fix(native-chat): a chat cut off by a crash, quit or restart reads Interrupted until its state changes

[brennanb2025]

	Files	Added	Deleted	Net
Test	44	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​1124	$\color{#cf222e}{\Huge{\mathbf{−}}}$​230	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​894
Prod	38	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​449	$\color{#cf222e}{\Huge{\mathbf{−}}}$​265	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​184

ELI5

When Orca crashes, quits, restarts, or evicts a chat while the agent is mid-answer, that answer is cut off. Orca marked it with a red "Failed" in the sidebar and everywhere else, but nothing failed: the chat was interrupted. Now it reads "Interrupted" everywhere, and it keeps saying so until the chat actually does something else. If the agent itself crashes, that is still a red "Failed".

What Changed

The problem: a turn cut off by an Orca crash, quit, restart, or eviction (Orca proved the agent process gone and nobody pressed Stop) showed a red Failed dot and "Failed" line on the sidebar card and agent row, the tab bar, Activity, the notes send menu, and the phone's agent list. The chat's own turn bar said "Worked for N", which disagreed with all of them.

The notice under the cut turn also blamed the wrong party: after an Orca quit or restart it read "Codex stopped while this response was in progress. You can continue in this conversation." in error red, though nothing shows the agent did anything wrong.

Separately, on desktop a native chat's settled status (a Stop's Interrupted, a Failed, a "Couldn't confirm") vanished from the worktree card and tab bar about 30 minutes after the chat's last activity, and right away after a restart for an older chat, while the agent row under the card and the phone kept showing it.

What you see now: a cut turn shows the existing Interrupted state on every status surface (the muted dot and the word "Interrupted", the same one a Stop uses, without "by user"). The turn bar reads "Interrupted after N", and the row under the turn reads "This response was interrupted. You can continue in this conversation." as a muted line, translated into the app's language. That line stays visible when the settled turn is collapsed, under the turn's partial answer.

The agent's own crash is unchanged in substance: when the agent process exits on its own while Orca runs, its row still reads " stopped while this response was in progress…" in red, and the status surfaces and the turn bar read red Failed / "Failed after N" to match it. A person's Stop, a real failure and "Couldn't confirm" read as before. A cut sends no notification, as before.

How long it lasts: a native chat's settled status is read from its latest turn, so it stays until the chat's state changes, for example when a new message starts a turn and it reads Working, or until the chat is closed. That includes after Orca restarts: a chat whose last turn was stopped, failed, cut off or could not be confirmed shows that on its tab and worktree card however long ago it happened. Opening or viewing the chat changes nothing, and no timer clears it. A clean "done" still fades from the card after 30 minutes as before. Only the latest turn's mark is shown: once a newer turn runs, an earlier turn's Failed, Stop or cut lives only in the transcript, under that turn.

On the worktree card, which sums up every chat and terminal agent in a workspace, a native chat's mark now ranks the same however old it is. Live work in another chat or terminal always shows over it. Another agent's fresh finish shows over an Interrupted or Couldn't confirm, but not over a Failed, which stays until that chat starts a new turn or is closed. Alone, the mark shows. Before, a native chat's mark ranked higher for its first 30 minutes and then dropped, so the card could change with nothing new happening. For example, chat A failed 10 minutes ago and chat B is working: the card read Failed, then switched to Working once A's failure turned 30 minutes old. Now it reads Working from the start, while chat A's tab and agent row read Failed. Likewise, a chat that was just stopped no longer hides another agent's fresh Done on the card.

Mechanism, the mark: agentVerdictDisplayMark (desktop) and its phone mirror agentRowVerdictMark map the host's interruption verdict to interrupted, as a Stop or a replaced turn, instead of failed. The turn bar's settledTurnStatusKey maps it to interruptedAfter. Activity titles follow the same grouping.

Mechanism, the agent's own crash: the journal records the same "interrupted, no verdict" turn for the agent's own exit as for a quit, so the cause is read from the row that explains the cut. One shared reader (structuredAgentTurnVerdictReader in native-chat-cut-turn-explanation.ts) gives a cut turn that the agent's own exit row explains the verdict failure. The host's status projection and the chat's turn bar both use it, so the published row, older clients and the turn bar all read Failed. Nothing is stored, so no turn-completion event (and no notification) is sent for it, as before. If the exit's own bookkeeping failed and a later reopen wrote the row instead, the cause cannot be told apart from a quit, and the turn reads Interrupted.

