[teamleaderleo]

A machine-side change (cmux-tui, cmux-host or acpmux) can now be checked end to end on a real Cloud machine in about a minute, with no image bake or nightly: bun scripts/cmux-next/cloud-devloop.ts --credentials FILE.

Leo, 2026-11-05: "Just running commands and then we're seeing, Oh, okay, we gotta fix this. Then we gotta fix that ... we hill climb within, like, you know, several minutes, we have multiple iterations within that instead of, like, one iteration every couple minutes."

What changes

scripts/cmux-next/cloud-devloop.ts: one kept Cloud machine (cloud-devloop, created on first use) is the dev box. One command:

1. rsyncs the worktree's changed files to a warm checkout on a Linux build host (--build-host or CLOUD_DEVLOOP_BUILD_HOST);
2. builds cmux-tui, the guest's one multi-call binary (dev profile, stripped, the guest's Ubuntu release, baseline zig CPU for libghostty-vt);
3. pushes a zstd --patch-from delta of what the machine already runs (0.3 MB after a one-line acpmux edit) in 512 KiB fs/write pieces, SHA-256-checked;
4. flips /opt/cmux/current to a dev profile and restarts cmux-host and acpmux.service; --restore undoes it;
5. runs the Cloud journey on that machine and prints the first red step with the journal lines since the run began.

cloud-journey.ts:

• --machine ID runs on an existing machine, keeps it, and resets an earlier run's model stub.
• --keyed-start checks that the app's keyed acpmux start still holds after acpmux.service's restart delay, which is the take-over race of cmux-platform#845.

Measured (production, agent test account)

iteration	build	push	install and restart	journey	total
one-line acpmux edit	20.7 s	5.9 s	14.6 s	22.4 s	66.8 s
no change	0.9 s	none	none	20.9 s	27.2 s

--keyed-start was red on production before #845 deployed and green after, 36 s each run.

Tests

cloud-devloop.test.ts and cloud-journey.test.ts: 21 pass and 0 fail on the fleet.

Changelog

none

Agent session: agent session

🤖 Generated with Claude Code

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Machine-side changes (cmux-tui, cmux-host, acpmux) can now be checked end to end on a kept Cloud machine in about a minute, with no image bake or nightly. bun scripts/cmux-next/cloud-devloop.ts --credentials FILE rsyncs the worktree's changed files to a warm fleet Linux build host, builds the multi-call cmux-tui binary there, pushes a zstd --patch-from delta in 512 KiB pieces, flips /opt/cmux/current to a dev profile, restarts the host and acpmux units, and runs the Cloud journey on that same machine, printing the first failing step with the machine's journal lines.

• One-line acpmux edit: build 20.7 s, push 5.9 s, install+restart 14.6 s, journey 22.4 s, about 67 s total; a no-change rerun is about 27 s.
• --restore puts the image's binary back; --snapshot creates the dev machine as the next base image.
• cloud-journey.ts gains --machine ID (run on an existing machine and keep it, resetting an earlier run's model stub) and --keyed-start, which verifies the keyed acpmux start still holds after acpmux.service's restart delay — the take-over race of cmux-platform#845.
• 21 tests pass, 0 fail on the fleet.

^{Written for commit db0ef2f. Summary will update on new commits.}

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[github-actions]

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

Review db0ef2f: 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=pr36251 -f ref=lane/cloud-devloop -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4200417289001 (42abbd2b81)
• cmux NEXT nightly 1.0.0-nightly.4201868763201 (5052d6545e)

The feat-cmux-next push run https://github.com/manaflow-ai/cmux/actions/runs/41992440601 at c293fc3 failed: cmux-next crash ends the app.
Those jobs last passed at 3deee62. This pull request is one of 3 merged since: #36285, #36207, #36251.
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.

[teamleaderleo]

Owned by an agent session · session

[blacksmith-sh]

Found 3 test failures on Blacksmith runners:

Failures

Test	View Logs
nightly cmux-tui resolve/nightly cmux-tui resolve	View Logs
test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)/ test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)	View Logs
the six-hour schedule must allow 90 minutes for a cold cache build and use the matching runner, Xcode, and visible timing output/ the six-hour schedule must allow 90 minutes for a cold cache build and use the matching runner, Xcode, and visible timing output	View Logs

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

[teamleaderleo]

Review (codex gpt-6.1-sol xhigh, next-review, since=20261105T073655Z): SHIP at db0ef2f

SHIP at db0ef2f

Visual verdict: no images; Leo tries the build

1. • [P1] [follow-up] [category:correctness] [finding:F1] [outcome:pending] cloud-devloop.ts:503 uses the nonexistent cloud.machine.snapshot operation, so every --snapshot invocation fails with an unknown-operation error. Call cloud.snapshot.create instead.
2. • [P2] [follow-up] [category:correctness] [finding:F2] [outcome:pending] cloud-journey.ts:497-503 resets only Claude’s stub endpoint. The Codex stub writes /home/cmux/.codex/config.toml, which remains pointed at the old loopback provider on later --machine --agent codex runs without --model stub.
3. • [P2] [follow-up] [category:correctness] [finding:F3] [outcome:pending] cloud-devloop.ts:270-276 searches only the first cloud.machine.list page. Since the API is paginated (default limit 50), an existing named dev machine beyond that page is missed and another machine is created.
4. • [P2] [follow-up] [category:reliability] [finding:F4] [outcome:pending] cloud-devloop.ts:280-286 proceeds after the 300-second status loop even if the machine is still not running, causing later install/exec failures instead of reporting a provisioning timeout.

Follow-ups:

• [finding:F1] Fix the snapshot operation.
• [finding:F2] Reset the Codex stub configuration.
• [finding:F3] Page through machine listings.
• [finding:F4] Fail explicitly when the machine-start timeout expires.