[teamleaderleo]

What

A chat sent to cmux Cloud gets its machine chosen from the chat's folder, with no machine, size or environment prompt.

CloudWorkPick (CmuxNextCloud) reads, from the chat's working directory:

• the repo key (host/owner/name, lower-cased) from the remote, never a login or password from the URL;
• the GitHub repo a new machine can clone, at the current branch when it is pushed (an unpushed branch clones the default branch);
• the checkout's git object size and its root file names (Package.swift, Cargo.toml, package.json, …), which give the server the toolchain;
• the task kind. The rule is simple and documented on CloudTaskKind.inferred(from:): a whole word naming testing makes it test, else a word naming building makes it build, else chat. It reads the chat's title;
• whether .cmux/machine.toml exists, and whether the agent is Codex (which needs the Codex broker).

Every git command is local, with no network, and bounded at 3 s. A command that fails just leaves its fact out. A folder outside git gets no repo, and its root files still go.

Where it's used:

• Move to Cloud when there is no machine to reuse. The existing reuse order (the named machine, else a connected one, else any) is unchanged.
• New Cloud Machine asked from a chat (the composer's row names its tab). The setup offers the chat's own repository as the starting point. The machine is sized for the chat only when the person keeps that repository.

Both send autosize: {taskKind, repo: {key, sizeMb, rootFiles}} with POST /api/vm (#26617, #26630). A memoryMb still wins.

The answer's sizePick {vcpus, memoryGb, pick, reason} is decoded on CloudMachine.sizePick. It is also kept on CloudMachineSession.sizePick, which a list read does not overwrite. That copy is what the hand-off reads: CloudSizePick.handoffSize is POST /v1/handoffs's size.

Why

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: "like as auto as chatgpt work and not NEEDING to deal with picking environments... the auto thing situation". And: "You know how we did that automatic routing, automatic sizing ... We want to have that for the cloud machines too ... Like ChatGPT work."

Risk

• Everything here is behind Labs preview features, the design's rule for new work-mode surfaces. With Labs off, nothing is read and every create is byte-for-byte as before.
• With Labs on, a moved chat's new machine takes the server's pick instead of the plan's default size. A plain chat may get a smaller machine than before; that is the server table's call (web: auto-size a handed-off task's cloud machine (fleet sizing model) #26617).
• Reading the pick runs up to five local git commands and one directory listing, only when a machine is about to be created.
• No new strings and no UI change. The setup dialog's starting point is prefilled only for a chat in a GitHub repo, with Labs on.

Tests

CloudWorkPickTests covers:

• a Swift package, Rust and Node repos with GitHub and non-GitHub remotes;
• unpushed branches;
• a non-git folder;
• the task-kind rule;
• that remote keys never carry credentials;
• the upstream track parse;
• sizePick decoding, where a malformed one drops the line and never the machine;
• that the setup is sized only for the chat's own repository.

CloudWorkspaceSetupWizardTests covers the suggested starting point. These ran as package tests on the Mac fleet.

No capture: with Labs preview features off (the default), nothing changes on screen. With Labs on, nothing new is drawn either. The one visible difference is a field's starting value: when New Cloud Machine is opened from a chat's composer, its setup dialog starts on "From repository" with the chat's own repo filled in (CloudWorkspaceSetupWizardTests.aChatsRepositoryIsTheSuggestedStartingPoint). No capture scene opens that dialog from a chat. The size line in the chat is the follow-up UI PR, which comes with before/after captures.

Agent session: agent session

🤖 Generated with Claude Code

[teamleaderleo]

Owned by an agent session · session

[github-actions]

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

Review 01564b7: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr26740 -f ref=leo/cloud-autosize-pick -f platform=macos -f backend_mode=local

web and size checks: green

[teamleaderleo]

Bead cx-egdz7x created for this PR by the cmux coordinator.

[teamleaderleo]

Review (codex gpt-6-astra): SHIP at d3b4891

Three rounds of review at high effort with fast mode. Round 1 found three issues, all fixed in d3b4891:

• The clone ref now names the upstream's branch, not the local branch.
• A local-path remote names no GitHub repo.
• No pick is read on the app-server link, whose create takes no autosize.

Round 2 asked about isDefault. The answer, with code lines, is that autosize is set only from a pick that requires the legacy link, and the configuration is immutable. Round 3 verified it and returned SHIP with no findings.

[teamleaderleo]

Review (codex gpt-6-astra): SHIP at 9b73ba6

The same reviewed change, rebased onto feat-cmux-next to pick up #26754, which fixes the base Release-compile red (CloudHandlers+Account.swift:146, not this change's line). The diff is identical to the reviewed head d3b4891: git patch-id --stable gives 5fcae646 for both.

[blacksmith-sh]

Found 2 test failures on Blacksmith runners:

Failures

Test	View Logs
test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)/ test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)	View Logs
test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)/ test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)	View Logs

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

[teamleaderleo]

Review (codex gpt-6-astra): SHIP at 8a93094

Delta review of the hand-resolved merge of feat-cmux-next into this branch: the remerge diff, the full PR diff against the merged base, and the full resolved files. No findings. Still held by the fleet-wide Swift merge hold (HOLD line in the body).

[teamleaderleo]

Orphaned: no activity for 15 h, and no lane holds a lease on it. Adopted by the orphan sweep: merged feat-cmux-next in at 2b98a87. A later pass queues it for the merge hub if it is reviewed, small and red only on base reds.

[teamleaderleo]

Review (codex gpt-6-astra high, fast, via review-router): BLOCK at dcf23b5

1. P2 — Incorrect branch selection for remapped tracking refs — CloudWorkPick.swift:77. upstream:lstrip=3 strips components from the local tracking ref, which need not match the remote branch. For example, remote team/origin tracking feature produces origin/feature. Accepting the suggested repository then provisions a machine whose bootstrap fails because that branch does not exist. Use upstream:remoteref, remove its refs/heads/ prefix, and add tests for slash-containing remote names and custom fetch mappings.

Static review only; tests were not run in this sandbox.

[teamleaderleo]

Review (codex gpt-6-astra high, fast, via review-router): SHIP at 01564b7

1. P3 — CloudWorkPick.swift:127: maxSplits: 1 leaves the remote-name field inside track. A synchronized branch tracking a remote named ahead is therefore classified as unpushed, causing the wizard to suggest the default branch. Inspect only the second NUL-delimited field and add a three-field regression test.

No P1/P2 findings. Static review only; tests were not run.

[github-actions]

The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/40999472385 at c62a8db failed: cmux-next checks (god files, concurrency, crash safety, l10n).
Those jobs last passed at cfcbe8d. This pull request is one of 1 merged since: #26740.
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.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4100000843001 (e58eb649c8)
• cmux NEXT nightly 1.0.0-nightly.4100052035301 (1cde021b9e)