Mechanism, lifetime: the worktree card and tab bar ignore an agent row once its evidence is 30 minutes old, because a hook stream that goes quiet proves nothing. A native chat's row is not a live report; its host re-derives it from the chat's journal whenever that changes. The status bridge now carries the host's own marker for a native chat's row (structuredHost, which the host already stamps on that row), and isSettledNativeChatVerdict keeps a done row with a settled verdict from such a row on those two surfaces past the window. On the worktree card such a mark, at any age, joins the tier a departed agent's mark already uses (applyRetainedAgentMark in worktree-agent-activity-summary.ts): live work always ranks above it, a fresh finish ranks above an Interrupted or Couldn't confirm but not above a Failed (the order the card already uses for a departed agent's failure), and alone it still shows. It no longer moves between tiers when its evidence turns 30 minutes old. Hook and terminal rows, a clean done, and live work decay exactly as before, and the shared freshness check used for subagent rows, attention sorting and cleanup is untouched.

Mechanism, notice: the chat cannot tell a quit, an eviction or a restart apart, because the host publishes no cause for them, so the notice uses one wording true for every cause. It is a host status line (response-interrupted in agent-session-host-status-rows.ts), like the existing "Part of this chat's history couldn't be loaded" line: desktop words it in the reader's language, and its notice tone keeps it muted on a client that does not know the line. The collapsed-turn fold never hides it, as it never hides a compaction report. The host writes its reopen row (left when a reopen finds the previous agent process gone) with the same words and presentation, stored red for older clients, and an older host's red reopen row is shown in the new words; any other reopen row, such as one a newer host words itself, is shown as written.

Replaces #25293, which cleared the red mark once the chat was viewed; that approach was rejected because viewing a chat does not change whether it was interrupted. Builds on #25043, which introduced the one derived notice row for a cut turn.

Why

The status should describe the chat's state. An Orca crash, quit or restart interrupts a turn; it is not the agent failing, while the agent's own crash is. The turn bar and the sidebar should agree, and the mark should hold for exactly as long as the chat stays in that state, whoever ended the turn.

Alternatives considered: keeping red Failed until the chat is viewed (#25293) was rejected because it makes a status depend on whether you looked. Recording a failure outcome in the journal for the agent's own exit was rejected because a recorded outcome sends a turn-completion event, which would add a notification that is not sent today. Exempting native-chat rows inside the shared freshness check (isExplicitAgentStatusFresh) was rejected because that check also decides whether subagent rows under a parent may claim live work; a dead parent's leftover children would have read Working forever. Exempting only the cut from the 30-minute window was rejected because a Stop shows the same Interrupted mark, and the two would have vanished at different times. Showing old marks only in the transcript was rejected because the latest turn's failure is the chat's current state, and the chat's row shows it. On the card, keeping a fresh native-chat Failed above live work (main's order for a fresh failure) was rejected: with no expiry it would pin the card red over live work for as long as the chat sits failed.

Differences from the common pattern

• Intended: the common pattern shows a turn cut by a restart as a plain finished turn ("Worked for N") and marks the thread only while it is unread (a red mark, or Done, that clears once opened). Orca shows it as Interrupted, on the turn bar and every status surface, until the chat's state changes, and viewing it changes nothing. Reason, in the product owner's words: an interrupted chat stays interrupted until its state changes; viewing it changes nothing.
• Intended: a native chat's Failed stays on its row until the chat's state changes, also after the chat is opened; it is never treated as an unread alert. Reason: a chat's status is its current state, not something you have or haven't seen.
• Intended: on the worktree card, which sums up several chats and agents, a native chat's Failed ranks below live work in another chat, not above it. Reason: the mark has no expiry, so ranking it above live work would hide that work for as long as the chat sits failed.
• Matches the common pattern: the status is derived from the chat's latest turn on every read; once a newer turn runs, an earlier turn's failure, Stop or cut appears only in the transcript; and nothing clears or re-ranks a settled status as time passes. Before this change the 30-minute window cleared it on two desktop surfaces, and the card re-ranked a native chat's mark at 30 minutes. The agent's own crash reads as a failure, a cut sends no notification, and the row explaining the cut is never hidden by the collapsed turn.

Linked Issue

Fixes #

Visual Proof

Live QA on macOS: an isolated dev build with stand-in agents, driven over the Chrome DevTools Protocol by a separate agent. Waits past 30 minutes ran on real time; the clock was not faked.

