Repository navigation
Conversation
|
Important Review skippedWe couldn't safely recover the incremental review. No full review was started, and the last reviewed checkpoint was preserved. Retry later, or explicitly request a full review by commenting You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 Walkthrough
Priority: ➖ Normal Merge Risk: 🟡 Moderate · up to
|
| Check name | Status | Explanation | Resolution |
|---|---|---|---|
| Description check |  |
The description explains the incident, root cause, implementation, tests, and known limitations. However, it does not follow the repository template and omits the required Linked Issue entry, Visual P… | Restructure the description using the repository template. Add the actual linked issue in the Linked Issue section, provide visual proof or write N/A with a reason, complete the Testing and checklist sections, and address the remaining requ… |
| Docstring Coverage | |
Docstring coverage is 13.33% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 15 functions across 7 files. | Write docstrings for the functions missing them to satisfy the coverage threshold. |
✅ Passed checks (3 passed)
-
Check name Status Explanation Title check The title clearly identifies the primary change: reseeding a remote workspace after its last session exits on the host. 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.
Full details: Description check
-
Explanation
The description explains the incident, root cause, implementation, tests, and known limitations. However, it does not follow the repository template and omits the required Linked Issue entry, Visual Proof or an explicit N/A explanation, checklist confirmations, and several required sections such as ELI5, What Changed, Why, and Testing.
Resolution
Restructure the description using the repository template. Add the actual linked issue in the Linked Issue section, provide visual proof or write N/A with a reason, complete the Testing and checklist sections, and address the remaining required sections or mark them as not applicable.
- Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Comment @coderabbitai help to get the list of available commands.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (1)
tests/e2e/paired-remote-terminal-session-exit-reopen.spec.ts (1)
237-248: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick winPoll for recovery instead of sleeping for 45 seconds.
Every passing run waits the full 45 seconds before checking the recovery conditions. Poll both counts with a bounded 60-second timeout so fast recoveries finish early and recoveries that take slightly longer can still pass. The test’s 300-second budget allows a longer poll; the fixed wait does not itself threaten that budget. The small observation dump has no separately established workflow impact.
Suggested fix
// The reopen: the user clicks the workspace in the sidebar again. await clickSidebarWorkspaceRow(client, worktreeId) - await client.page.waitForTimeout(45_000) + await Promise.all([ + expect + .poll(() => readHostTerminalCount(host, worktreeId), { + timeout: 60_000, + message: 'Reopened workspace has no live host surface' + }) + .toBeGreaterThan(0), + expect + .poll( + () => readClientWorkspaceState(client, worktreeId).then((state) => state.tabIds.length), + { timeout: 60_000, message: 'Reopened workspace renders no tabs' } + ) + .toBeGreaterThan(0) + ]) const observed = await readClientWorkspaceState(client, worktreeId)
ℹ️ Review info
⚙️ Run configuration
- Configuration used: Repository UI
- Review profile: CHILL
- Plan: Advanced
- Run ID:
2d258272-7470-4b78-abf6-b28f621c603e
📥 Commits
Reviewing files that changed from the base of the PR and between 9fdd90f and ddabb367f0308d403dfb4e8e825f2c354a99f573.
📒 Files selected for processing (5)
src/renderer/src/runtime/web-runtime-session-tab-lifecycle.tssrc/renderer/src/runtime/web-session-close-intent.tssrc/renderer/src/runtime/web-session-tabs-sync-emptied-workspace-tombstone.test.tssrc/renderer/src/runtime/web-session-tabs-sync/apply-preparation-base.tstests/e2e/paired-remote-terminal-session-exit-reopen.spec.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.
| // Why remember the emptying here: this is the one place that knows the | ||
| // close is the user's own. When the intents cover every terminal tab the | ||
| // row still holds, this close empties the workspace, and the mirror's | ||
| // affirming-empty frame may then leave its closed-last-terminal tombstone | ||
| // row. A session the host ends on its own never passes through here, so | ||
| // its emptying leaves no marker and the row is deleted instead. | ||
| const rowTabs = state.tabsByWorktree[args.worktreeId] ?? [] | ||
| const terminalTabsWithoutCloseIntent = rowTabs.filter((tab) => { | ||
| if (!isWebTerminalSurfaceTabId(tab.id)) { | ||
| return false | ||
| } | ||
| return !closeIntentTabIds.has(toHostSessionTabId(tab.id)) | ||
| }) | ||
| if (terminalTabsWithoutCloseIntent.length === 0) { | ||
| rememberUserEmptiedWorktree(args.worktreeId) | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Clear the user-empty marker when a close fails or the host refuses it.
