[teamleaderleo]

HOLD: stacked on #26920, which is on #26801 and Swift #26791, all held. This PR changes Swift too.

Composer Cloud: a chat starting in Cloud shows how the start goes as metadata under the person's message. The card shows:

• the step: getting a machine, copying the folder, opening the chat;
• the folder;
• the machine and its size;
• the secrets the machine gets;
• its network.

It is never a modal and has no Full Access warning. Working is a spinner. A failure shows the host's words, and the message stays in the composer.

Leo, 2026-10-28: "whatever's convenient for cloud, like that can be productionized, operationalized, we ought to just put out there"

Leo, 2026-10-28: "you can just say ... where should this chat run and stuff, and that's just about it ... the sort of metadata that you get to see, the UI elements ... their little warning for the full access thing, but I don't know if we really need to do that"

Gallery: agent-pane.cloud-start

What changes

• cloudStartProgress.ts (new): the card's state.

  • The page starts it from the send: the message and the folder's name.
  • It takes the host's pushes field by field, each one checked: only for the same start (its cloud.startChat key), a step never goes back, text is one line and capped, and anything else is ignored.
  • An answer clears it when the chat opened (the tab closes), or marks it failed with the host's words.

• CloudStartView.tsx and cloudStart.css (new): the card, in three layouts behind the cloud-start-look experiment:

  • chips (default): a spinner line with the step, then one row of small chips;
  • rows: label and value rows, like the Sandboxed receipt;
  • steps: a three-step checklist with the facts on one line.

  I picked chips because it reads as metadata, like a "Building environment" line. It stays one or two lines while the start runs for minutes, and it doesn't push the composer down. The rows layout spends four lines on facts that matter less than the step. The steps layout spends three lines on a checklist.

• App.tsx:

  • the card shows in the chat view while a new chat's Cloud start runs, and over the box on the New Tab page (without the bubble, since the message is still in the box);
  • the bridge takes cloudStartProgress;
  • another chat clears the card, and so does picking another place to run after a failure.

• pageHost.ts: the shared page host hands the cloudStartProgress event to the bridge.

• Swift CloudStartChat pushes each step as the start reaches it (machine, then folder, then opening), with the folder's name and, once known, the machine and its size. AgentPaneCloudStart.progress(_:folder:machine:size:to:) delivers the event.

• Swift AgentPageChats.seed: the Cloud start closure is written out with its type. Xcode 26.6 could not type the map form that cmux NEXT: Cloud as where a new chat runs, a row in the composer's environment picker #26801's base has.

Not yet filled

With each step the host pushes the folder, the machine and its size ("4 CPU · 16 GB"). Secrets and network have fields and show when pushed. They come from the derived environment, which is being built on the server side and isn't pushed yet.

Before / after

• Before: a Cloud send left the new chat empty while a toast said it was moving.
• After: the message with the start's progress under it. See the agent-pane.cloud-start gallery entry:
  • variants: sent, machine, copying, opening, failed;
  • Compare arms shows the three layouts.

The gallery advisory captures only PRs into feat-cmux-next, so it runs once this PR's base retargets.

Tests

• cloudStartProgress.test.ts: begin, pushes (checked fields, other keys, a step that never goes back, after a failure), and the answer.
• CloudStartView.test.tsx: every arm shows the message, a spinner and the same facts; the steps arm marks done, current and next; a failure shows the host's words with no spinner; no bubble on New Tab.
• test/cloud-start-host-event.test.ts: the host event reaches the bridge.
• cmux-ci (Linux pool): those tests, cloudRun, NewTabScreen and design-selection pass, and bun run check is green.
• Fleet: AgentPaneCloudStartTests and CloudWorkPickTests.

Risk: low, behind Labs and only for a Cloud start.

Agent session: agent session

🤖 Generated with Claude Code

[teamleaderleo]

Owned by an agent session · session

[teamleaderleo]

Review (codex gpt-6-astra): SHIP at 072682b

Three rounds, run with high effort in fast mode:

• Round 1 (fixed): the host pushes the folder, the machine and its size; a failed card shows whenever it is set and goes when the mode leaves Cloud.
• Round 2: one finding misread the code. HOST_EVENTS is the one stream name and carries every kind; there is no list of kinds. It was answered with the code line and the host-event test.
• Round 3: SHIP.

Known limit: secrets and network show once the host pushes them. Neither has a source on the client yet; it comes from the derived environment on the server side.

At this head:

• cmux-ci (Linux pool): 31 tests pass and bun run check is green.
• Fleet: AgentPaneCloudStartTests and CloudWorkPickTests pass.

Held per the body's HOLD line, since it is Swift and stacked.

