mirror of
https://github.com/stablyai/orca.git
synced 2026-09-29 16:02:50 +00:00
2c0dcc83d1cfe5250bc4f2eff35cfb60b9818221
10508
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jinwoo-H
|
2c0dcc83d1 | Simplify mobile push delivery and mirror desktop notification categories | ||
|
Jinwoo-H
|
35864ab077 | Merge main and preserve push schema dependencies | ||
|
Jinwoo-H
|
48fb41b324 | fix(push): verify successor-first Cloud Run recovery | ||
  Brennan BensonandMerge Sim
|
c1e15c4008 |
feat(native-chat): read a tool batch as a group (#19372)
* feat(native-chat): read a tool batch as a group A run of several tool calls collapsed to one joined string: names and arguments run together, separated by a middle dot that also occurs inside `browser.open` and `tools/read`, with the overflow cut mid-token. Opened, the member rows sat flush with the header and with the message content around them, so the batch had no visible end. Two presentation changes, no new derivation: - Each member gets its own bounded pill in the collapsed header, carrying its own category glyph, so the boundary between calls is a shape rather than a character. Pills wrap instead of truncating, and members past the summary cap are counted in `+N more` rather than dropped silently. - Opened members are indented under the header, which is what marks where the run ends. `toolRunSummaryMembers` keeps the run's leading calls apart instead of pre-joining them; `summarizeToolRun` now derives its string from it, so mobile's header is byte-identical and the two cannot disagree about which calls speak for a run. Two existing behaviours are pinned by test rather than changed, both being naming decisions rather than layout ones: the header still prints the raw `mcp__linear__list_issues` while the row beneath prints the split name, and a call carrying only a `url` still falls through to a JSON preview clipped at 28 characters. * fix(native-chat): bundle hidden tool count copy * fix(native-chat): drop the filled pill for a glyph-led member list Rendered in the app, the filled chips were wrong twice over. `bg-accent` is reserved for hover/active row backgrounds, and the only full-strength use of it in native chat is on payload and diff surfaces — so each member read as a shrunken content block, and a run became the loudest thing in the transcript. Worse, `flex-wrap` degenerated: at a 297px pane each member is 274-288px, so every one took its own line, the header grew 24px to 72px, and the `5x` count centred against the block landed beside the second member as though it counted that call alone. The glyph already marks where a member starts, so the fill was carrying no information the icon wasn't. Members are now inline, glyph-led, and separated by spacing; the list stays one line and truncates as a whole, as it did before this branch. `+N more` moves outside the truncating span so the count of what is not shown survives a pane too narrow to print the list. Members carry `data-tool-run-member` rather than being found by their fill. * fix(native-chat): let the run summary size to its content `flex-1` on the truncating member list made it claim the header's slack, so `+N more` was pushed to the far right edge with a gap between it and the last member it counts. Without it the span still shrinks and truncates — `min-w-0` plus the default shrink is what drives the ellipsis, which is how the header worked before this branch — and the count now sits directly after the list at every width. * fix(native-chat): separate run-header members with real whitespace An `ml-3` margin marks the boundary on screen but is invisible to a copied selection and to the button's accessible name, so the header read `ls -latools/read`. Adds a space text node between members and trims the margin to pay for its width. `+N more` also picks up the hover transition every other header segment already had. --------- Co-authored-by: Merge Sim <sim@local> |
||
  BingZandNeil
|
5bd0247aaa |
fix(xterm): remove scrollback decorations by identity (#13178)
* fix(xterm): fire Marker dispose before clearing line (#10879) Scrollback trim under search highlights was O(k²) because dispose set marker.line to -1 before onDispose, collapsing SortedList keys. Fire listeners first so delete still sees the real line, then clear the line. Fixes #10879 * fix(xterm): remove scrollback decorations by identity * perf(xterm): avoid index arrays for unique decorations --------- Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com> |
||
|
Neil
|
cbc7bb418c |
fix(ci): read the changed-path list past the first pipe buffer (#19409)
`pr-code-change-scope.mjs` read stdin with `readFileSync(0, 'utf8')`, a single read of fd 0. Once the writer outgrows the 64 KB pipe buffer that read returns early or throws EAGAIN, the script exits 0 having emitted no `name=value` pairs, and `tee -a "$GITHUB_OUTPUT"` records nothing -- so every lane the classifier gates is silently skipped rather than failing loudly. A PR opened long ago carries a stale `pull_request.base.sha`, so the gate's merge-base diff spans the whole base branch. PR #13178 diffed 13,294 files (773 KB) against a base 1,592 commits behind main and lost typecheck, test, static analysis, xterm patch sync, package and e2e to this. Stream stdin instead, matching how the sibling `pr-e2e-source-routing.mjs` already reads the same list in the same workflow. |
||
  OrcaWinandm4air
|
102402e41e |
fix(deps): update DOMPurify sanitizer hardening (#19377)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
OrcaWinandm4air
|
b8f6c7cabe |
fix(deps): update react-i18next for TypeScript 7 and parser fixes (#19378)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
OrcaWinandm4air
|
a4e5106e14 |
chore(docs): patch brace expansion resource exhaustion fixes (#19382)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
OrcaWinandm4air
|
1dae024ab2 |
chore(mobile): patch xmldom security fixes in plist tooling (#19380)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
OrcaWinandm4air
|
91fbc1529a |
fix(deps): harden cloud HTTP and WebSocket dependencies (#19362)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
OrcaWinandm4air
|
cfd59f2ca0 |
fix(deps): update desktop parser security dependencies (#19361)
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local> |
||
|
Brennan BensonandMerge Sim
|
3b8128df04 |
fix(orchestration): index per-PTY mailbox reservation cleanup (#19390)
Co-authored-by: Merge Sim <sim@local> |
||
|
Jinwoo-H
|
8fd9947c1a | test: align push fixtures with CI database and secret scanning | ||
|
Jinwoo-H
|
794b4373e6 | fix: close push delivery races and isolate gateway validation | ||
|
Neil
|
588240043e | fix(sidebar): reveal collapsed workspaces without clearing filters (#19398) | ||
|
Brennan BensonandMerge Sim
|
9f044031fc |
fix(native-chat): render compaction notices, plan documents, and images (#19228)
* fix(native-chat): render compaction notices, plan documents, and images * fix(native-chat): avoid repeating notice text in details * fix(native-chat): journal canonical and legacy compaction events * test: add digest to native chat notice payload fixture * chore(native-chat): drop the planning doc from the PR --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Jinwoo-H
|
d1b3ae8086 | test: align hook notification fixtures with mobile delivery | ||
|
Jinwoo-H
|
761d2a046c | fix(test): allow isolated CI PostgreSQL in registration race | ||
|
Jinwoo-H
|
1fa630103a | fix(ci): assign push deployment identity to Terraform root | ||
|
Jinwoo-H
|
ff4e8d1092 | fix(push): address review races and isolate deployment authority | ||
|
Neil
|
2265fce591 | chore(mobile): remove stale max-lines exceptions (#19366) | ||
|
Jinwoo-H
|
eac2a9dba1 | fix(push): fence cancelled registrations and close unregister drain race | ||
|
Jinwoo-H
|
1b2b0c63aa | Simplify mobile push delivery and unify notification policy | ||
|
Jinwoo-H
|
5d4faa5efa | Merge main and resolve mobile push integration conflicts | ||
|
Brennan BensonandMerge Sim
|
d0506bf5de |
feat(native-chat): add execution details and tool row identity (#19226)
* feat(native-chat): annotate tool rows with execution and source details * fix(native-chat): require explicit MCP identity for tool annotations * test: add required state to MCP projection fixture --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Brennan BensonandMerge Sim
|
0b3af9ddbc |
Show native chat message timestamps on hover and keyboard focus (#19218)
Co-authored-by: Merge Sim <sim@local> |
