[lawrencecchen]

Review (claude-review-subagent): SHIP at cb8920a

Summary

The nightly-guards / workflow-guard-tests / release-notary job (and its wrapper nightly-guards / Guard status) is red on every feat-cmux-next push and on unrelated PRs.

Which command fails, per run of "cmux-next guards" on feat-cmux-next:

Run	Head	Failing commands in release-notary
41496150525	bbb2b89 (#32457)	tests/test_nightly_universal_build.sh
41500471225	ae802cf	test_nightly_universal_build.sh, test_nightly_promotion_admission.py
41506508608	55c0b5c (#32514)	the two above, plus tests/test_nightly_next_arm64_only.py
41507839428	480eb3c (after #32529)	tests/test_nightly_next_arm64_only.py only
41512462969	5b91230	tests/test_nightly_next_arm64_only.py only

#32529 already fixed the two reds that #32457 started. The red that is left comes from #32514: the "Build nightly app (Release)" step now sources scripts/ci/nightly-steps/app-host-build-root.sh. test_nightly_next_arm64_only.py runs that step's real text in a temp root with recorder executables, and that root did not have the script. Bash stopped before xcodebuild with app-host-build-root.sh: No such file or directory, and the test raised CalledProcessError. (The FileNotFoundError: 'xcodebuild' and JSON "Expecting value" lines in the log come from other tests that pass; they are expected output.)

The fix: the recorder root gets a copy of app-host-build-root.sh, as it already gets compilation-cache.sh. With no BUILD_TRACK the script keeps the slot's own tree (build_root=$PWD), so xcodebuild still runs in the recorder root. No assertion changes; the guard checks the same ARCHS and helper architectures for all four tracks.

Testing

Command (Linux, no build; CI sets GITHUB_STEP_SUMMARY, so set it locally):

GITHUB_STEP_SUMMARY=/tmp/s.md python3 tests/test_nightly_next_arm64_only.py

• Base 562235a: red. subprocess.CalledProcessError on the "cx-f58x ... browser-import" Release step; stderr /bin/bash: line 82: scripts/ci/nightly-steps/app-host-build-root.sh: No such file or directory.
• Head cb8920a: green. PASS: nightly-next arm64 only.

python3 scripts/cmux-next/prepush-gate.py: no new violations.

Observed / Not run

• Observed: the command above red on base and green on head on this Linux VM; the per-run failing-command table above from gh run view --log-failed.
• Not run: the full release-notary group locally; the CI run of this PR shows it.

Changelog

none

[github-actions]

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

Review cb8920a: 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=pr32548 -f ref=fix/base-red-notary-32457 -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4152029422301 (8219b57698)
• cmux NEXT nightly 1.0.0-nightly.4152127188001 (ec14e3bc84)
• cmux NEXT nightly 1.0.0-nightly.4152325530501 (5a3ac69758)
• cmux NEXT nightly 1.0.0-nightly.4152550616901 (edcadab7a6)
• cmux NEXT nightly 1.0.0-nightly.4152605376201 (91798945c6)

[lawrencecchen]

Review (claude-review-subagent): SHIP at cb8920a

Independent review: red on base and green on head confirmed; a mutation (archs="arm64 x86_64" in a copy of app-build-release.sh) makes the test fail, so the guard does not pass vacuously. #32514 added the same copy to tests/test_ci_fleet_compilation_cache.py but not to this test; this PR completes it. Note: the job has been red since #32457, but the red that is left comes from #32514 (55c0b5c); #32529 cleared the reds #32457 started.