them.

CI passes on 39ea26e8ea (run 37866107424 attempt 1).
macOS jobs did not run: compile and app tests were skipped: Fast static checks skipped (the macos job needs both to succeed). Re-run CI once they pass, before merging.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); signatures are its SIGNATURES table. A machine verdict is the runner's fault, not this PR's; yours means the failing file is one this PR changes, also red on main that main's latest full suite fails the same way, seen on other PRs that it failed on another pull request's run lately.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[lawrencecchen]

Review (correctness read, nx-docked-column lane). Concrete defects:

1. The chat dock with no strip still takes tabs. .newColumn handles only a terminal (guard case .terminal = respawn else { return controller }). New Browser from that state, and Cmd-T (.agent/.page, respawn: nil), get controller back, so the browser tab, the New Tab page or a new chat opens inside the chat dock. That breaks the invariant in the summary ("it never gets tabs of its own"). For a browser, a new column can be opened with the browser as its first tab (move-tab-to-column after the tab is created, or a respawn-style create). For Cmd-T, the New Tab page can take the same path.
2. .newColumn drops the invocation's cwd and keep. route builds respawn: .terminal(SpawnOptions(cwd:keep:)), but .newColumn calls layoutModel.newColumn(after:), which spawns with the chat pane's selected-tab cwd and no keep. A person's New Terminal that carries a cwd (for example a Finder "New cmux Workspace Here"-style entry or a command action) opens in the wrong folder. Pass the respawn's options to the new column.
3. Minor: in newBrowser the early return when route gives nil (dockChat) skips noteUserChoice. tabs.newTabKind: auto therefore does not learn this browser choice. The terminal path has the same gap.

[cursor]

Stale Bugbot comment from a previous run.

[blacksmith-sh]

Found 1 test failure on Blacksmith runners:

Failure

Test	View Logs
unittest.case.FunctionTestCase (test_cmux_tui_artifacts_coalesces_queued_feat_cmux_next _pushes)/ unittest.case.FunctionTestCase (test_cmux_tui_artifacts_coalesces_queued_feat_cmux_next _pushes)	View Logs

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

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 5d3bca2. Configure here.}

[github-actions]

The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/37871997540 at 9e3a1a2 failed: cmux-next swift test.
Those jobs last passed at cfe3a43. This pull request is one of 4 merged since: #18716, #18718, #18124, #18170.
If the failure is in your change, fix forward on feat-cmux-next. Pull requests run only the tiers their paths reach (docs/ci/cmux-next-tiers.md); label a batch PR full-ci to run them all before merging.