[shivamhwp]

Problem

Refs #33552.

A slow device hub can keep its detached agent-device daemon alive after desktop Quit. LocalDeviceHost waited for the hub to exit before requesting daemon shutdown, while desktop gives the backend five seconds to stop. The daemon has no idle timeout.

Fix

Stop the hub and daemon concurrently under the existing host lifecycle lock. Each helper retains its existing shutdown and ownership checks.

Evidence

Verified head 801f686fa67825157305b2c99af492fe5e4157c8, base ad9e87111fd9b2f10ec320b4de0a6ad3a092278a:

• GitHub CI and test output passed. All 20 LocalDeviceHost tests pass. The new Deferred-based regression fails on base because daemon cleanup waits behind the stalled hub finalizer.
• Boat job k4drzd7ejkmvuih4 ran vp run check:changed origin/main at this head and merge base. Format, lint, knip, GraphQL, affected typechecks, and tests passed, exit 0. The VM was confirmed deleted.
• Real desktop builds on macOS 27.0.1 arm64 on shiv and Ubuntu 26.04 x64 on corex. Both used Electron 44.10.1, agent-device 0.21.31, hub 0.19.1, isolated state and mock keychain. Helpers were enabled through the real Devices settings with background CUA controls. macOS used the unsigned repo-packaged T3 Code Beta 0.1.7 app, embedding the exact commits above.
• macOS Quit used AXPress on the native Quit menu item. Linux SIGTERM follows DesktopLifecycle through app.quit and before-quit. Each controlled stall used SIGSTOP only after checking the app-owned hub PID, argv and isolated state file.

Platform	Case	App/backend and hub after Quit	agent-device daemon
macOS	Base, ordinary	Exited	Exited
macOS	Base, hub paused	Exited after desktop grace	PID 70959 survived, HTTP 200 after Quit
macOS	Head, ordinary	Exited	Exited 132 ms after Quit
macOS	Head, hub paused	Exited after desktop grace	Exited 119 ms after Quit, while hub remained paused
Linux	Base, ordinary	Exited	Exited
Linux	Base, hub paused	Exited	PID 4101264 survived
Linux	Head, ordinary	Exited	Exited
Linux	Head, hub paused	Exited	Exited

In the macOS stalled case, base Quit was at 21:32:45.966Z. The backend exited at 21:32:51.039Z, while daemon 70959 still returned HTTP 200 at 21:33:00.923Z. At head, Quit was at 21:39:06.525Z and daemon 76506 exited at 21:39:06.644Z. The paused hub and backend exited at 21:39:11.588Z. The captured daemon argv used the packaged T3 app executable in both builds. All test processes were stopped afterward, including the surviving baseline daemons.

macOS controlled reproduction before the fix. Settings enables the real helpers, then native Quit ends the window recording. The process observations above establish the surviving daemon.

macos-base-stalled.mp4

macOS after the fix, same Settings and native Quit flow with the hub deliberately paused. The daemon exits before the hub.

macos-head-stalled.mp4

Ordinary native Quit also passes before and after:

macos-base-ordinary.mp4

macos-head-ordinary.mp4

The macOS window recordings finalize playable video when Quit removes the capture target, with recorder error SCStreamError -3815. They show helper startup and end at Quit; native AX action output and OS process observations establish shutdown. Linux full-display recordings returned black frames, so its process observations and per-window capture are the usable evidence. The installed GitHub CLI accepts media attachments only, so text logs could not be attached.

Not shown: the reporter's original build or reason for its nine-day daemon lifetime, naturally occurring hub stalls, active device sessions, or a signed release installation. Earlier shutdown work and daemon stop itself exceeding the desktop grace period remain outside this fix. A paused hub can still leave stale hub.json after forced backend shutdown. This change prevents it from blocking daemon cleanup.

Two independent Codex review sessions completed. Round 1 found no production correctness or test-signal issues. Round 2 resolved the macOS evidence blocker with no new findings.

Implemented by Claude Opus 5.5 through Claude Code. Investigated and verified by GPT-6.1 Sol through Codex.

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: 9a35ab0
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: c9c917a3-2c64-4a8c-8a84-3975ce9e40a2

Commits

Reviewing files that changed from the base of the PR and between 00c597b and de873a4.

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

No actionable comments were generated in the recent review.

Walkthrough

LocalDeviceHost.stop now requests hub and agent-daemon shutdown concurrently. It removes hub.json after hub shutdown completes. A test checks that the daemon stop request occurs while hub cleanup is pending.

Changes: Local device shutdown

Layer / File(s)	Summary
Parallel hub and daemon shutdown apps/server/src/device/LocalDeviceHost.ts, apps/server/src/device/LocalDeviceHost.test.ts	LocalDeviceHost.stop starts hub and agent-daemon shutdown concurrently. Hub-state file removal remains sequenced after hub shutdown. The test verifies the daemon stop command while hub cleanup is pending, then confirms that hub.json is removed.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Possibly related issues

• pingdotgg/t3code issue 33552: The PR requests daemon shutdown before hub cleanup completes, matching the issue's reported graceful-Quit daemon behavior.

Possibly related PRs

• pingdotgg/t3code#33587: This PR changes the daemon-stop cleanup that runs concurrently with hub shutdown here.

Suggested labels: bugfix

Suggested reviewers: only21mil, juliusmarminge

Merge Risk: Minimal · previous review 0eac8

The change allows daemon shutdown to begin without waiting for the hub. No actionable merge-blocking issue remains after normal checks.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly summarizes the main change: stopping the device daemon alongside a slow hub shutdown.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.
Description check		The description follows the template's main sections: Problem, Fix (Change), Evidence (Verification), and an issue reference. It states the root cause, the fix, and extensive verification with CI resu…

Full details: Description check

Explanation

The description follows the template's main sections: Problem, Fix (Change), Evidence (Verification), and an issue reference. It states the root cause, the fix, and extensive verification with CI results, platform test tables, and recordings. It also lists what was not tested, which the template asks for. Minor gaps: the Scope section does not clearly say why the change is a bug fix rather than a feature, and the title is not shown in the description. The template's required sections are largely covered, so the check passes.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: 17a6a01
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 6ecbb7ff-55fd-4caa-91f5-b51e8d662c28

Commits

Reviewing files that changed from the base of the PR and between 801f686 and 0eac8c9.

Files selected for processing: 2

• apps/server/src/device/LocalDeviceHost.test.ts
• apps/server/src/device/LocalDeviceHost.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.

[macroscopeapp]

Approvability

Verdict: Approved at 0eac8c9

Macroscope's review found this PR approvable — Local-device shutdown now stops the agent daemon alongside hub cleanup, preventing a slow hub exit from delaying daemon shutdown. Existing locking, state-directory scoping, and cleanup ordering are preserved, keeping the change narrowly scoped.

No code changes detected at de873a4. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}

[maria-rcks]

Note

Written by claude-opus-5-5 on behalf of Maria

Merged current main (no rebase) to clear the conflict with #33587, which rewrote stopAgentDeviceDaemon to read the recorded daemon file instead of agentToolRef.

• LocalDeviceHost.stop keeps this PR's concurrent hub/daemon shutdown and now calls main's stopAgentDeviceDaemon.
• The test fixture keeps main's runs log and previous-daemon argument, plus this PR's hub.cleanup hook and daemonStopRequested deferred. All four daemon-stop tests (main's three and this PR's one) pass.
• vp run check:changed passes (format, lint, knip, typecheck, tests). No open review findings.

Dismissing prior approval to re-evaluate 0eac8c9