Sleep stops a workspace's native chat agents

[brennanb2025]

	Files	Added	Deleted	Net
Test	22	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​677	$\color{#cf222e}{\Huge{\mathbf{−}}}$​112	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​565
Prod	58	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1206	$\color{#cf222e}{\Huge{\mathbf{−}}}$​313	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​893

ELI5

The problem. Sleeping a workspace stopped its terminals and browsers but left its native chat agents running: the claude or codex process behind each chat tab. A workspace that only had chats couldn't be slept at all, because Sleep was greyed out.

What's different for you. Sleep now also stops the workspace's chat agents. This applies to Sleep from the sidebar menu, bulk and descendant Sleep, the phone and the command line. The chat tabs and their history stay. Your next message in a slept chat starts the agent again in the same conversation. Chat-only workspaces can now be slept.

• A chat that is replying is stopped right away, without waiting for the turn to finish. In QA, claude exited about half a second after Sleep.
• After Sleep, the chat's row keeps its last result (for example "Done") with nothing running.
• If a chat can't be confirmed stopped, for example because its host is unreachable, everything else still stops. You get one toast, and the workspace isn't counted as asleep.
• When the same workspace is open on two hosts (say this Mac and an SSH machine), Sleep follows main's rule from [hardening 4/5] fix(sidebar): confirm before sleeping a workspace that another host's copy shares #30047: it sleeps every copy, after the existing "also sleeps the copy on " confirmation. It stops the chats on every host that has a copy.

Marking the slept workspace as asleep and hiding it under "Hide sleeping workspaces" is the stacked follow-up, #28457.

What Changed

• The host stops the chats. Sleep asks each execution host to stop the workspace's chats, through a new agentSession.sleepWorkspace call. The execution host is the machine running the agents: this desktop, a paired server, or an SSH machine reached through a paired server. The host finds every chat it runs in that workspace, including hidden chats and starts still in progress. It closes them through the existing ordered close path, which ends the process's input, then signals it, then forces an exit if needed. It answers with the chats it can prove are stopped, plus a typed failure for each of the rest.

• What counts as stopped. A chat counts as stopped only with proof:

  • the host stopped the running process itself;
  • it released an acquisition it owned and saw the exit;
  • the chat's claim is already released, with recorded exit evidence;
  • or the chat never started.

  A release without proof stays unverified, and that's what produces the one toast.

• Only real newer work overrules a stop. A stop result counts as superseded only on facts: the chat's session restarted, or newer work exists (a running child, a delivery, a pending submission, a create or attach in flight, or an open dispatch). The provider's own "I've ended" report doesn't count. A message you send after the stop still runs and resumes the chat.

• Sleep acts on the workspace, as main does. The sidebar flow is main's id-based flow from [hardening 4/5] fix(sidebar): confirm before sleeping a workspace that another host's copy shares #30047. It adds one step: when terminal teardown is admitted, it asks every host that publishes the workspace to stop its chats, from main's own list of a workspace's copies. If main's terminal-owner check refuses (it can't tell which copy runs the terminals), no chat is stopped. A workspace's VM is suspended only after its chats have stopped.

• A failed Claude start is retired after a proven exit. If a Claude chat's start failed and Sleep later proves its process gone, the failed attempt is dropped, so your next message starts a fresh agent instead of being refused.

• Status rows have one writer. Chat status rows stay owned by the host's status feed. Sleep no longer deletes them from the app window, so there's no race between the two writers. A chat makes a workspace sleepable only while the host reports that it holds the chat's process.

• Project moves. The shared Sleep routine is also used when a project moves into WSL (feat(wsl-mover): pause a moving project's terminals through Sleep and prove they exited (D1b-5d) #28820), so that move now stops the project's chats too.

Why

The host is the only side that knows which agent processes it owns. The visible tab list misses hidden chats and starts still in progress. Reusing the host's ownership and its existing close path avoids a second way of stopping processes. Taking main's whole-workspace Sleep model keeps one meaning of what Sleep acts on. An earlier version of this PR slept one host's copy at a time; that contradicted main's #30047 decision and would have broken its confirmation flow.

Mixed versions. The new call is checked against the host's advertised capabilities. An older host that lacks it returns an ordinary failure, with the one toast, and is fixed by updating the host. The failures field in the answer is optional, so older peers read the answer unchanged. Nothing else on the wire changes.