The code writes the marker before the RPC call. Two paths can leave the terminal alive but keep the marker:
- The host refuses the close and republishes the snapshot (lines 165-172).
- The RPC call fails with an error other than
tab_not_found(the catch block).
In both paths, the code clears the close intents but leaves the marker. The marker clears only when a nonempty terminal row lands in prepareWebSessionTabsSnapshotBase. If the terminal later exits on the host before such a snapshot is applied, the code keeps the tombstone, and the workspace stays empty. Call forgetUserEmptiedWorktree(args.worktreeId) in the refusal branch and in the non-definitive branch of the catch block.
| export function rememberUserEmptiedWorktree(worktreeId: string): void { | ||
| const trimmed = worktreeId.trim() | ||
| if (trimmed) { | ||
| userEmptiedWorktreeIds.add(trimmed) | ||
| } | ||
| } | ||
|
|
||
| export function forgetUserEmptiedWorktree(worktreeId: string): void { | ||
| userEmptiedWorktreeIds.delete(worktreeId) | ||
| } | ||
|
|
||
| export function wasWorktreeEmptiedByUserClose(worktreeId: string): boolean { | ||
| return userEmptiedWorktreeIds.has(worktreeId) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
🔎 Supported by static analysis
🏁 Script executed:
rg -n 'rememberUserEmptiedWorktree|forgetUserEmptiedWorktree|wasWorktreeEmptiedByUserClose|worktreeId.*trim\(' src/renderer/src/runtime/web-session-close-intent.ts src/renderer/src/runtime/web-runtime-session-tab-lifecycle.ts src/renderer/src/runtime/web-session-tabs-sync/apply-preparation-base.tsRepository: stablyai/orca
Length of output: 1406
🏁 Script executed:
printf '%s\n' '--- intent ---'
nl -ba src/renderer/src/runtime/web-session-close-intent.ts | sed -n '1,220p'
printf '%s\n' '--- lifecycle ---'
nl -ba src/renderer/src/runtime/web-runtime-session-tab-lifecycle.ts | sed -n '1,180p'
printf '%s\n' '--- sync prep ---'
nl -ba src/renderer/src/runtime/web-session-tabs-sync/apply-preparation-base.ts | sed -n '145,215p'
printf '%s\n' '--- all relevant ID declarations and constructors ---'
rg -n -g '*.{ts,tsx}' 'worktreeId\s*[:=]|worktreeId\??:|worktree_id|worktree ID|worktreeId.*(trim|normalize)|trim\(\).*worktree' src/renderer/src src/shared src/main | head -250
printf '%s\n' '--- worktreeId references in changed area ---'
rg -n -g '*.{ts,tsx}' 'worktreeId' src/renderer/src/runtime | head -250Repository: stablyai/orca
Length of output: 41469
🏁 Script executed:
set -o pipefail
printf '%s\n' '--- relevant files ---'
rg --files src | rg 'worktree|workspace|runtime-types|session-tabs-sync' | head -250
printf '%s\n' '--- selector and ID validation symbols ---'
rg -n -g '*.{ts,tsx}' 'toRuntimeWorktreeSelector|is.*Worktree|validate.*worktree|normalize.*worktree|canonical.*worktree|worktree.*(trim|parse)|worktreeId.*(trim|normalize)' src | head -300
printf '%s\n' '--- worktree ID inputs and state construction ---'
rg -n -g '*.{ts,tsx}' 'tabsByWorktree|activeWorktreeId|worktreeId:' src/renderer/src src/shared | head -350
printf '%s\n' '--- selector definition and runtime type declarations ---'
rg -n -g '*.{ts,tsx}' 'export function toRuntimeWorktreeSelector|type Runtime.*Worktree|interface .*Worktree|worktreeId: string' src/renderer/src/runtime src/shared | head -250
printf '%s\n' '--- lifecycle callers ---'