||
 Jinwoo Hong
|
bcd4076dd1 |
fix(relay): never cache a region hint from a one-region catalog (#19349)
* fix(relay): never cache a region hint from a one-region catalog The director lists only regions with a serving cell, so a roll wave shortens the catalog to one entry. The resolver required only every *listed* region to be measured, so that lone region won against nothing and was cached for 24 h: a US desktop refreshing while US cells rolled published asia-east2 for a day, the incident #19233 was written to end. Now fewer listed regions than the fleet serves withholds the hint (1 h no-hint TTL), the same outcome as an unmeasurable peer. * test(relay): give the unstable-probe case a two-region catalog so it has one cause |
||
|
Jinwoo Hong
|
d936d8da82 |
revert(mobile): pull the relay connect-speed mobile pass pending a smaller, verified re-land (#19348)
* Revert "feat(mobile): time relay dial stages so diagnostics say where a slow connect went (#19245)" This reverts commit |
||
|
Brennan BensonandMerge Sim
|
6a47d2831f |
fix(native-chat): scope composer file drops to the pane that received them (#19328)
* fix(native-chat): scope composer file drops to the pane that received them A native OS file drop resolving to `target: 'composer'` carried no pane identity, so the window-wide payload was attached by every mounted composer. Because inactive chat tabs stay mounted (hidden), one drop populated every chat pane's attachment cache, and those chips replayed whenever the user returned to a tab they never dropped into. The workspace-creation composer and chat composers also leaked into each other, since neither could tell which surface actually received the drop. Composer drops now carry a `scopeKey` the way a terminal drop carries its tab and pane leaf id: the composer publishes its pane key as `data-composer-scope-key`, the preload harvests it during the composedPath walk, and each composer attaches only its own. The workspace composer's last-wins ownership stack now claims unscoped payloads only. * test(native-chat): supersede the bug-asserting drop repro with the scoping test The repro that landed on main asserts the pre-fix behavior (a drop reaching every mounted composer), so it fails once drops are scoped to the pane that received them. Its scoping cases now live in native-chat-composer-drop-scope.test.tsx, which keeps its editor-target control case verbatim and adds coverage for unscoped composers and a scope key published inside the drop-target marker. * test(native-chat): cover workspace composer drop isolation * fix(native-chat): authorize external attachment paths before preview --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Brennan BensonandMerge Sim
|
5cefb440bf |
fix(native-chat): stop an unanswered host from reading as one that refuses structured chat (#19321)
* fix(native-chat): stop an unanswered host from reading as one that refuses structured chat `readLocalRuntimeCapabilities()` returned `[]` both before the first status probe landed and after one failed, so "not asked yet" and "host says no" were the same value. Every structured-chat launch route consumed it, and an unprobed host was routed to legacy chat exactly as a refusing one is. Keep the two apart: the cache holds `null` until a probe succeeds, a failed probe leaves it `null` rather than emptying it, and the launch route names the case with its own blocker instead of borrowing `runtime-capability`. No routing outcome changes — both cases still decline structured chat. The point is that the reason is now truthful, which is what the routing work needs to build on: once a launch can target a runtime peer, capabilities come from that host, and an unanswered remote must not be indistinguishable from one that refuses. `hostCapabilities` on the launch route stays local-only at every call site; a per-target resolver replaces it when the route learns to reach a peer. * test: cover unknown runtime capability lifecycle and launch fallback --------- Co-authored-by: Merge Sim <sim@local> |
||
 Jinjing
|
1d1b73c408 |
Defer inactive browser pages across worktree switches (#19326)
* Defer inactive browser tabs while retaining their viewport slots Restore worktrees and tabs on demand instead of mounting the full tree. Only render active pages and those required by automation, mobile drivers, or remote viewers. Inactive panes stay deferred with persistent viewport slots so their webview guests survive chrome unmounts, reducing memory overhead when opening workspaces with many tabs. * Defer browser pages until active and recover if evicted Pages defer rendering until active, then retain state when inactive. Add recovery logic to restore guests evicted by workspace memory pressure when pages are reactivated. * Stop retaining browser content when worktree is inactive - Browser panes and pages now unmount when their worktree transitions to inactive, except for pages claimed by automation/mobile/viewer consumers - Prevents unwanted restoration of all hidden browser tabs when switching between worktrees - Tests verify proper cleanup at scale and correct page lifecycle across worktree switches * Preserve document-preview guests when switching browser tab profiles Document previews use a fixed partition and should not be recreated when the profile changes. Only URL-based pages need their webviews destroyed and rebuilt with the new profile. Includes test coverage. * Create browser pages cold to defer guest initialization Pages created in the background now start with loading: false, since they don't own a guest until first shown. Only live guests can report loading status, so background tabs sit idle until activation triggers navigation. * Prevent document preview from swallowing pointer events during drag Move webview registration to attachDocPreviewWebview before append, ensuring it's enrolled in drag passthrough before becoming hittable. When a document preview tab remounts mid-drag, the previous hook-based enrollment landed too late. Also refactor mountEligible into isBrowserPagePanePaintable for clarity. |
||
|
Brennan BensonandMerge Sim
|
ce4a3a4186 |
feat(chat): add structured session rewind backend (#19235)
* feat(chat): add structured session rewind backend * fix(chat): make interrupted session rewinds recover safely * fix(native-chat): negotiate rewind runtime capability * fix(native-chat): consolidate remaining adapter imports --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Jinwoo-H
|
e53b95e26a | fix(push): reconcile fully handled notification summaries | ||
|
Jinjing
|
9fed61e5c2 |
Persist agents sidebar search visibility as pairing-local preference (#19313)
* Persist agents sidebar search field visibility as pairing-local preferen - Add `agentsShowSearch` to workspace UI state with default on - Include in pairing-local fields so preference syncs across clients - Convert search from menu action to checkbox menu item for explicit toggle - Update activity thread options menu to reflect checkbox state - Add localization strings across all supported languages - Update RPC schemas and preference persistence layer - Includes readiness validation reports confirming feature is clean * rm review * fix documentation |
||
|
Jinwoo-H
|
8da335d778 | fix(mobile): reconcile dismissed pushes on first connection | ||
|
Jinwoo-H
|
1bb74e4599 | fix(notifications): preserve away alerts and recover missed dismissals | ||
|
Jinwoo Hong
|
83b1558ecc |
feat(mobile): time relay dial stages so diagnostics say where a slow connect went (#19245)