Differences from the common pattern:

• Intended: stopping is decided by the host's per-chat proof rather than a fire-and-forget close. Orca's Sleep can partly fail across several hosts, and A slept native-chat workspace reads asleep and hides #28457 marks only chats that were proven stopped.

Known limitation: main's terminal-owner check also refuses a workspace open on two hosts that has no terminal bound in this window, so such a workspace (including one with only chats) still can't be slept. Main shows "can't tell which host runs this workspace's terminals". A chat-only workspace on a single host can be slept.

Linked Issue

Reported in the workspace Sleep thread ("Sleeping a workspace with a native chat should really sleep it ... should be properly marked as slept and hidden"). The marking and hiding half is #28457.

Visual Proof

The same input on each build: a workspace with only a Claude chat (ws-chat), and one with a terminal plus a Claude chat (ws-mixed). Both chats were idle after replying.

Before, on main: a chat-only workspace can't be slept (Sleep is greyed out).

Before, on main: after Sleep, ws-mixed shows Inactive and drops its chat row, but that chat's claude is still running.

After: Sleep is offered for the chat-only workspace.

After Sleep on ws-chat: its claude exited, there's no toast, and the workspace is still listed.

Reopening keeps the history, and a new message resumes claude and gets a reply.

After Sleep on ws-mixed: the chat's claude and the terminal's processes all stopped, with no toast.

Testing

Final head (5b829e9), live QA: a hidden test copy of Orca on the M4Air, with the real signed-in Claude and Codex driving the UI. Main (70d00aa) got the same input as the comparison.

• Main: Sleep is disabled for the chat-only workspace. Sleeping the mixed workspace stops its terminal, but its claude keeps running until the app quits.
• Chat-only Sleep: claude exits 0.46 s after Sleep, and the host lists the chat as stopped. No toast, and the workspace stays listed. Reopening shows the history, and a new message resumes claude and gets a reply.
• Mixed Sleep: the chat and the terminal's processes all stop within 0.54 s. No toast.

Earlier rounds also covered:

• a control chat in another workspace surviving;
• a forced stop failure: exactly one toast, the workspace isn't counted as asleep, and retry works;
• the same over SSH;
• an already-stopped chat after an app restart, which sleeps without a toast.

Automated:

• The full PR workflow is green on 5b829e9.

• Focused tests cover: the host's Sleep answer, Sleep across the chat's lifetime, the RPC surface, the sidebar Sleep flow, main's twin-Sleep tests, and the failed-start retirement.

• Each regression from review has a test that failed before its fix: a Sleep stopping a newer turn, a second Sleep after a failed start, and the next send after a retired start.

• Final build manually verified through the changed native-chat flow

• Existing automated contracts exercised and targeted changes covered

Not verified:

• native phone apps;
• a Windows desktop run;
• Sleep of a workspace open on two hosts in a live rig (covered by unit tests only).

Seen but not caused by this PR:

• About 1.5 s after Sleep, an SSH workspace reads Active again; main does this too.
• Three real-Claude CLI tests fail on model names, because the installed CLI now reports newer IDs. Those files are unchanged from main.

Review

This went through eleven review rounds with independent reviewers, plus design panels on the hard calls. Defects found and fixed here:

• an already-stopped chat failed Sleep;
• the status row raced between the host feed and the app window;
• the first Sleep of a running chat answered "stopped nothing";
• after main's [hardening 4/5] fix(sidebar): confirm before sleeping a workspace that another host's copy shares #30047, per-copy Sleep broke main's confirmation flow, so the PR was rebuilt onto main's whole-workspace model;
• in the rebuild, a Sleep could stop a turn sent after waking, and a VM could be suspended before its chats stopped;
• a failed Claude start could lose its exit proof, or be reused by the next send.

The branch was synced with main at the end (main at 70d00aa) and has no conflicts. The pre-commit formatter also reformatted 13 of main's files that arrived unformatted in an earlier merge (formatting only).

Agent skill upstream boundary

• This change copies no upstream skill-installer source or documentation.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

• Last reviewed commit: 797d7a9
• Latest commit, not reviewed: 9439a9e

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 5cf0a255-ca6e-42c7-a176-a505585d25a8

📥 Commits

Reviewing files that changed from the base of the PR and between a9aef85 and 797d7a9.