rg -n -g '*.{ts,tsx}' 'closeWebRuntimeSessionTab\(|activateWebRuntimeSessionTab\(' src/renderer/src | head -250Repository: stablyai/orca
Length of output: 41459
🏁 Script executed:
printf '%s\n' '--- runtime selector ---'
nl -ba src/renderer/src/runtime/runtime-worktree-selector.ts
printf '%s\n' '--- snapshot apply entry and worktree ID source ---'
rg -n 'prepare.*Snapshot|apply.*Snapshot|worktreeId|snapshot' src/renderer/src/runtime/web-session-tabs-sync.ts src/renderer/src/runtime/web-session-tabs-sync src/renderer/src/runtime/web-runtime-session-snapshot.ts | head -160
printf '%s\n' '--- runtime worktree selector, catalog, and contracts ---'
rg -n 'RuntimeWorktree(List|Create)|worktreeId|id:' src/renderer/src/web/preload-api/web-runtime-worktree-catalog.ts src/shared/runtime-worktree-contracts.ts src/shared/worktree src/renderer/src/store/slices/worktrees-create*.ts src/renderer/src/store/slices/worktrees-fetch*.ts 2>/dev/null | head -240
printf '%s\n' '--- lifecycle call site context ---'
nl -ba src/renderer/src/runtime/browser-workspace-tab-close.ts | sed -n '70,115p'
nl -ba src/renderer/src/components/terminal/terminal-tab-actions.ts | sed -n '155,195p'
printf '%s\n' '--- representative worktree ID constructors ---'
rg -n -g '*.{ts,tsx}' 'id:\s*`(folder|worktree|repo)|worktreeId:\s*`|worktreeId\s*=\s*`|return\s+`(folder|worktree|repo)|\\btrim\\(\\)' src/renderer/src/store src/renderer/src/web src/shared/worktree src/shared/folder-workspace-types.ts | head -260
printf '%s\n' '--- worktree list API and ID schema candidates ---'
rg -n -g '*.{ts,tsx}' 'z\.string\(\).*worktree|worktree.*z\.string|worktreeId.*z\.|id: z\.string|id:\s*string' src/shared/runtime-worktree-contracts.ts src/shared/worktree src/preload/api src/renderer/src/web/preload-api | head -220Repository: stablyai/orca
Length of output: 41494
🏁 Script executed:
printf '%s\n' '--- shared worktree identity type and constructors ---'
nl -ba src/shared/worktree/types.ts | sed -n '45,95p'
rg -n -g '*.{ts,tsx}' 'WORKTREE_ID_SEPARATOR|::\$\{|make.*Worktree.*Id|worktree.*Id.*=.*path|id:\s*.*repo.*path|id:\s*.*worktreeId' src/shared/worktree src/renderer/src/store/slices src/main src/relay | head -240
printf '%s\n' '--- worktree identity producers ---'
rg -n -g '*.{ts,tsx}' 'id:\s*[^,]*::|WORKTREE_ID_SEPARATOR|worktreeIdFrom|toWorktreeId|createWorktree.*id|path.*worktree.*id' src | head -260
printf '%s\n' '--- snapshot result contract and producers ---'
rg -n 'RuntimeMobileSessionTabsResult|session\.tabs\.list|listSessionTabs|worktree:.*worktreeId|worktree: .*selector' src/shared src/renderer/src/runtime src/relay src/main | head -250
printf '%s\n' '--- workspace path input normalization ---'
rg -n -g '*.{ts,tsx}' 'realpath|resolve\(|normalize.*path|trim\(\)' src/shared/worktree src/renderer/src/store/slices/worktrees-create* src/renderer/src/store/slices/worktrees-fetch* src/relay/git-handler-worktree-list.ts src/relay/workspace-session-handler.ts | head -260Repository: stablyai/orca
Length of output: 41679
🏁 Script executed:
printf '%s\n' '--- canonical worktree ID implementation ---'
nl -ba src/shared/worktree/id.ts | sed -n '1,115p'
printf '%s\n' '--- runtime path normalization implementation ---'
rg -n 'function normalizeRuntimePathForComparison|export function normalizeRuntimePathForComparison' src/shared