* feat(mobile): time relay dial stages so diagnostics say where a slow connect went A 10s connect was unattributable from a shared report. Relay dial stages carried no timestamps, so nothing could tell "the cell never answered relay-hello" from "the E2EE handshake was slow", and the per-state dweltMs the client already computed went only to console.log — invisible without a debug build. RelayDialStageTracker now stamps each stage entry from a monotonic clock (performance.now where present, wall clock otherwise) and returns the duration of the stage it just left. The session logs one entry per stage, and settles the in-flight stage on connect, failure, or close, so a dial that dies mid-way still names the stage it never finished. dweltMs joins the same buffer as a structured field instead of console. Durations ride the existing per-host log buffer and its cap, so memory is unchanged and no new storage appears. The report derives two lines from them: the latest dial's stage breakdown (a reconnect loop must not average away the attempt being reported) and total dwell per connection state. Both are numbers and closed-enum names, and the entries still pass through the existing redaction. * fix(mobile): never let a diagnostics sink break a dial, and pin timing names to their enums Review follow-ups on the dial-stage timing work. The stage timing emitted on the confirm's success path ran inside the try that calls fail(), so an onLog sink that threw would have turned a good connect into a failed session. The same hazard existed on the direct path, where the dwell emit sits in publish() ahead of the listener loop and the connect waiters. Both sink calls are now isolated: a broken sink loses a log line and nothing else. The persisted-log validator accepted any string as a timing name, and the report echoes that name unredacted. Names are now checked against the closed enum for their kind, backed by Record<Union, true> tables so adding a stage or a state breaks the build rather than silently widening what a corrupted store can inject. Entry volume: every reconnect cycle walks four connection states, so logging each one would roughly double what a slow-connect report holds against the unchanged 200-entry per-host cap. Transitions under 100ms are therefore not buffered. They cannot be where a slow connect spent its time, and console still shows all of them. States that flap slowly, which is the case support cares about, still land in the log. RpcClientConnectionState takes an optional clock so dwell thresholds are testable without sleeping. * fix(mobile): reject a negative stored stage duration when hydrating the log A persisted timing only had to be finite to survive hydration, so a corrupted `ms: -1` reached the diagnostics report, where the dial summary sums the stage durations and a negative would subtract from the total. Producers clamp at 0 (`elapsedMs`), so anything below it is corruption. 0 itself still hydrates: a stage the dial passes through instantly is real. * refactor(mobile): move the relay liveness profile out of the session so the dial log fits * fix(mobile): never let the liveness-timeout log line keep a dead relay connected * test(mobile): prove the throwing timeout sink was actually reached |
||
|
Jinwoo Hong
|
ceafdcad2f |
perf(mobile): race the direct and relay dials from t=0 on every reconnect (#19308)
* perf(mobile): race the direct and relay dials from t=0 on every reconnect A foreground reconnect gave the direct dial a fixed 2.5s head start, and while that dial sat in 'connecting'/'handshaking' the supervisor refused to open a relay socket at all. A phone that is off the LAN paid the full head start on every reconnect and got nothing for it, and a phone whose relay dropped could only return to the LAN through three hysteresis probes. Both dials now start together and the first authenticated socket is adopted through the existing migrateTo cutover. Nothing about the migration machinery changes: only who is allowed to start a dial. - The relay dial now yields to a live session and to nothing else. An unfinished direct dial is progress on the other runner, not a reason to stand still. - The direct return probe grows a second adoption policy. Against a live relay hysteresis still has to prove direct stable; during a reconnect there is no session to protect, so an authenticated direct socket wins outright. probeNow pre-empts a pending 15s tick so that dial starts with the relay dial, not after it, and the dial itself no longer waits for the operation mutex — a relay dial holding it is exactly the case the race exists for. - A loser closes and books nothing. The relay dial withdraws inside migrateTo and returns 'aborted', so no backoff is booked against it; a direct socket that loses leaves the promotion streak untouched. Only a reconnect that both paths lose books a failure, once, on the relay cadence. Kept: the 30s background grace and the foreground gate, because a backgrounded phone must not open a billed relay splice; the shared failure cooldown, because a genuine relay failure still has to be paced; the hysteresis dwell after a migration, because it is what stops a marginal LAN flapping a healthy session. The accepted cost is one relay socket per reconnect for a phone that is on its LAN. It closes as soon as the direct path authenticates, before the resume confirm, because migrateTo only checks the abort predicate after E2EE auth. Tests that encoded the removed rules: - 'fails over when the direct retry loop publishes reconnecting' asserted no relay dial while direct was handshaking. The failover now precedes the direct client giving up, so it asserts the dial instead of its absence. - 'does not spend a queued relay retry while direct authentication is progressing' encoded the block outright; it now asserts the retry runs on the failure cadence while a handshake drags on. - The four grace-race cases move to mobile-endpoint-reconnect-race.test.ts as t=0, direct-wins, background/resume and both-lose cases. - Five relay-bookkeeping cases now state their premise with unreachableDirect. They describe a phone with no LAN, which used to be implicit and is now load-bearing: with a reachable LAN the direct socket wins those reconnects. * fix(mobile): withdraw a lost relay dial pre-handshake and damp blip races Review follow-up to 4e31130471. Racing both paths from t=0 was correct but charged the LAN case twice: once per reconnect in cell work, and again whenever the LAN flapped. Withdraw before the handshake. migrateTo only consults its abort predicate after E2EE authentication, so a dial that had already lost still made the cell reserve a splice and the desktop finish a key exchange. The establisher now watches the logical client across the dial and closes the cell socket the moment direct authenticates. In the common window, after relay-auth is on the wire and before the hello lands, nothing of the key exchange has started, so the withdrawal costs the desktop nothing. The dial still reports itself aborted and still books nothing. The watch is dropped once migrateTo returns, because past the cutover this session is the active path and a later direct promotion must not read as a reason to close the client's own socket. Damp races that a blip started. relayDialAllowed yields only to a live session and a lost race books nothing, so a flapping LAN drove one cell socket per blip with only the relay's per-host rate limiter as a backstop, and reaching that limiter would have converted a benign race into a booked relay failure. After a race is lost to direct, the next unforced race is suppressed for 2s, doubling per consecutive loss to a 30s cap. This is not backoff and is kept separate from it: a forced replacement is never damped, a relay dial that wins clears the streak, and a foreground resume clears it too, so the path the user is watching never waits. The window arms its own lapse timer, so a LAN that dies inside the window still reaches relay without a new trigger. A superseded cutover no longer escapes probe() as an unhandled rejection. Only the probe timer calls it, and it discards the promise, so the routine end of a lost race would have surfaced as one. Credential rotation moves to MobileRelayCredentialRefresh. The supervisor crossed the 300-line cap; rotation is a self-contained responsibility that only runs over a live direct connection, so it splits cleanly instead of taking a max-lines bump. * fix(mobile): end a damper window as soon as the direct path is really gone Round-2 review follow-up to 9a21da4326. The damper armed its window when direct won the race, and nothing shortened it. A LAN that died inside that window left the phone waiting out the whole thing, up to 30s at the cap, with only a log line to show for it. My previous commit body claimed the path the user watches never waits; that was true only of a foreground resume, and it is corrected here. Losing the direct path now collapses the wait to a 250ms floor, so the next recovery races almost at once. The floor is not zero because the reason the damper exists is a LAN that drops and comes straight back, and a disconnect is how such a blip begins. So the rest of the window is kept aside rather than spent: if direct returns inside the floor it was a blip and the window resumes, and if the floor lapses with direct still gone it was an outage and the held window is void. Without the second half, one blip would have bought a flapping LAN a free pass on every race that followed, which is the case the damper was added for. The streak itself is untouched by the clamp. A LAN that flaps all afternoon still escalates toward the cap; only the current wait is cut short. record() now takes the same forceReplacement guard as suppresses(), so a forced replacement that stands down cannot grow the streak or be read as a loss to direct. A lease rotation or a reconsidered network change is not a LAN that flapped. Also documents that a genuine relay failure deliberately does not reset the streak, and that the damper and the failure backoff serialize rather than stack: a damped attempt never reaches the dial that would book a cooldown. * fix(test): give the direct-probe fixture the race-era hooks The phase-1 probe test predates canDial and adoptsOutright, so its hooks literal threw at the first dial. These cases model a live relay session. * docs(mobile): say why a finished credential refresh races relay instead of waiting on direct |