📒 Files selected for processing: 6

• src/main/runtime/rpc/methods/structured-agent-session-gate-classification.test-fixture.ts
• src/main/runtime/rpc/methods/structured-agent-session.test.ts
• src/main/runtime/rpc/methods/structured-agent-session.ts
• tests/e2e/cross-version-wire/structured-agent-session-host-fixture.ts
• tests/e2e/cross-version-wire/structured-agent-session-surface-calls.ts
• tests/e2e/cross-version-wire/structured-agent-session-surface-manifest.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

📝 Walkthrough

Workspace sleep now carries execution-host identity through runtime and renderer paths. The structured-session host closes matching sessions and reports stopped IDs after checking stop receipts and remaining work. The renderer coordinates host-scoped session stops and panel teardown, and terminal connections track sleep intents and teardown by host. The change also updates POSIX process-tree cleanup and Codex managed-process closure.

Priority: ➖ Normal

Merge Risk: 🔵 Low · up to 797d7

Sleep can show a plural failure message for one workspace. An adapter that omits the optional ownership check may also miss a release attempt; confirm the affected adapter paths before merging.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 23.73% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 59 functions across 65 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
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.
Title check		The title clearly and concisely describes the primary change: workspace Sleep now stops native chat agents.
Description check		The description is detailed and covers the user impact, implementation, rationale, visual proof, testing, limitations, and review history. The Linked Issue section does not provide a direct issue link…

Full details: Description check

Explanation

The description is detailed and covers the user impact, implementation, rationale, visual proof, testing, limitations, and review history. The Linked Issue section does not provide a direct issue link or completed Fixes # reference, and several checklist items remain unchecked, but the description is otherwise substantially complete.

✨ Finishing Touches: 💡 1

📝 Generate docstrings: 💡

• Commit to this branch
• Create a new PR

• 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.}

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 66d46363-cb90-45fc-8db5-71e2733ee1b3

📥 Commits

Reviewing files that changed from the base of the PR and between 099afc8 and 224d130.

⛔ Files ignored due to path filters: 1

• src/shared/rpc-contract/rpc-params-catalog.generated.ts is excluded by !**/*.generated.*

📒 Files selected for processing: 15

• src/main/runtime/rpc/methods/structured-agent-session-gate-classification.test-fixture.ts
• src/main/runtime/rpc/methods/structured-agent-session-pi-dialog-capability.test.ts
• src/main/runtime/rpc/methods/structured-agent-session.test.ts
• src/main/runtime/rpc/methods/structured-agent-session.ts
• src/renderer/src/components/sidebar/sleep-workspace-chats.ts
• src/renderer/src/components/sidebar/sleep-worktree-activation-race.test.ts
• src/renderer/src/components/sidebar/sleep-worktree-flow.test.ts
• src/renderer/src/components/sidebar/sleep-worktree-flow.ts
• src/renderer/src/components/sidebar/use-worktree-context-menu-model.tsx
• src/renderer/src/components/sidebar/workspace-lineage-menu-actions.test.ts
• src/renderer/src/components/sidebar/workspace-lineage-menu-actions.ts
• src/renderer/src/components/sidebar/worktree-context-menu-policy.ts
• src/renderer/src/i18n/locales/en.json
• src/renderer/src/runtime/structured-agent-session-close.ts
• src/shared/structured-agent-session-surface-capabilities.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

[coderabbitai]

Actionable comments posted: 2

---

ℹ️ Review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: fff368ec-bc04-4d95-b920-9cc76924c31e

📥 Commits

Reviewing files that changed from the base of the PR and between 224d130 and 4f5bd1f.