After main's Continue feature was merged in (round 3, all scenarios pass):

Crash mid-reply, then relaunch. The turn reads "Interrupted after 31s", with the cause line and Continue. The card and tab dots are grey.

After pressing Continue. The agent carries on and finishes.

Graceful quit mid-reply, then relaunch. The quit wording of the cause line, with Continue.

32 minutes later, after viewing the chat. Still Interrupted.

Neighbouring states and the workspace card (round 2, all nine scenarios pass):

Stop pressed (unchanged). "Interrupted by user", with no notice line.

The agent process dies while Orca runs (unchanged). Red Failed.

One chat failed, another working, in one workspace. The card shows Working; the failed chat's tab stays red.

After the working chat finishes. The card shows the failure, and doesn't change on its own.

Noticed during QA, not changed by this PR: no resume offer after a crash, only after a graceful quit; after a quit, the resume dialog, Continue and the "1 chat to resume" pill are all offered at once; Continue adds a message bubble styled like the user's own.

Testing

• I manually tested these changes locally
• Automated tests added/updated, or explained why not below

New tests: a Stop, a cut, a failure and "Couldn't confirm" on a native chat stay on the worktree card and tab bar more than 30 minutes after the chat's last activity, while a clean done, a terminal hook row and a stale working row still fade, and the subagent liveness check still reads the row stale; on the card, another agent's live work reads Working over a native chat's Failed and a fresh Done reads Done over its Stop, both for a mark minutes old and one past 30 minutes, and a cut alone reads Interrupted; the status bridge carries the host's marker on its rows; a cut that the agent's own exit explains reads failed on the status row and "Failed after N" on the turn bar and sends no completion event, while a quit, a reopen and an exit row answering a later message keep it Interrupted; a collapsed cut turn shows its partial answer and the notice (derived, the host's reopen row, and an older host's re-worded red row), and the notice alone for a turn with no answer; a reopen row with its own presentation, and an early build's untoned row, are kept as written; Activity titles a cut "Agent interrupted"; the phone's row reader shows Interrupted for a cut (also checked through the cross-version phone test, which loads the phone's real row reader).

Updated tests, each pinning the old rule (red Failed or "Worked for N" for a cut): the shared verdict table, the turn-bar labels (desktop and phone), the tab bar, the worktree card summary and agent rows, Activity grouping, the dashboard row, the paired-client mirror, and the host tests for a close, quit, restart and crash (structured-agent-session-close-verdict, -crash-turn-end, -recovered-turn-clock, -stop-event-restart).

Each rule was ablated (the change reverted with the tests kept) and its tests went red: the verdict mark, the turn-bar label, the agent-exit failure reading and its precedence over a later reopen row, the settled-verdict lifetime and its done-only guard, the retained tier for old marks on the card, the bridge's host marker, the never-folded notice, the phone mirror, the Activity title, the notice words, the host's reopen row, the re-wording of only legacy stored rows, and the notice tone.

AI Disclosure

Review

Agent skill upstream boundary

• Not applicable, or this change follows docs/reference/agent-skill-sharing-upstream-boundary.md and copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.

Notes

Remote and SSH: the execution host still owns the verdict and publishes it on the same row as before. There is no wire change: failure and interruption are existing verdicts, and the row marker is the host's existing structuredHost field. Losing contact with a remote host keeps the last settled status, as for every settled row.

Mixed versions: a phone or desktop client older than this change still draws a cut chat as Failed, because the drawing happens on the client. A newer client paired with an older host draws a cut as Interrupted; for an agent's own exit it shows "Failed after N" on the turn bar but Interrupted on the sidebar, until the host updates. An older client reading a newer host shows the new reopen row in red with its neutral English words, because the host stores it red so a client that folds every other row under a collapsed turn still shows it; a newer client draws that row muted, and shows an older host's red reopen row in the new words.

Unchanged on purpose: attention sorting still ranks a cut like a completion (news you have not seen), unlike a Stop, which it demotes. Cmd+J shows no status for native chats, before or after this change.