||
|
Jinwoo Hong
|
5857357fcf |
feat(relay): log the region probe and name the assigned cell (#19307)
* feat(relay): log the region probe and name the assigned cell A desktop silently pinned itself to a far relay region for a day and every phone connect paid the round trip. Nothing in the desktop logs said which regions were probed, what they measured, why one was rejected, or which cell the host landed on, so the only way to diagnose it was a bench harness. The resolver now emits one line per outcome. A refresh carries every region's probe origins, the discarded warm-up, the kept samples, the minimum, the spread, and a verdict, then the chosen region or no-hint with the reason it withheld one. Cache hits, diagnostic overrides, and a director that cannot list its regions each get their own line so a quiet run is never ambiguous. Self-heal logs the cached region, the best measured region, the assigned cell's round trip, and whether it kept or deleted the cache. Only a refresh reports a catalog failure; a self-heal never chose a region, so a line saying it withheld a hint would be a lie. Relay status now carries the assigned cell so the pairing panel can name it. The field is optional because an offline host holds no assignment and the web client answers from a stub that never has one. Splitting catalog fetching out of the preference module keeps both files inside the line budget without a lint disable. * fix(relay): drop the assigned cell from statuses not served on it The origin pool publishes offline while it still holds the assignment it is about to rotate, so the panel kept naming a cell nothing was served from. The same class of bug hid a second instance: the coordinator republishes registered right after the broker announces its cell, and that republish carried no cell, blanking the value moments after it was set. The cell would never have reached the panel in the real flow. Deriving the cell from the status at each publisher removes both. The rule lives beside the status type because it defines when the optional field is populated, and the coordinator reads the owned broker's endpoint rather than trusting a call site to remember to pass it. * i18n: add the relay cell label to the English catalog * test(relay): audit the relocated region catalog fetch call site * fix(relay): report a self-heal whose catalog request failed instead of staying silent |
||
|
Jinwoo Hong
|
643571def6 |
feat(mobile): draw the last known tab strip while a session reconnects (mobile pass) (#19281)
* feat(mobile): draw the last known tab strip while a session reconnects Reopening a workspace the phone has already visited threw away everything it knew. The route clears its tabs on mount, so until the reconnect lands and the first snapshot is applied the session screen has an empty header and a bare spinner, even though the strip it is about to be handed is the one it drew a minute ago. Persist the four fields the strip actually draws -- id, type, title, agent -- per host and workspace, and add a reconnecting-with-cache shape to the route state so those rows render immediately, disabled, under the ids the live snapshot will reuse. Live tabs always outrank the cache, so a mid-session drop keeps its mounted terminals; an exhausted retry loop or a rejected pairing outranks it the other way, because a strip the user cannot reach is worse than the existing offline affordance. With nothing cached the screen behaves exactly as before. The body stays a placeholder. Replaying stored scrollback into the terminal WebView would double-render the same rows once the live stream replays them, so the strip is the cached content and the body waits for the stream. * fix(mobile): keep shell titles and unpaired hosts out of the cached tab strip Review of the reconnect strip cache found two ways it leaked. A terminal's title is whatever the shell last set, which is routinely the command line: a psql URL with an inline password, a curl with a bearer token. Both fit well inside the 64-character cap and both were written to plaintext AsyncStorage verbatim. Browser tabs carried their page title the same way. Terminals and browsers now collapse to a fixed label, with a resolved agent naming itself because that lookup is a closed enum. The rule lives in the storage module rather than its caller, so it holds for entries an older build already wrote, and a tab type this build cannot draw is dropped instead of having its title trusted. The cache also survived forgetting a host. Nothing expired an entry, and the module-global memory map meant a later save from any surviving host serialized the forgotten host's rows straight back to disk. Both cleanup paths now evict by host, dropping the in-memory rows and rewriting storage, with a pending debounced write cancelled so it cannot restore them. Also: the storage key digests the workspace id, which ended in a filesystem path, and cached rows carry the same de-emphasis as the disabled tab-bar buttons beside them, so an inert row does not pass for a live one. * fix(mobile): make a forgotten host's cached tab strip actually leave disk Review finding on this PR, fixed here so it rides along with the rest. writeFile swallowed its own rejection, so deleteCachedSessionTabStripForHost resolved successfully while the unpaired host's plaintext tab titles stayed on disk, and removeHostAndCloseClient discarded the promise with void so nothing could have observed the failure anyway. The write now throws. The debounced save keeps a best-effort catch, since a dropped cache refresh costs one repaint and the next save rewrites the whole map, so only the deletion path needs the failure. Host removal awaits the deletion and logs a failure but never rethrows: the metadata removal has committed and the client is closed by that point, so reporting a finished removal as failed would be wrong. The unpaired-host credential sweep already awaited the deletion and now sees the rejection, consistent with its sibling credential deletions. Two ways the rows could come back are closed as well. The cache refuses saves for a host it has been told to forget, so a snapshot racing the deletion cannot re-insert it, and the deletion awaits any debounced write already on the wire, since that write built its blob from the map as it was and would otherwise race the purge for the last word on disk. The refusal lasts for the process, so re-pairing the same host caches again from the next app launch, which is the cheap direction for a deletion the user asked for. * fix(mobile): order the tab-strip cache writes so a purge is the last word Two debounced writes could sit on the AsyncStorage bridge at once, and the second replaced the in-flight handle. A host purge then awaited only the newer write, so the older blob -- snapshotted while the forgotten host was still in the map -- could commit after it and restore the host's titles to disk. Writes now queue behind one chain and the purge queues last. The unpaired-credential sweep also aborted on a cache-purge failure, stranding the write revision and onDeleted after every credential was already deleted. It now warns and finishes, as removeHostAndCloseClient already did. |
||
|
Jinwoo Hong
|
6729f3a0b0 |
tools: add a phone-vantage relay connect benchmark (#19251)