⛔ Files ignored due to path filters: 1

• src/shared/rpc-contract/rpc-params-catalog.generated.ts is excluded by !**/*.generated.*

📒 Files selected for processing: 65

• mobile/src/host-screen/host-screen-overlays.tsx
• src/main/acp/acp-structured-session-adapter.ts
• src/main/claude/claude-structured-session-acquisition.ts
• src/main/claude/claude-structured-session-adapter.ts
• src/main/codex/codex-app-server-connection.ts
• src/main/codex/codex-structured-session-adapter.ts
• src/main/ipc/dashboard-payload-validation.ts
• src/main/ipc/ephemeral-vm-runtime-handler-cleanup.test.ts
• src/main/ipc/ephemeral-vm-runtime-handlers.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-adapter-router.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-adapter-router.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-adapter.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-attach-orchestration.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-conversation-lifetime.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-create.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host-delivery.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host-mutation-wiring.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-surface-lifetime.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-task-queue.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-workspace-sleep.ts
• src/main/pi/rpc-session-adapter.ts
• src/main/provider-process/provider-process-teardown.test.ts
• src/main/provider-process/provider-process-teardown.ts
• src/main/runtime/agent-session-record-founding.ts
• src/main/runtime/orca-runtime-list-managed-worktrees.ts
• src/main/runtime/rpc/methods/structured-agent-session-rpc.test-fixture.ts
• src/main/runtime/rpc/methods/structured-agent-session-schemas.ts
• src/main/runtime/rpc/methods/structured-agent-session.test.ts
• src/main/runtime/rpc/methods/structured-agent-session.ts
• src/main/runtime/rpc/methods/worktree.ts
• src/main/runtime/runtime-notifier-contract.ts
• src/main/runtime/structured-worker-authority.ts
• src/main/window/runtime-window-lifecycle.ts
• src/preload/api/ephemeral-vm-api.ts
• src/preload/api/ui-bridge-clipboard-and-window-controls.ts
• src/preload/api/ui-command-event-api.ts
• src/renderer/src/components/dashboard/useDashboardPopoutBridge.ts
• src/renderer/src/components/sidebar/sleep-worktree-activation-race.test.ts
• src/renderer/src/components/sidebar/sleep-worktree-flow.test.ts
• src/renderer/src/components/sidebar/sleep-worktree-flow.ts
• src/renderer/src/components/sidebar/sleep-worktree-sidebar-position.ts
• src/renderer/src/components/sidebar/use-worktree-context-menu-commands.ts
• src/renderer/src/components/sidebar/use-worktree-context-menu-model.tsx
• src/renderer/src/components/sidebar/workspace-lineage-menu-actions.test.ts
• src/renderer/src/components/sidebar/workspace-lineage-menu-actions.ts
• src/renderer/src/components/terminal-pane/pty-connection-deliberate-sleep-guard.test.ts
• src/renderer/src/components/terminal-pane/pty-connection/run-deferred-connect.ts
• src/renderer/src/hooks/ipc-events/mobile-terminal-close-ipc-bridge.ts
• src/renderer/src/i18n/locales/en.json
• src/renderer/src/lib/ephemeral-vm-runtime-cleanup.ts
• src/renderer/src/lib/sidebar-worktree-activation.test.ts
• src/renderer/src/lib/sidebar-worktree-activation.ts
• src/renderer/src/lib/worktree-sleep-intent.ts
• src/renderer/src/runtime/structured-agent-session-client.ts
• src/renderer/src/store/slices/worktrees/session/set-active-worktree.ts
• src/renderer/src/store/terminals/terminal-pty-bindings.ts
• src/renderer/src/store/terminals/terminal-pty-release.ts
• src/renderer/src/store/terminals/terminal-tab-creation.ts
• src/renderer/src/store/terminals/terminal-worktree-sleep-owner.test.ts
• src/renderer/src/store/terminals/terminal-worktree-sleep-owner.ts
• src/shared/dashboard-snapshot.ts
• src/shared/ephemeral-vm-runtime-host-match.ts
• src/shared/rpc-contract/structured-agent-session-params.ts
• src/shared/rpc-contract/worktree-params.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 41ddb650-64f5-446d-9f92-0b61548d700b

📥 Commits

Reviewing files that changed from the base of the PR and between 4f5bd1f and 9d4018c.

⛔ Files ignored due to path filters: 1

• src/shared/rpc-contract/rpc-params-catalog.generated.ts is excluded by !**/*.generated.*

📒 Files selected for processing: 10

• mobile/src/host-screen/host-server-workspaces-wiring.test.ts
• mobile/src/host-screen/use-host-worktree-actions.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-create.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host-delivery.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host.ts
• src/main/runtime/agent-session-record-founding.ts
• src/main/runtime/rpc/methods/worktree-sleep-method.ts
• src/main/runtime/rpc/methods/worktree.ts
• src/renderer/src/i18n/locales/en.json
• src/shared/rpc-contract/structured-agent-session-params.ts

🚧 Files skipped from review as they are similar to previous changes: 1

• src/renderer/src/i18n/locales/en.json

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 3 remain after this review.