Checklist

• This PR is small and focused
• I explained what changed and why (ELI5, the user-facing before/after, the mechanism, and why over the alternatives)
• Before/after screenshots or videos attached for UI changes, or N/A with reason
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred)

[brennanb2025]

Review status

The problem. When Orca crashes, quits, restarts or evicts a chat while the agent is mid-answer, the answer is cut off. On main the sidebar card, agent row, tab bar, Activity, the notes send menu and the phone's agent list all marked that chat red Failed, while the chat's own turn bar said "Worked for N". On the worktree card and tab bar the red mark also vanished about 30 minutes later, though nothing about the chat had changed. The line under the cut turn said "Codex stopped while this response was in progress…" in error red, blaming the agent for something Orca did.

What changes for you.

• A cut chat reads Interrupted everywhere (the muted dot and word a Stop already uses), and the turn bar reads "Interrupted after N".
• One muted line sits under the cut turn: "This response was interrupted. You can continue in this conversation." It stays visible when the turn is collapsed.
• The mark stays until the chat's state changes: a new message (it reads Working, then the new result), or closing the chat. Opening the chat, time passing and restarting Orca change nothing.
• If the agent process itself crashes while Orca is running, the chat still reads red Failed with its red " stopped…" line, as on main.
• A person's Stop, a real provider failure and "Couldn't confirm" read as before. A cut sends no notification, as before.
• A native chat's other settled marks (a Stop's Interrupted, a Failed, "Couldn't confirm") also no longer vanish from the card and tab bar after 30 minutes. On the worktree card, live work in another chat always shows over an older mark.

Fixed during review

• An agent's own crash had become a calm Interrupted on every status surface, sitting above a red error line. It reads Failed again. The Failed comes from the chat's history at read time and is not saved, so no new notification fires.
• The muted line was folded away inside a collapsed turn, so by default nothing explained the cut. It now always shows.
• The same Interrupted mark had two lifetimes: a cut stayed while a Stop vanished at 30 minutes. Every settled native-chat mark now follows the same rule, using the host's existing marker for a native-chat row instead of a new field.
• An old kept Failed could pin the worktree card red over live work in another chat. Kept marks now sit in the existing lower tier for older marks: live work shows over them, and a fresh finish shows over an older Interrupted or "Couldn't confirm".
• Older clients would have folded away the host's new muted reopen row and shown no explanation. The host now stores that row red with the neutral words, so older clients keep it visible, and newer clients draw it muted.
• Only an older host's red reopen row is re-worded. A newer host's own wording (including the "Orca stopped" cause rows being added in feat(native-chat): say in the chat why Orca stopped a reply, and offer Continue #25675) is kept, along with every stored field.
• The failed-start row added on main in fix(native-chat): mark a chat whose start failed on its tab and workspace row #24904 now follows the same lifetime rule.
• A Cmd+J change that could never take effect was removed, and the PR body's claims about Cmd+J, notifications and departed agents were corrected.
• Carried main's one-line fix for a duplicate test key (resolveLaunchArgs) so CI can run; it matches what main will do.

Deferred

• When the agent exits on its own, a sub-millisecond gap separates the turn ending from its red line being written (measured median 0.26–0.52 ms, max 1.9 ms, over 40 runs). That is under one frame, so it isn't a visible flash. Removing it would mean changing how a Stop that meets an exit is recorded.
• The rule deciding a chat's status mark is duplicated across the desktop and the phone. One shared function would make the next change a one-line edit; it's a follow-up.
• Older stored rows from late September lose the provider detail they carried once re-worded. Small comment and translation-register nits.
• With a newer client and an older host, an agent's own crash shows "Failed after N" on the turn bar but Interrupted on the sidebar, until the host updates.

Open question

• A native chat's old Failed, Interrupted or Couldn't confirm now stays on its tab and the worktree card until that chat changes, however long ago it happened, and after restarts. On main these faded after 30 minutes. This applies the "status is the chat's state" rule beyond the cut case, so one mark doesn't have two lifetimes. An old Failed also stays on the card over a fresh finish in another chat, which is main's existing order for older failures. If only the cut should persist, that part comes out.

Differences from the common pattern

• Intended: the common pattern shows a cut turn as finished and marks it only until the chat is read. Here the chat reads Interrupted until its state changes, because an interrupted chat is still interrupted after you look at it.