* tools: add a phone-vantage relay connect benchmark Connect-speed work on the phone had no way to attribute latency to a hop. Timing the mobile app end to end only says "connect is slow", and a synthetic WebSocket probe does not exercise the credential check, the E2EE handshake, or the RPCs the phone blocks on before it publishes connected. This replays the shipped mobile wire sequence from Node against a real desktop over the production relay, so each phase gets its own number. The handshake is a plain-JS port of the mobile client session, which is only trustworthy if it stays byte-identical to what ships; a parity test runs it against the real desktop responder in the normal unit suite so drift in the transcript encoding, key schedule, or frame layout fails there rather than producing a bench that measures a handshake nobody uses. Adds a foreground mode for the resume-after-background question the phone lanes need: connect, go silent past the relay's client silence watchdog, then report whether the retained socket still answers and what the fallback redial costs. The bench writes a resume-credential bundle at runtime. That file carries a live device token for a real paired desktop, so the directory ignores it outright. * tools: make the relay bench name its target and opt in to dialing The supporting scripts carried production defaults: the director origin was hardcoded in both, and the hop-latency probe defaulted to a named production cell. Running either with no arguments sent live traffic at production, and the region probe did it on import, before any argument was read. A default like that is the wrong shape for a bench, because the operator never states what they are measuring against and a stray invocation is indistinguishable from an intended one. Every script now refuses to open a socket unless ORCA_RELAY_BENCH_LIVE=1 is set, and the director comes from --director or ORCA_RELAY_BENCH_DIRECTOR with no fallback. The cell origin is a required argument. Refusals print one line of usage and exit 2, so an accidental run is inert rather than live. The remaining host-id default is an id no desktop owns, which is the point of that probe: it measures the cell hop without reaching a desktop at all. * tools(relay-bench): type refuse() as never so origins are strings * tools(relay-bench): fail closed on hostile input and bounded arguments Review found the harness trusted whatever it was handed: the DevTools port and the director-supplied probe origins went straight into a URL, http origins were accepted, repeat counts came from a bare Number() cast, and the state file kept its existing mode. - Validate the DevTools port as a 1-65535 integer, so '80@attacker.example' cannot move the fetch off loopback via URL userinfo. - Require https for every origin, and refuse loopback, link-local, private, and multicast destinations. Region probe origins and the cell URL the director returns go through the same check, so a compromised director cannot aim the harness at the operator's own network. - Bound --runs, --rounds, runs, --gap, and --hold as whole numbers, so 'Infinity' exits 2 instead of looping forever against the relay. - Report a region as UNREACHABLE when every probe fails, rather than letting Math.min([]) spread into NaN and read as ok. - Bound the director /v1/resolve and /v1/regions fetches and report timeouts. - Return null openMs when the socket never opened, and clear dial, cell, and RPC timers on the first terminal event so Node exits promptly. - Default handle.rpc() to RPC_TIMEOUT_MS, not DIAL_TIMEOUT_MS. - Write the state file through a helper that creates the parent directory, refuses a symlink, and forces 0600 on an existing file; refuse to read one that is readable beyond the operator. - Read the pairing link from stdin or a 0600 file, never argv. - Reject missing and invalid positionals with usage and exit 2. Adds unit tests for the pure guards: argument parsing, bounded integers, port and origin classification, DNS vetting, state-file modes and symlinks, region verdicts, and pairing-link decoding. None opens a socket, and every network path stays gated on ORCA_RELAY_BENCH_LIVE=1. * tools(relay-bench): settle in-flight rpcs and guard an empty region catalog Follow-up to the review fixes. Clearing a pending rpc timer without a resolution swapped a 15 s timeout for an await that never returns, so the teardown paths now settle each waiter with a closed result. A director that answers /v1/regions with no regions now reports that and exits 1 instead of printing an empty round. * tools(relay-bench): attach the origin-vetting doc to the function it describes * fix(tools): resolve a director-named cell through DNS and fail cdp-eval clearly |
||
|
Jinwoo Hong
|
e60d9aaac9 |
fix(relay): attach a phone whose accept straddles a control rebind (desktop side) (#19238)
* fix(relay): attach a phone whose accept straddles a control rebind (desktop side) The cell announces a connection with a single conn-open. When the desktop's control socket dies mid-accept the phone waited out the 10s attach deadline and was closed HOST_OFFLINE, even though the desktop was online. host-hello-ack already restates those connections in pendingConns; the desktop parsed the field and threw it away. Desktop replays pendingConns on activation. It needs kind and relayDeviceId to dial: they decide the pairing authority a connection carries and the E2EE device binding, so neither may be guessed. Two sources, in order: - the ack entry, when the cell states them. This is the case where the desktop never received the conn-open at all, i.e. the headline scenario, and it needs the cell change in #19266. - the conn-open this process already saw, when the frame arrived but the data socket died with the control. Works against every deployed cell. An entry described by neither is skipped with a warning rather than dialed. Entries already spliced or already holding a data socket are skipped, so a rebind never double-dials. The desktop advertises x-orca-host-capabilities: pending-conn-details on the control upgrade so a cell knows the ack entries will be read. The header name and token are duplicated by hand because the desktop cannot import the relay contract; both sides assert the literals so drift fails a test rather than silently disabling the feature. Also stop dropping a conn-open that lands while the control is draining. A drain-only cell refuses new phones, so such a frame predates the drain and only that cell holds the waiting phone. * test(relay): pin the attach deadline the desktop mirrors by hand RELAY_HOST_ATTACH_DEADLINE_MS duplicates RELAY_PROTOCOL_LIMITS.hostAttachDeadlineMs, which the contract suite already pins to 10_000. Now that the cell and desktop halves land as separate PRs the two can drift independently, and drift would silently shorten both the observed-open eviction window and the deadline a replayed dial states, with nothing failing. |
||
|
Jinwoo Hong
|
c37413271e |
perf(mobile): open a session with parallel startup RPCs and a pre-warmed terminal engine (#19260)
Startup RPCs now fan out in parallel and the xterm engine pre-warms inside the real terminal frame while they are in flight, so the first pane inherits a warm WebView and an already-measured viewport instead of paying a round trip for it. The pre-warm opens its engine before measuring: web-ready only reports that the bundle loaded, and the WebView answers a measure with null until a terminal exists. It also pre-warms at the user's saved text size, because cell size is what the frame height gets divided by. Host writes such as worktree.activate wait for an evaluated status.get reply. Navigation still fails open when a host cannot answer one, but that fallback no longer reads as a passing compatibility verdict. |
||
|
Jinwoo Hong
|
e628090ad4 |
perf(mobile): cut the relay reconnect critical path and admit dead sockets faster (mobile pass) (#19280)
* perf(mobile): cut the relay reconnect critical path and admit dead sockets faster
Phone medians put E2EE authentication at ~424ms but `connected` at ~630ms,
because the session serialized two RPC round trips behind it: the resume
confirm (`pairing.getEndpoints`) and the capability advisory. Both now ride
the authenticated socket concurrently and off the critical path, so the
session publishes `connected` as soon as E2EE authenticates. Peer identity
is already proven by then — the confirm carries credential/lease bookkeeping
and the cell assignment check, and it still fails the session on a bad answer
or a foreign relayHostId, only later. `persistResumeConfirmation` awaits the
new `whenResumeConfirmed()` instead of assuming the answer is present at
`connected`.