printf '%s\n' '--- runtime snapshot contract ---'
rg -n -C 8 'type RuntimeMobileSessionTabsResult|interface RuntimeMobileSessionTabsResult|RuntimeMobileSessionTabsResult\s*=' src/shared
printf '%s\n' '--- worktree selector consumers ---'
rg -n -g '*.{ts,tsx}' 'RUNTIME_WORKTREE_ID_SELECTOR_PREFIX|startsWith\\(.id:|id:.*worktree|worktreeSelector|resolve.*Worktree.*Selector' src/renderer/src src/shared src/relay src/main | head -220
printf '%s\n' '--- concrete repo/worktree path assignment ---'
rg -n -g '*.{ts,tsx}' 'repo\.path\s*=|path:\s*.*realpath|realpathSync|realpath\(|worktreeIdForRepo|makeWorktreeId|worktreeIdFrom|createWorktreeId' src/main src/relay src/renderer/src/store src/shared/worktree | head -260
printf '%s\n' '--- repository path and ID validation entry points ---'
rg -n -g '*.{ts,tsx}' 'repoPath.*trim|path.*trim\(\)|z\.string\(\).*path|path: z\.string|worktree.*path.*(schema|parse|validate)' src/main src/relay src/shared src/renderer/src/store | head -260Repository: stablyai/orca
Length of output: 41782
🏁 Script executed:
printf '%s\n' '--- path comparison normalization ---'
nl -ba src/shared/cross-platform-path.ts | sed -n '1,90p'
printf '%s\n' '--- ID constructor callers ---'
rg -n -F 'getRepoMainWorktreeId(' src
printf '%s\n' '--- repo path input schemas ---'
nl -ba src/main/ipc/repos/repo-ipc-arg-schemas.ts | sed -n '35,105p'
nl -ba src/main/ipc/repos/repo-ipc-arg-schemas.ts | sed -n '170,200p'
printf '%s\n' '--- snapshot contract and repository-owned producers ---'
nl -ba src/shared/runtime-session-contracts.ts | sed -n '248,280p'
rg -n -F 'RuntimeMobileSessionTabsResult' src/main src/relay src/shared/runtime-session-contracts.ts
rg -n -F 'worktreeId:' src/relay src/main/runtime src/main/ipc | rg 'session|snapshot|worktree' | head -160
printf '%s\n' '--- git worktree path parsing ---'
nl -ba src/main/git/worktree-list-reader.ts | sed -n '195,260p'
printf '%s\n' '--- marker and related exact-key operations ---'
nl -ba src/renderer/src/runtime/web-session-close-intent.ts | sed -n '18,38p;158,175p'
nl -ba src/renderer/src/runtime/web-session-tabs-sync/apply-snapshot.ts | sed -n '1,65p'
nl -ba src/renderer/src/runtime/web-session-tabs-sync/apply-preparation-base.ts | sed -n '175,201p'Repository: stablyai/orca
Length of output: 42494
🏁 Script executed:
printf '%s\n' '--- snapshot producer ---'
nl -ba src/main/runtime/orca-runtime-schedule-mobile-session-tabs-changed.ts | sed -n '1,155p'
printf '%s\n' '--- snapshot result finalization ---'
nl -ba src/main/runtime/runtime-mobile-session-result-finalization.ts
printf '%s\n' '--- runtime worktree identity passed into snapshot producer ---'
rg -n -F 'getSnapshot(' src/main/runtime
rg -n -F 'buildSnapshot(' src/main/runtime/orca-runtime-schedule-mobile-session-tabs-changed.ts src/main/runtime/orca-runtime-prune-mobile-session-tab-group-layout.ts
printf '%s\n' '--- repo path source schema and path resolution ---'
rg -n -C 3 'path: z\.string\(\)\.min\(1\)' src/main/ipc/repos src/main/ipc | head -100
printf '%s\n' '--- path trailing slash helper only ---'
rg -n -F 'function trimRuntimePathTrailingSlash' src/shared/cross-platform-path.ts
nl -ba src/shared/cross-platform-path.ts | sed -n '105,140p'Repository: stablyai/orca
Length of output: 17823
Keep the exact worktree ID in the marker set.