Verified

Live on a hidden, isolated test build on a second Mac at c4c8446, driven over the app's debugging protocol, with a scripted stand-in agent (no real login):

• Crash (force-killed Orca mid-reply, then relaunched): every surface read Interrupted, the turn bar "Interrupted after 38s", and the muted line was visible with the turn collapsed.
• Viewing the chat changed nothing.
• 31 minutes of real time later (the clock was not moved), the card still read Interrupted.
• A new message went to Done; the line stayed only under the old turn.
• Quit behaved the same as the crash.
• Stop: "Interrupted after N" and "Interrupted by user", with no line, as on main.
• Killing only the agent's process: red Failed everywhere and the red "Claude stopped…" line.
• Provider error: red "Failed after 3s".
• Live work over a 49-minute-old Failed: the card read Working.

Crash, collapsed: muted line under the partial answer

Crash, sidebar: card and agent row read Interrupted

31 minutes later: card still Interrupted

New message: the new turn is Done, the old turn keeps its line

Quit, collapsed

Stop: unchanged

Agent process killed: red Failed

Provider error: red Failed

Live work over an old Failed: card reads Working

After that work finished: the old Failed shows again on the card

Automated: typechecks (node, web, mobile) and the named test set (12,453 desktop and 357 mobile tests) pass locally at c1962a0 (before the last main merge), and each fix was reverted with its tests kept to show they fail without it. CI at e7820df: typecheck, package, cross-version wire compatibility, Mobile Checks and test shards 3–5 pass. Shard 1 fails only on config/scripts/vitest-sqlite-runtime-boundary.test.ts:70. That failure is main's own: two test files from #25888 (orchestration-party-location.test.ts and runtime-orchestration-sender-names.test.ts) aren't listed in NODE_RUNTIME_INCLUDE, and main still has the gap. Shard 2 lost its runner (a shutdown signal) about 4.5 minutes in, on both attempts. This PR's tests in that shard are small, and three of them had finished before it died. The latest head, ee43985, only merges main, and its CI is running.

Not verified

• Live QA ran at c4c8446, before the last three merges of main. Those merges changed only conflict resolutions and test wiring, which were checked by tests, not live.
• The notes send menu (no entry point on that build), the phone in an emulator, a real SSH host and Windows/Linux were not exercised live; the phone was covered by unit tests through its real session hook.
• The in-progress frame of the new message was not captured, only its settled state.

[brennanb2025]

Live QA, round 2, at a56b6f1: all nine scenarios pass. Screenshots are in the PR description under Visual Proof.

Setup: a macOS dev build with an isolated profile and home folder and stand-in agents, driven over the Chrome DevTools Protocol by a separate agent. Waits past 30 minutes ran on real time.

	Scenario	Result
S1	Orca crashes mid-reply, then relaunches	Interrupted on the card, tab, Activity row and turn bar. The muted "This response was interrupted…" line shows with the turn collapsed.
S2	Open the chat	Still Interrupted.
S3	30+ minutes later	Card and tab still Interrupted.
S4	Send a new message	Working, then Done. The notice stays only under the old turn.
S5	Graceful quit mid-reply	Same as S1.
S6	Stop pressed	"Interrupted by user", with no notice line (unchanged).
S7	The agent process dies while Orca runs	Red Failed, with the red "Claude stopped…" row (unchanged).
S8	Provider error	Red "Failed after 3s" (unchanged).
S9	One chat failed, another working, same workspace	The card shows Working while the other chat runs, then the failure once it's Done. It doesn't change on its own over time.

CI at this head: 21 checks pass. The one red on the first run, package (windows), was a timeout in profile-state-writer-stall.electron.test.ts, which this PR doesn't touch; the same failure hit unrelated PRs that hour, and it passed on rerun.

Noticed during QA, not changed by this PR: no resume offer after a crash, only after a graceful quit; the "1 chat to resume" pill stays after Close; an "Enjoying Orca?" card can cover the composer's Send/Stop corner.

[brennanb2025]

Main merges

Round 3: merged main (2460068) at 09f9ea0.

The main change that overlaps this PR is #25675 (Orca says why it stopped a reply, and offers Continue). The two now combine like this:

• What the chat shows: when the row knows both the cause (quit, update or crash) and the machine's name, it reads "Orca on was closed while this response was in progress." and Continue is offered. Otherwise it shows this PR's neutral line: "This response was interrupted. You can continue in this conversation."
• What the host stores: the row about Orca's own stop is now stored in that neutral wording, with the cause beside it, instead of naming the agent ("Codex stopped ..."). It is still stored red, so older clients keep it on screen. With no known cause, it is the same row as any other cut turn.
• Row order fix: in NativeChatNoticeRow, this PR's neutral wording for these rows was being applied before feat(native-chat): say in the chat why Orca stopped a reply, and offer Continue #25675's cause wording, so the cause line never showed. A row with a cause and a machine name now gets the cause wording first.
• Conflicts: in structured-agent-session-dead-generation-settlement.ts I kept main's turn scope and cause, with the neutral words. In native-chat-transcript-slots.ts I kept this PR's check, which already covers main's rows.