Foreground liveness on a retained relay: `notifyForeground('app-resume')`
now probes past the 10s voluntary minimum on urgent bounds (2s, one miss),
so a socket that died while the process was suspended is admitted in ~2s
instead of ~8s. Focus and network nudges keep the old minimum and bounds.
Relay sessions also gain a 25s idle sweep, gated on foreground so a
backgrounded app spends no probes.
Recovery is no longer blocked by the direct return probe. The probe's 12s
dial is a pure observation on its own socket, so it takes the supervisor's
operation mutex only for the cutover; a relay recovery landing during a
foreground return now starts immediately instead of waiting the budget out.
Requests that do land during the cutover are queued in a new
RelayRecoveryIntentQueue and replayed on release — an owning forced
replacement keeps its intent, everything else replays as a plain recovery.
Tests updated deliberately, for the new ordering:
- 'sends no periodic traffic while an authenticated relay is idle' asserted
the absence of any relay idle probe, which is exactly the gap D3 closes.
Replaced by a sweep test plus a backgrounded no-probe test.
- 'rate-limits foreground sequences without suppressing a retry' asserted
that app-resume was suppressed inside the 10s minimum. An app resume is
now the one nudge that must never be rate-limited.
- the session helpers waited for the confirm answer before `connected`;
they now authenticate, read both concurrent frames, and settle them.
* fix(mobile): book backoff when a relay resume confirm fails after the cutover
Review round 1 on 352bfd2300.
P1: publishing `connected` at E2EE authentication made `migrateTo` resolve
before the resume confirm answered, so a confirm that failed afterwards —
a `relayHostId` mismatch from a rehomed desktop is the live case — was still
reported as an `established` dial. registerFailure was skipped, no cooldown
was booked, recordMigration()/setActiveSession() ran for a dying session, and
the queued-recovery replay redialled immediately: a tight loop with a
connected→disconnected blip per pass. The establisher now awaits
whenResumeConfirmed() after the cutover and, if the session is no longer
connected, reports a failed dial (or an aborted one when direct won or the
supervisor went inactive) exactly as a rejected migrateTo used to. The UI
still connects early; only the supervisor's bookkeeping waits.
The state check, rather than getFailure(), is the oracle: a live session can
carry a latched failure without having failed yet, and "is this session still
alive once the confirm settled" is precisely the question migrateTo used to
answer.
P2: the resume probe profile goes to two 2s misses instead of one. The first
frame after a resume rides a cold radio and a possibly distant cell, so one
slow answer is not proof of a dead link; the verdict still lands at 4s rather
than the previous 8s.
Nits: the direct probe's two early returns no longer close the candidate the
finally also closes (the second shape pre-existed); RelayRecoveryIntentQueue
is cleared in the supervisor's stop().
Mutex-hold note: persistResumeConfirmation, and now the establisher's own
await, are bounded by the confirm's request timeout. That would have been the
session's 30s default, so the confirm is pinned to RELAY_CONFIRM_TIMEOUT_MS
(12s) — the same bound migrateTo's waitForAuthenticated applied before.
Test: a supervisor-level case where every dial authenticates then fails the
confirm must book 250/500/1000ms backoff with no immediate redial, and must
never record a migration. It fails on the pre-fix establisher.
* fix(mobile): close three relay probe and liveness gaps from review
Review findings on this PR, fixed here so they ride along with the rest.
Direct return probe: schedule() guarded only the pending timer, so a caller
asking for an immediate probe while a dial was in flight started a second one
that overwrote activeProbe. stop() then reached only the newest socket and left
the earlier dial running out its 12s budget. Releasing the operation mutex for
the dial removed the only thing that had been serializing probes, and the
background bounce hits it directly: background() cancels the timer but leaves an
in-flight dial alone, and the matching foreground return asks for a probe at
once. The in-flight probe now owns the next slot and re-arms on the soonest
delay any caller asked for, so an urgent request is deferred rather than dropped
on the 15s floor.
Liveness watchdog: both retry paths in handleProbeTimeout, the tolerated-miss
one and the unfair-window one, retried without rechecking shouldIdleProbe. An
idle-sweep probe that started in the foreground could therefore keep spending
probes after the app backgrounded and terminate a healthy relay on misses that
were really iOS suspending the socket, which is the exact reading the foreground
gate exists to prevent. Probes now carry their origin, and an idle-sweep probe
that times out while backgrounded clears its state and re-arms the sweep with no
misses carried forward. Caller probes still reach a verdict.
Relay RPC session: whenResumeConfirmed() handed a pre-authentication caller an
already-resolved promise, so the documented contract only held after
authentication. No caller can reach that window today, since publishAuthenticated
assigns the promise before publishing 'connected' and both readers run after
migrateTo resolves, but the type comment promised more than the code delivered.
The deferred now exists from construction and settles on the confirm, on fail(),
and on close(), which are the only ways the session can end. Both endings had to
settle it and already shared nearly all of their teardown, so they are unified
behind one terminate().
