[teamleaderleo]

Settings and palette agent updates now create terminal jobs on their captured machine and session through one shared handler. Rust validates the command; the handler preserves the intent key and refuses scripts and App callers.

Settings shows receipts and exit outcomes, and blocks another update while the outcome is unsettled. Intent retention is currently in memory.

Validation: three fleet suites, nine web tests and TypeScript checks pass; regressions fail before the fix. Status messages cover 21 locales. Existing localization and package-group checks still fail.

Stacked on #21439.

Changelog

Changed: Settings and palette agent updates run on the selected machine with visible outcomes.

[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: bf16ed46-d28e-4a9a-bed6-cda6c8a2463b

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

• Autofix · 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.}

[github-actions]

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

Review 6f00b00: 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=pr21447 -f ref=codex/agent-update-shared-ui-b9db4d92 -f platform=macos -f backend_mode=local

Pending: CI is running on 6f00b0087e (run); the result below was for 135d11526f and no longer applies.

Last result: Unclear: web / react-apps-check failed: no known signature; failed step: Check webview test shards.

Written by scripts/ci/classify_failures.py (ci-failure-attribution.yml); it is rewritten when this head's CI completes.

red: cmux-next checks (god files, concurrency, crash safety, l10n); NEW: red check: Tier routing (+4 more) is not red on the base 550ef1d; do not merge until green; see lane-rules

CI fast guards failed on 135d11526f (https://github.com/manaflow-ai/cmux/actions/runs/39411418068). 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.

[github-actions]

CI fast guards failed on ff11579374 (https://github.com/manaflow-ai/cmux/actions/runs/39201256654). 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. This comment is updated in place on each push.

[teamleaderleo]

Visual preview of 6f00b00: in the maintainers' private review feed.

[teamleaderleo]

Bead cx-sqy1l created for this PR by the cmux coordinator.

[teamleaderleo]

cmux-next guards / nightly-guards / workflow-guard-tests / release-notary is red on 21 of 21 fresh feat-cmux-next PR heads in the last 2:00:00, so it is most likely a base red, not this PR's. The base-reds lane has it; a base merge picks up the fix once it lands.

[greptile-apps]

Failed starts block later updates

When terminal creation fails with operation.failed, start() keeps an unknown job and saves its key for replay. The owner can store that failure and return it on every replay, but refresh() only clears agent_cli_job.* errors. That CLI then stays blocked until the app restarts, even though no terminal was created.

Treat a confirmed, stored creation failure separately from a lost reply.

[greptile-apps]

Finished updates become blocked again

When two refreshes overlap, an older pending reply can arrive after another refresh saved exited. refresh() writes each reply without checking the saved state or revision, so the older reply marks the finished update as running and blocks further updates until another successful refresh.

Keep exited final, or reject older observations.

[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.

[teamleaderleo]

Fixes from the review, at 6f00b0087e1d

Each failing test was committed before its fix:

1. The major finding: a lasting failure kept an update unsettled forever. Red test 7502ddfdcc5f, fix 0ad8305a5bbf, narrowed by red test 0902535c4db2 and fix 445644ece38f, then by red test 47f662a2ec34 and the fix after it.
   • First start (a fresh key): any coded answer from the owner now settles the job, for example operation.failed or daemon_shutdown_pending. The one exception is mutation.indeterminate: the owner says the effect may have run, so that job keeps blocking its CLI.
   • Replay of a lost reply: settles only on an agent_cli_job.* refusal, as before. A second review showed that a coded answer sent before the owner reads its journal can arrive while the lost start has already created the terminal.
   • Trade-offs left for Recover agent updates across app restarts #21469:
     • A lost reply whose creation then failed is still replayed until the owner session ends. Recover agent updates across app restarts #21469's journal resolve settles that case.
     • Rarely, a first-start operation.failed follows a terminal that has already gone public, through a failed launch activation. Here only the in-memory job is dropped. Recover agent updates across app restarts #21469 keeps its ledger row instead.
2. The minor finding: jobs only advanced on a Settings refresh. A start that finds an earlier job of the same CLI still unended now refreshes once before refusing. A job that has already exited no longer blocks the next palette or Settings update. One that still runs keeps refusing.

Reviews. Two independent delta reviews traced every new test by hand: red before its fix, green after. No relay, socket or allowlist change.

Verification gap. The Swift tests did not run. Pull requests into feat-cmux-next run Swift tests only with full-ci or a dispatch, and this host has no Swift.

Reds at this head all fail on the base too:

• react-apps-check 1-4 and Web status: the failing tests are in areas this PR doesn't touch:
  • settings/strings.test.ts: the base's DensityCard uses settings.appearance.density.help, which no catalog has at base 550ef1d99f6c.
  • check-layers: ui/findPill.css.
  • The markdown, What's New, status-glyph and composer suites.
• defaults-snapshot: red on every fresh head.
• crash-class rules: the findings are in base FilesPane/MarkdownShare files.
• cmux-next checks: the aggregate of the above.
• autofix repairs: the CI installation hit GitHub's API rate limit. I re-ran it.
• release-notary: base was red earlier, then fixed. Its test passes on the base tree at d3a9df0c733d.