fix(ci): keep cmux-next minis first through bot retries

[teamleaderleo]

Routing-only follow-up to #18182. Keep cmux-next macOS jobs on the owned minis through bot attempts 1 to 3, then use the Blacksmith fallback only after a fourth bot retry or an explicit non-owned route. The test covers every affected job and keeps human retries minis-first.

Changelog

Fixed retry routing so owned minis remain the first route.

Validation: python3 -m unittest tests/test_ci_side_lane_placement.py (16 passed).

---

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

---

Summary by cubic

Keeps cmux-next macOS jobs on the owned minis through bot run attempt 3 before falling back to Blacksmith on attempt 4.

• A rerun reuses the placement job's runner output, so bot attempt 3 previously dropped the side-lane label and went to Blacksmith; the label now survives until the rescue reruns a refusal as attempt 4.
• Human retries stay minis-first at any run attempt, and the test covers every affected job at attempts 3, 4, and 5.

^{Written for commit 109a9ac. Summary will update on new commits.}

---

Note

Medium Risk
Changes when cmux-next macOS jobs leave the owned fleet for Blacksmith during automated rescue retries; mis-routing could queue jobs longer or spill to paid runners earlier than intended.

Overview
Extends cmux-next macOS job routing so owned-pool placement (macos-placement output) is honored through bot attempts 1–3, not only through attempt 2. runs-on and CMUX_NEXT_RUNNER for swift-test, daemon-test, generated-files, release-compile, and cmux-scheme-compile now gate on (github.run_attempt <= 3 || triggering_actor != github-actions[bot]) before falling through to the Blacksmith overflow path—so a rescue re-run at attempt 4 (after a refusal on attempt 3, with placement output reused) is what moves jobs off the minis.

Workflow header comments document that retry sequence. test_ci_side_lane_placement.py adds coverage that bot attempt 3 stays on the placement label, bot attempt 4+ uses fallback, and human retries remain minis-first.

^{Reviewed by Cursor Bugbot for commit 109a9ac. 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: 8331ffaa-4793-4512-bb8d-f72d87942b34

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.}

[blacksmith-sh]

Found 23 test failures on Blacksmith runners:

Failures

Test	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='backend/packages/home-core/conformance/ conversation-cases.json')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='backend/packages/home-core/conformance/ conversation-import-cases.json')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='backend/packages/home-core/conformance/ conversation-search-cases.json')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='mux/packages/brain/conformance/ chief-cases.json')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='scripts/ci/cmux-tui-darwin-builder')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='scripts/ci/macho_weaken.py')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='scripts/ci/macos-cross.sh')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='scripts/ci/macos_stubs.py')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='scripts/ci/macos-stubs/x')	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='skills/cmux-browser/SKILL.md')	View Logs

...and 13 more test failures. View all on Blacksmith

^{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 109a9ac. Configure here.}

[cursor]

Unwatched retry stays on minis

High Severity

The new run_attempt <= 3 gate keeps a bot retry on needs.macos-placement.outputs.runner through attempt 3, but owned_rerun only watches cmux-next bot retries through attempt 2. The sweeper never adopts that third attempt or creates the attempt 4 Blacksmith overflow, so a refused or queued job can stay on the minis.

Additional Locations (2)

• .github/workflows/cmux-next.yml#L296-L299
• tests/test_ci_side_lane_placement.py#L197-L211

 

^{Reviewed by Cursor Bugbot for commit 109a9ac. Configure here.}