* fix(mobile): give a resume probe its own miss budget
A resume probe supersedes an ordinary probe already in flight, but startProbe
carried the ordinary profile's missedProbes across the switch. Relay uses 2
misses for both profiles, so one earlier 4s miss plus a single slow 2s answer
terminated the session -- consuming the tolerated cold-radio answer the urgent
profile exists to provide. Switching profile now resets the count.
|
||
|
Jinwoo Hong
|
33af0af4ea |
fix(relay): stop rejecting the near region on a cold first probe (#19233)
* fix(relay): stop rejecting the near region on a cold first probe The region probe counted the process's first /health request, which pays TCP and TLS setup, as a latency sample. The resulting spread rejected the near region on essentially every cold run, leaving the far region as the sole survivor and pinning US desktops to Asia cells for 24 hours. Discard a warm-up probe per origin, compare regions by minimum latency, and keep the spread check only for a genuinely flapping path. A region now wins only against a measured competitor; a rejected or unmeasurable peer sends no hint, which is remembered for an hour so a reconnect does not re-probe. An origin that fails its warm-up is dropped before the sampling rounds, so an unreachable region costs one probe timeout instead of four. After a control socket registers, probe the cell we landed on once per process, and delete the cache only when it names a region other than the best measured one and that cell is more than 3x slower -- a far cell under a correct cache is the director declining the hint, and re-measuring would return the same answer. * test(relay): audit the region probe's global fetch call site |
||
|
Brennan BensonandMerge Sim
|
a87a19c996 |
fix(terminal): remount a pane left unbound by a spawn that returned no PTY id (#19223)
* fix(terminal): remount a pane left unbound by a spawn that returned no PTY id A restored-PTY reattach that resolves without a PTY id leaves the pane mounted with no transport binding, so registerData never runs. Main keeps pushing pty:data for the id; the dispatcher finds no handler and parks the bytes in the pre-handler buffer, which claims no delivery credit and so ACKs them anyway — main's flow control reads healthy while the pane shows its last frame forever. The visibility reconciler skips unbound panes, so nothing rebinds one until the user remounts the tab. Every startFreshColdRestoreAgentResume call site is floating (no await, no catch) and startFreshSpawn resolves null rather than rejecting, so no caller could see the failure. Settle it at the completion hook they all funnel through, whose guards already mean "no pty, pane alive, still unbound, nobody else spawning" — it settled the direct-SSH lease there and did nothing for local panes. Route those to the existing remount seam instead; the SSH retry ledger keeps ownership so the two never race. Observed in the field on three concurrent panes, each parked just past the 64KB pre-handler warn threshold. * test: drop a mock-only assertion that failed typecheck --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Brennan BensonandMerge Sim
|
da4da8e60a |
fix(agent-status): stop a stale self-authored agent title from faking a pending question (#19237)
* fix(agent-status): stop a stale self-authored agent title from faking a pending question A workspace card could show the amber "agent is asking you something" icon while every pane sat idle and its only agent row read `done`. Orca injects its own `<Agent> - action required` OSC title when a hook reports blocked/waiting, then classifies that same title back as evidence. Two gaps let that one-shot string outlive the state it described: - The pane-id sets that suppress the title heuristic were built only from FRESH rows, so once a row aged past AGENT_STATUS_STALE_AFTER_MS the pane stopped suppressing its own title and `permission` — which outranks `done` — decided the indicator. Pane identity is not a liveness fact, so it is now tracked separately and never expires. Stale rows suppress `permission` only; a working spinner re-renders, so its stale-row fallback is preserved. - The hook-driven tab-title write compared the resolved title against the pane's layout slot (`titlesByLeafId`, which only a mounted pane updates) while writing `tab.title`. Once those slots diverged, `done` resolved to a title equal to the pane slot, the no-op guard skipped the write, and the tab kept the stale label. The guard now compares against the slot it actually overwrites. All three status surfaces read the same suppression inputs, so all three showed it: the workspace card, the terminal tab glyph, and the cmd-J palette dot. Tests: each fix has a regression test that fails without it (the sidebar, tab-bar and palette tests all go red from a single ablation of the stale-set lookup). * fix(agent-status): preserve native permissions and cover palette fallbacks --------- Co-authored-by: Merge Sim <sim@local> |
||
|
Brennan BensonandMerge Sim
|
0252fe5c36 |
feat(native-chat): show Codex subagent activity instead of opcode rows (#18773)
* feat(native-chat): show Codex subagent activity instead of opcode rows
Codex spawns subagents and reports their lifecycle, but Orca rendered only
gray `codex · item:subAgentActivity` opcode rows. Build the real display: one
summary row per spawn group with a live working count and token usage.
State is accumulated from `subAgentActivity.kind` alone. A live probe against
app-server 0.152.1 showed `agentsStates` arrives empty even in a real subagent
run, and that every activity item is delivered twice (item/started and
item/completed), so every transition is idempotent and terminal states latch.
Children never receive `thread/started`, so there is no nickname, role, or
depth to read; the row labels from the trailing segment of `agentPath`.
Two sweeps keep a row from claiming work forever: the parent turn's terminal
event settles still-running children, and session start marks a pre-restart
roster unverifiable rather than exited, since Codex resume replays no
non-message items and no event can ever settle them.
The roster rides a new NativeChatBlock variant paired with a plain-text twin.
A journal item kind could not be used: that union is closed, and an unknown
kind parses as malformed, which is the corrupt-journal class that can hide the
chat tab. Block types are explicitly admissible when unknown, so an older
client drops the block and renders the sentence.
MessageRow moves out of NativeChatMessageList to keep both files under the
max-lines budget without a disable.
* feat(native-chat): give the subagent summary row its bot glyph
The row led with a glyph that swapped on state — a check once every child
completed, a group icon otherwise — so a group appeared to change identity
the moment it settled. Per the approved mock, the glyph names the category
and never moves: state is carried by the status dot and the tone of the
words beside it.
Use lucide `bot`, the same glyph the individual `subAgentActivity` rows take
in the eight-category vocabulary, so the summary reads as their parent. Slot
and glyph are the mock's 16px/14px, muted by default, and the svg is
`aria-hidden` — the headline is what a screen reader announces, so the icon
never stands alone.
* fix(native-chat): correct the Codex subagent roster's build, journal write, and failure reporting
* Restore the exhaustive block handling that adding `subagent-group` to
`NativeChatBlock` broke. `formatWorkerTranscriptMessage` and `boundBlock`
both fell through to `image-ref` field access, so `tsc -p` failed for the
CLI and node projects and `build:cli` could not emit. Both now guard on
`image-ref` explicitly and give the roster block its own branch.
* Stop the roster's publish from evicting its own append. The sink queue
coalesces by `coalescingKey` alone with no op-kind check, so passing the
append's key to `tryPublish` spliced the queued append out and the row
never reached the journal — permanently, since `lastSerialized` was
already set. `tryPublish()` now takes no argument, matching every other
call site. The regression test's fake sink honours the key, which the
previous fake did not.
* Keep `collabAgentToolCall` substantive. Only the MultiAgentV2 path emits
`subAgentActivity`, so a V1 turn has no roster row; suppressing its collab
tool calls too would have left a V1 fan-out showing nothing at all.