Four restart tests failed after that merge: #25675's tests expected the stored quit row to match what the chat shows. With this PR the stored row is red and the chat shows it muted. cdcc646 updates those tests; production code is unchanged. CI on cdcc646 passed (21 passed, 16 skipped, 0 failed).

Round 4: merged main (e4cd14d) at 7fa999c.

• NativeChatNoticeRow.tsx: kept this PR's order (cause line and Continue when the cause and machine are known, otherwise the neutral line) and added main's Claude sign-in button (fix(claude): activate account profiles and remove credential replay (Step 4 of 4) #24434).
• Locale catalogs (en, es, fr, ja, ko, zh): merged key by key. Each now differs from main only by this PR's notices.responseInterrupted string, and main's new "Stopped manually" wording is kept.

These main commits in round 4 touch nearby code:

• fix(claude): activate account profiles and remove credential replay (Step 4 of 4) #24434: Claude sign-in button on chat notice rows.
• feat(native-chat): compact tool runs with a bounded call list #26285: compact tool runs; spacing between rows in the same turn.
• feat(native-chat): show each interrupted chat's workspace as its read-only sidebar card #25652: the restart dialog shows each interrupted chat's workspace card.
• Label a message stopped before the agent began on it plainly as "Stopped" #26381: a message stopped before the agent began now reads "Stopped manually".
• feat(native-chat): preview each rail tick with its reply, and land jumps at the top #26056: a preview of each reply on the transcript rail.

The earlier QA still holds: none of these change the cut-turn notice, the row Orca writes when it stops a reply, or Continue. The only one in the same file adds a separate sign-in button beside them.

Checks on 7fa999c: running; I'll update this comment when they finish.

[coderabbitai]

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: 1a8054fc-a269-44b5-b6ac-bd8a6cdaffc9

📥 Commits

Reviewing files that changed from the base of the PR and between e4cd14d and 7fa999c.

📒 Files selected for processing (82)