When a worktree path ends in whitespace, the close path can store a trimmed ID while snapshot application queries and clears with the original ID. An affirming-empty snapshot can then remove the row instead of keeping its closed-terminal tombstone. A later snapshot with that same ID cannot clear the marker either. Store the original ID and use trim() only to reject blank input.
Suggested fix
export function rememberUserEmptiedWorktree(worktreeId: string): void {
- const trimmed = worktreeId.trim()
- if (trimmed) {
- userEmptiedWorktreeIds.add(trimmed)
+ if (worktreeId.trim()) {
+ userEmptiedWorktreeIds.add(worktreeId)
}
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| export function rememberUserEmptiedWorktree(worktreeId: string): void { | |
| const trimmed = worktreeId.trim() | |
| if (trimmed) { | |
| userEmptiedWorktreeIds.add(trimmed) | |
| } | |
| } | |
| export function forgetUserEmptiedWorktree(worktreeId: string): void { | |
| userEmptiedWorktreeIds.delete(worktreeId) | |
| } | |
| export function wasWorktreeEmptiedByUserClose(worktreeId: string): boolean { | |
| return userEmptiedWorktreeIds.has(worktreeId) | |
| export function rememberUserEmptiedWorktree(worktreeId: string): void { | |
| if (worktreeId.trim()) { | |
| userEmptiedWorktreeIds.add(worktreeId) | |
| } | |
| } | |
| export function forgetUserEmptiedWorktree(worktreeId: string): void { | |
| userEmptiedWorktreeIds.delete(worktreeId) | |
| } | |
| export function wasWorktreeEmptiedByUserClose(worktreeId: string): boolean { | |
| return userEmptiedWorktreeIds.has(worktreeId) |
…t-side The mirror wrote its closed-last-terminal tombstone row for any host snapshot affirming zero terminals, but the host cannot tell a user close from a session it ended itself. After a remote session exited cleanly, every seeder read the tombstone as a deliberate emptying: the workspace reopened empty, with no terminal, no agent, and no error, and only an app restart recovered it. Remember the user's own emptying instead: the tab-close path marks the worktree when its close intents cover every terminal tab the row still holds, and only that marker may leave the tombstone row. A host-side exit deletes the row, so the existing after-wake ensure re-creates the surface on the next activation. The marker clears when a terminal row returns or the worktree's close intents do.
Address review feedback on the remote-workspace reseed change. - Keep the exact worktree ID in the user-emptied marker. Snapshot application checks and clears the marker with the ID it was given, so a trimmed key never matched a padded worktree ID; trim() now only rejects blank input. - Forget the marker when the host refuses a close and republishes its snapshot, and when the close RPC fails with a non-definitive error. The terminal stays alive in both cases, so the user did not empty the workspace. Tests cover the exact-ID round trip and the marker after each close outcome.
ddabb36 to
08a0a55
Compare
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to GitHub limitations.
🟡 Minor · Preserve the missing-versus-empty row distinction at… · apply-preparation-base.ts:191
src/renderer/src/runtime/web-session-tabs-sync/apply-preparation-base.ts:191
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winPreserve the missing-versus-empty row distinction at
apply-preparation-base.ts:191.The
[]selected for a user-empty snapshot can compare equal to a missing row, sowithWorktreeEntryskips inserting the tombstone. Conversely, an existing empty row can compare equal to thenullselected for a host-side exit, so deletion is skipped. Wake seeding uses row presence and can therefore make the wrong reseeding decision.Distinguish nullish values from explicit empty arrays while preserving
null/undefinedequality and nonempty-row comparisons.Suggested fix
const left = a ?? [] const right = b ?? [] + if ((a == null) !== (b == null)) { + return false + } if (left.length !== right.length) {
ℹ️ Review info
⚙️ Run configuration
- Configuration used: Repository UI
- Review profile: CHILL
- Plan: Advanced
- Run ID:
082feb1a-5864-4e30-a8ac-14cc61129ed2
📥 Commits
Reviewing files that changed from the base of the PR and between ddabb367f0308d403dfb4e8e825f2c354a99f573 and 08a0a55.