[blacksmith-sh]

Found 42 test failures on Blacksmith runners:

Failures

Test	View Logs
test/accent-glass-invariants.test.ts/accent-glass-invariants.test	View Logs
test/accent-glass-invariants.test.ts/accent-glass-invariants.test	View Logs
test/agent-pane-agents-pill.test.ts/agent-pane-agents-pill.test	View Logs
test/agent-pane-agents-pill.test.ts/agent-pane-agents-pill.test	View Logs
test/agent-pane-peer-chat.test.ts/agent-pane-peer-chat.test	View Logs
test/agent-pane-peer-chat.test.ts/agent-pane-peer-chat.test	View Logs
test/agent-pane-queue-row-click.test.ts/agent-pane-queue-row-click.test	View Logs
test/agent-pane-queue-row-click.test.ts/agent-pane-queue-row-click.test	View Logs
test/agent-pane-transcript-spacing.test.ts/agent-pane-transcript-spacing.test	View Logs
test/agent-pane-transcript-spacing.test.ts/agent-pane-transcript-spacing.test	View Logs

...and 32 more test failures. View all on Blacksmith

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

[cmux-bot]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Probably yours: agent-pane-dock-gutters.test.ts, composer-caret-focus.test.ts, composer-caret-focus.test.ts and 3 more fails here but not on main's latest full suite. Seen on other PRs too (likely flaky): agent-pane-dock-gutters.test.ts, agent-pane-question-records.test.ts, agent-pane-queue-steer.test.ts and 15 more.

CI failed on bb706c72de (run 40764881815 attempt 1): 3 code, 1 unknown.

Job	Verdict	Why
web / react-apps-base-check (2)	code	a web test failed
web / react-apps-base-check (4)	code	a web test failed
web / react-apps-base-check (3)	code	a web test failed
web / react-apps-check	unknown	no known signature; failed step: Check webview test shards

Failures:

• new in this PR agent-pane-dock-gutters.test.ts chromium > the gutter beside every part of the stack scrolls the transcript, and the part keeps the ...: failed (job)
• seen on other PRs agent-pane-dock-gutters.test.ts webkit > the gutter beside every part of the stack scrolls the transcript, and the part keeps the po...: failed (job)
• seen on other PRs agent-pane-question-records.test.ts answered live, the card goes and the record takes its place: failed (job)
• seen on other PRs agent-pane-queue-steer.test.ts queued and steering prompts > the queue takes its own place in the stack: nothing is cut off or draw...: failed (job)
• new in this PR composer-caret-focus.test.ts chromium: New Tab composer: hover gives the hovered button alone the soft pill; the box stays put: failed (job)
• new in this PR composer-caret-focus.test.ts chromium: New Tab stage composer: hover gives the hovered button alone the soft pill; the box stays ...: failed (job)
• new in this PR composer-caret-focus.test.ts chromium: chat composer: keyboard focus gives a button the soft pill and the bar's outline: failed (job)
• new in this PR composer-caret-focus.test.ts chromium: New Tab composer: keyboard focus gives a button the soft pill and the bar's outline: failed (job)
• seen on other PRs accent-glass-invariants.test.ts no orange or system accent and no opaque glass on the gallery stages: failed (job)
• seen on other PRs agent-pane-agents-pill.test.ts by default the running agents sit in the tray above the composer, not in the transcript: failed (job)
• ... 14 more in the job logs

Matched log lines

web / react-apps-base-check (2): (fail) chromium > the gutter beside every part of the stack scrolls the transcript, and the part keeps the pointer [8583.97ms]
web / react-apps-base-check (4): (fail) no orange or system accent and no opaque glass on the gallery stages [816296.44ms]
web / react-apps-base-check (3): (fail) the summary card lists the chat's background tasks: running ones first, the finished folded [956.97ms]

Not re-run automatically: web / react-apps-base-check (2), web / react-apps-base-check (4), web / react-apps-base-check (3), web / react-apps-check are not machine failures.

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.

CI fast guards failed on bb706c72de (https://github.com/manaflow-ai/cmux/actions/runs/40764881815). It does not block the merge; a red guard merged into main breaks it for every open PR.
The log named no failed step; see the run.

Agents: python3 scripts/ci/guard_attribution.py fix applies the mechanical fixes locally.

[teamleaderleo]

Review (codex gpt-6-astra): SHIP at bb706c7

A delta review of 072682b..bb706c7, given the full changed file. The agent-pane.cloud-start gallery entry's experiment script had no steps, which the gallery's validateEntries rejects, so the entry rendered an error card. It now has one step, which waits for the card's status line.

Checked on cmux-ci (Linux pool): gallery-coverage.test.ts "the entries are valid" passes, and all 5 variants × 3 arms render ready in Chromium.

[teamleaderleo]

Folded into #28686 (original commits kept, on current feat-cmux-next). Review and landing continue there; the owner closes this one.

[teamleaderleo]

Owned by an agent session · session · took over from an earlier session (its owning session is closed)

[teamleaderleo]

Closing as a duplicate: #28686 folds this PR (with #26801, #26920 and #26921) into one PR on feat-cmux-next and carries it from here. This branch stays on the remote.