• mobile/src/session/MobileNativeChatTurnStatus.test.ts
• mobile/src/session/mobile-native-chat-merged-snapshot-parity.test.ts
• mobile/src/session/use-mobile-structured-agent-session.cut-turn-notice.test.tsx
• mobile/src/session/use-mobile-structured-agent-session.ts
• mobile/src/worktree/agent-row-display.test.ts
• mobile/src/worktree/agent-row-display.ts
• src/main/acp/acp-structured-host-lifecycle.test.ts
• src/main/acp/acp-structured-host-stop.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-attach-orchestration.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-close-verdict.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-conversation-open.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-crash-turn-end.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-dead-generation-settlement.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host-lifetime.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-orca-stop-row-readers.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-orca-stop-row.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-provider-child-record.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-recovered-turn-clock.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-restart-failure-filing.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-restart-interruption-test-harness.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-restart-offer-withdrawal.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-restart-ownership.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-shutdown-cut-row.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-stale-turn-open.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-stale-turn-verdict.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-stop-event-restart.test.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-wedged-profile-migration.test.ts
• src/main/runtime/claude-structured-exit-mid-response.test.ts
• src/renderer/src/components/activity/ActivityPrototypePage.thread-grouping.test.ts
• src/renderer/src/components/activity/activity-status-group-header.test.ts
• src/renderer/src/components/activity/activity-thread-grouping.status-order.test.ts
• src/renderer/src/components/activity/activity-thread-presentation.test.ts
• src/renderer/src/components/activity/activity-thread-presentation.ts
• src/renderer/src/components/dashboard/DashboardAgentRow.test.tsx
• src/renderer/src/components/native-chat/NativeChatMessageList.cut-notice-fold.test.tsx
• src/renderer/src/components/native-chat/NativeChatMessageList.interrupted-turn-fold.test.tsx
• src/renderer/src/components/native-chat/NativeChatNoticeRow.test.tsx
• src/renderer/src/components/native-chat/NativeChatNoticeRow.tsx
• src/renderer/src/components/native-chat/StructuredAgentSessionStatusBridge.test.tsx
• src/renderer/src/components/native-chat/StructuredAgentSessionStatusBridge.tsx
• src/renderer/src/components/native-chat/native-chat-transcript-slots.ts
• src/renderer/src/components/native-chat/structured-agent-session-status-bridge-failed-start.test.tsx
• src/renderer/src/components/native-chat/use-structured-agent-session.cut-turn-notice.test.tsx
• src/renderer/src/components/native-chat/use-structured-agent-session.ts
• src/renderer/src/components/sidebar/use-worktree-activity-status.ts
• src/renderer/src/components/sidebar/use-worktree-activity-statuses.ts
• src/renderer/src/components/sidebar/worktree-agent-activity-summary.test.ts
• src/renderer/src/components/sidebar/worktree-agent-activity-summary.ts
• src/renderer/src/components/sidebar/worktree-card-agent-summary.test.ts
• src/renderer/src/components/tab-bar/terminal-tab-activity-status.test.ts
• src/renderer/src/components/tab-bar/terminal-tab-activity-status.ts
• src/renderer/src/i18n/en-runtime-required.json
• src/renderer/src/i18n/locales/en.json
• src/renderer/src/i18n/locales/es.json
• src/renderer/src/i18n/locales/fr.json
• src/renderer/src/i18n/locales/ja.json
• src/renderer/src/i18n/locales/ko.json
• src/renderer/src/i18n/locales/zh.json
• src/renderer/src/lib/agent-verdict-status-line.ts
• src/renderer/src/lib/pane-agent-evidence.test.ts
• src/renderer/src/lib/pane-agent-evidence.ts
• src/renderer/src/lib/recent-workspace-tab-rows.test.ts
• src/renderer/src/lib/worktree-status.ts
• src/renderer/src/runtime/paired-agent-row-turn-outcome-arms.test.ts
• src/renderer/src/store/slices/agent-status-contract.ts
• src/renderer/src/store/slices/agent-status-live-entry-builder.ts
• src/shared/agent-main-agent-verdict.test.ts
• src/shared/agent-main-agent-verdict.ts
• src/shared/agent-session-host-status-rows.ts
• src/shared/agent-status-types.ts
• src/shared/native-chat-cut-turn-explanation.test.ts
• src/shared/native-chat-cut-turn-explanation.ts
• src/shared/native-chat-cut-turn-notice.test.ts
• src/shared/native-chat-cut-turn-notice.ts
• src/shared/native-chat-orca-stop-cut.test.ts
• src/shared/native-chat-turn-status.test.ts
• src/shared/native-chat-turn-status.ts
• src/shared/structured-agent-session-latest-request.ts
• src/shared/structured-agent-session-projection.ts
• src/shared/structured-agent-session-turn-timing.ts
• tests/e2e/cross-version-wire/cross-version-worktree-ps-verdict.unit.test.ts

💤 Files with no reviewable changes (2)

• src/renderer/src/lib/recent-workspace-tab-rows.test.ts
• src/main/native-chat/agent-session-wire/structured-agent-session-host-lifetime.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

Walkthrough

The change distinguishes interruption verdicts from failures across turn status, activity labels, and worktree summaries. Journal explanations identify agent exits that classify a cut turn as a failure. Reader-facing cut-turn notices use generic interruption wording, and host-status handling normalizes qualifying older rows. Settled native-chat verdicts remain visible beyond the freshness window when they have structured-host provenance.

Priority: ➖ Normal

Merge Risk: ⚪ Minimal · up to 7fa99

Native chats cut off by a crash, quit or restart now read as Interrupted, while agent failures remain Failed. No merge-blocking risk was found in the supplied changes.

🚥 Pre-merge checks | ✅ 4 | ❓ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Docstring Coverage	❓ Inconclusive	Docstring coverage is 37.84% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 37 functions across 50 files. (30 skipped…	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: crash, quit, and restart cuts now read as Interrupted until the chat state changes.
Description check	✅ Passed	The description is detailed, focused, and covers the user impact, implementation, rationale, testing, visual proof, compatibility, and checklist. The Linked Issue field remains incomplete as Fixes #…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Docstring Coverage

Explanation

Docstring coverage is 37.84% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 37 functions across 50 files. (30 skipped: 7 unsupported, 23 over the file limit.)

✨ Finishing Touches 💡 1

📝 Generate docstrings 💡

• Commit to this branch
• Create a new PR

• 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.}