📒 Files selected for processing (5)
src/renderer/src/runtime/web-runtime-session-tab-activate-close.test.tssrc/renderer/src/runtime/web-runtime-session-tab-lifecycle.tssrc/renderer/src/runtime/web-session-close-intent.test.tssrc/renderer/src/runtime/web-session-close-intent.tssrc/renderer/src/runtime/web-session-tabs-sync-emptied-workspace-tombstone.test.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.
…session-exit-reseed # Conflicts: # src/renderer/src/runtime/web-runtime-session-tab-lifecycle.ts
… row sameTerminalTabs treated a missing row as equal to the explicit empty row, so a user-emptied workspace whose row was missing never got its tombstone. A missing row now differs from an explicit empty one; a null frame against an existing tombstone stays a no-op, so a synthesized unpublished frame cannot undo the user's own emptying.
Summary
When a remote workspace's last terminal session exits on its own (the agent process ends, the user exits the session), the client mirror wrote its closed-last-terminal tombstone row — so every seeder read the workspace as deliberately emptied, and reopening it showed an empty workspace: no terminal, no agent chat, no error. Only restarting the app recovered it. The tombstone now records only the user's own emptying; a host-side exit deletes the row and the existing after-wake ensure re-creates the surface on the next activation.
The incident
Reported twice against the same flow (a Proxmox LXC environment recipe, per-workspace cloud VM):
session-exited ... code 0, cause "exited".running, the serve stays up, the pairing connection stays healthy (verified: the app re-authenticated within seconds after a forced serve restart).Root cause
#21630taught the mirror to keep an explicit empty terminal row when a host snapshot affirms the workspace holds zero terminals, because a deleted row reads as "never initialized" and re-seeds the workspace on every focus (STA-6173). The empty row is the tombstone every seeder honors.But a host snapshot affirming zero terminals cannot distinguish two opposite situations:
Both produce the same affirming-empty frame, so both got the tombstone, and the workspace could never regain a surface mid-run. App restart escaped it only because startup hydration takes a different seeding path.
The e2e test added here reproduces the exact sequence against a headless paired runtime: create a workspace terminal, open the workspace from the sidebar, end the session cleanly server-side, reopen from the sidebar. On the pre-fix tree the host terminal count after reopen is 0 and the client renders no tabs and no errors.
The change
The client knows one thing the snapshot cannot: when a close is the user's own.
closeWebRuntimeSessionTabnow records, alongside its close intents, that this close covered every terminal tab the row still held — the user emptied the workspace. The mirror's affirming-empty decision consults that marker:ensureWebRuntimeWorktreeTerminalAfterWake, already wired to every sidebar activation) reads "never initialized", re-creates the surface on the host, and the workspace comes backThe marker clears when a terminal row returns for the worktree or when the worktree's close intents clear, so a later host-side exit still reseeds.
This stays inside
#21630's own vocabulary: removal frames and synthesized unpublished frames still delete, a never-initialized worktree still gets no row, and only positive evidence of the user's own close may tombstone.Tests
web-session-tabs-sync-emptied-workspace-tombstone.test.ts: the tombstone case now models the user's close (the old case asserted a host retraction as a user tombstone — the conflation itself); new cases for a host-side exit deleting the row and for the marker clearing once a terminal returns.paired-remote-terminal-session-exit-reopen.spec.ts(e2e, electron-headless): the incident end to end, red on the pre-fix tree, green after.typecheck:node,typecheck:web, and the changed-code quality gate pass.Scope and known edges
closeWebRuntimeSessionTabdoes not set the marker, so such a workspace may reseed on the next activation rather than staying empty. That is the pre-fix(session): remember closed and initialized runtime tabs #21630 behavior for that path only, never a strand.