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to GitHub limitations.

⚠️ Outside diff range comments (1)

🟡 Minor · Preserve active teardown counts when forgetting a worktree. · worktree-sleep-intent.ts:87-114

src/renderer/src/lib/worktree-sleep-intent.ts:87-114
🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Preserve active teardown counts when forgetting a worktree.

withWorktreeSleepTeardown keeps a finalizer pending during terminal shutdown. If deletion forgets the worktree, then the same repo::path is recreated and starts another teardown, the old finalizer decrements the recreated worktree's count and removes it. A later PTY bind can then clear the new sleep intent while terminal shutdown is still active.

Suggested fix

     intents.delete(key)
-    teardownCounts.delete(key)
     listeners.delete(key)

---

ℹ️ Review info

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 1631169b-b705-4add-866f-dd4885f7d62d

📥 Commits

Reviewing files that changed from the base of the PR and between 9d4018c and a9aef85.

📒 Files selected for processing: 2

• src/renderer/src/lib/worktree-sleep-intent.ts
• src/renderer/src/store/slices/workspace-activation-preparation.test.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 6 remain after this review.

[brennanb2025]

Status (head 9439a9e4f6f524ebf98d704693fc7585c922339c): review done, live QA passed, full PR workflow green.

This head is 5b829e9 plus one merge of main. The merge's only conflict was in the session host's setup, where one field main removed was dropped to match. Live QA ran on 5b829e9; the full PR workflow passed on this head, after one rerun of a job that failed uploading its build artifact.

Why the PR changed since round 9: main merged #30047, which makes Sleep act on the whole workspace. When the same workspace is open on two hosts, Sleep on either copy sleeps both, after a confirmation. This PR had made Sleep act on one host's copy, which broke that confirmation flow. It was re-planned and rebuilt onto main's model. Sleep now asks every host that has a copy to stop its chats. Main's terminal-owner check, confirmation and teardown order stay as they are on main. The rebuild removed about 300 production lines of per-host handling.

Rounds 10–11 reviewed the rebuild, and every finding is fixed and re-verified:

• an old Sleep could stop a reply sent after waking the workspace. Chat stops are now admitted when terminal teardown is admitted, and a later send still resumes the chat;
• a VM was suspended before its chats stopped. It now suspends after they have;
• cleanup after a failed Claude start could lose its exit proof, so a second Sleep reported a failure;
• after a proven exit, a failed Claude start stayed recorded, so the next message was refused. It is now retired, about 4 lines.

Three-model design panels settled the ordering and retirement calls, and independent host and flow reviewers agreed on each fix. The loop's fixer finished with CI green. The wrap-up (a final main sync and QA) was run by the lane coordinator after the loop session went idle.

Live QA on this head: a hidden test copy on the M4Air, real signed-in Claude, Codex driving the UI, with main 70d00aa as the comparison on the same input.

• Main: Sleep is greyed out for a chat-only workspace, and a mixed Sleep leaves the chat's claude running.
• This PR:
  • chat-only Sleep: claude exits 0.46 s later and the host lists the chat as stopped. There's no toast, and reopening shows the history and gets a reply;
  • mixed Sleep: the chat and the terminal's processes all stop within 0.54 s.

Screenshots are in the PR description.

The QA rig left out --password-store=basic --use-mock-keychain, so the test copies may have used the M4Air's real keychain. The real settings.json and hooks.json were unchanged.

CI on this head: the full PR workflow passes. An earlier floating-terminal test failure came from main and was fixed there (#30568).

Pre-existing, not from this PR:

• three real-Claude CLI tests fail on model names, because the installed CLI now reports newer IDs (those files are unchanged from main);
• an SSH workspace reads Active about 1.5 s after Sleep.

Disclosed limitation: a workspace open on two hosts with no terminal bound in this window keeps main's "can't tell which host" refusal. That includes one with only chats.