* Surface a settled failure while siblings still work. The summary now
reports the worst adverse outcome independently of the group verdict, so
the row shows `3 working +1 failed` with a failed-coloured dot instead of
a neutral pulsing dot. The plain-text twin names it too.
* Treat `/morpheus` as a child. Only `/root` is the turn itself; the old
segment-count test silently dropped a valid single-segment agent.
* Refresh token-usage recency on update so an active thread is not evicted
as the oldest entry, and scope the `agentsStates` comment to the V2 path.
* fix(native-chat): stop the subagent roster announcing a new duration every second
The roster row is an `aria-live="polite"` region and it contains the elapsed
clock, which reticks once a second for as long as the fan-out runs. A screen
reader therefore reads out a fresh duration every second, burying the state
changes the live region exists to report — the headline, the verdict, and the
`+1 failed` alert.
No other live region in the transcript does this. `NativeChatToolRun`'s live
button holds only the active tool label, and in `NativeChatWorkingStatus` the
variant that shows a duration is precisely the one with no `aria-live`.
Hide the clock from the accessibility tree only while it is moving. Once the
group settles the duration is fixed, so it stays readable and costs no
announcements.
* fix(native-chat): retry a refused roster publish, and stop two wrong readings
Four defects from a third review pass over the Codex subagent roster.
`write()` set `lastSerialized` before the append and rolled it back only when
the APPEND was refused. A refused PUBLISH left it set, so an identical replay
short-circuited and the revision was never published again. The repo's own
pattern is the opposite: `codex-structured-item-streams.ts` advances
`checkpointLengths` only once the append AND the publish are both accepted.
Roll back on either half.
That alone did not cover the sweep, which is the LAST event a group ever gets:
its `changed` guard skips the write on a retry because every child has already
latched, stranding the settled roster's final revision. Write when the previous
attempt was refused part-way, too.
`formatWorkerTranscriptMessage` read `block.agents` as its exhaustive fallback.
The journal schema deliberately admits block types this build does not know and
`client.call` casts the RPC result instead of validating it, so a newer remote
host's block reached that line and threw `agents is not iterable`, taking down
the whole `worker read`. It printed a harmless `[image omitted]` before. Match
`subagent-group` explicitly and degrade the unknown case.
The elapsed clock measured to `now` whenever no child carried a terminal
timestamp. That is exactly the roster restored from the journal after the host
died: the reconciler latches `unverifiable` without a `settledAt`, so a child
that ran four seconds reported the time since the crash as its run length, on a
row that is not even counting. Show no duration when none is known.
Also restores package.json to origin/main: the merge had deleted one of main's
two duplicate `bench:terminal-partial-escape-tail` keys. Behaviour-preserving
(JSON is last-wins and the deleted line was the dead one), but unrelated to this
PR and better left to its own change. No gate rejects duplicate JSON keys.
The new refusal tests also cover the append-side rollback, which had none.
* fix(native-chat): stop the subagent roster vanishing from every settled turn
`NativeChatToolRun` bailed out for a completed turn whose activity disclosure
is collapsed before it reached the branch that draws a roster-only run. That
guard exists to push TOOL activity behind the turn-status disclosure, and it
fires on exactly the shape a spawn group has: a roster message carries no tool
blocks, so `selectActiveToolCall` returns null and `isSettled` is true, while
the list passes `expandOverride={expandedTurnIds.has(turnKey)}` — false until
the reader opens that turn — and `activeTurnIsWorking={false}`.
That is the default state of every finished turn in the transcript, so the one
compact row this feature exists to leave behind ("Ran 3 subagents") disappeared
the moment its turn ended. Worse, `MessageRow` counts a spawn group as
renderable specifically so the row survives, then rendered a wrapper around a
component that returned null — the empty ghost bubble its own guard is written
to prevent.
Order the roster branch before the disclosure guard. A roster has no tool
activity to hide, and the guard's reasoning ("a failed child command looked
like the whole response was still running") does not reach it. Runs that do
carry tool blocks still fall through to the guard unchanged, and in practice a
roster never shares a message with them: it is its own `role: 'system'` journal
row and `isToolOnlyMessage` is false for it, so `foldToolMessages` never merges
tool blocks into it.
Also drop childless groups when building the rows, so `subagentRows.length`
stays an honest test of "something will draw" — the roster-only branch returns
a margin-bearing wrapper on the strength of it, and a group with no children
renders null.
Both tests fail with their fix reverted; the existing NativeChatToolRun suite
still passes, so the completed-turn disclosure behaviour is unchanged.
* test(native-chat): cover the subagent roster at the message-list level
Every defect this feature has shipped so far lived in the assembly between
rows, and the row-level suites kept passing through all of them. Loop 4's
regression — a settled roster swallowed by the completed-turn disclosure —
was found by reading the code, not by a test, and an independent visual-proof
run observed the same symptom in the real UI and routed around it rather than
reporting it. `NativeChatToolRun` rendered alone is handed `expandOverride`
and `activeTurnIsWorking` by the test author, so it agrees with whatever the
caller was assumed to pass.
Drive the real component instead. The roster is its own `role: 'system'`
journal row carrying the producer's two blocks (structured + plain-text twin),
so what reaches the DOM depends on `foldToolMessages`, the turn-key mapping
and the disclosure state `NativeChatMessageList` owns — none of which a row
test exercises.
Three cases, on one assembled transcript that holds tool calls AND a roster:
- a settled turn with activity collapsed, the resting state of the whole
transcript, still shows the row (fails with loop 4's reorder reverted);
- tool activity stays behind that disclosure and appears only on expand,
and expanding draws no second roster (fails with the guard removed);
- a working turn reads as a live spawn.
The first also pins that the plain-text twin is dropped rather than printed
beside the row it stands in for.
Timestamps are explicit and ascending: the list re-sorts by (timestamp, id),
so rows sharing a millisecond tie-break alphabetically and the user turn can
sort last, stranding the roster outside its own turn and reconciling live
children to `unverifiable`.
No production code changed.
* fix(native-chat): make "counts as renderable" and "actually draws" agree for a spawn group
`MessageRow` counts any `subagent-group` block as renderable, but
`NativeChatSubagentRun` renders null for a childless roster. A group with
`agents: []` therefore mounted a row that drew nothing — an empty div that still
costs the transcript one `gap-5` slot. The Codex producer never writes one (every
`write()` call site operates on a group that already holds an entry), but the
block schema admits `agents: []` with no `.min(1)`, and the wire is where such a
shape would arrive.
Narrow `subagentGroupBlocks` — whose only production caller IS that renderable
check — to the groups that will draw, behind a named `isRenderableSubagentGroup`
that `NativeChatToolRun` now shares in place of its own copy of the predicate, so
the two guards cannot drift apart again. A childless group carrying its
plain-text twin now prints the twin, which is what the twin is for; a bare one
skips the row entirely.
Also correct four comments that had stopped describing the code:
- the roster header called `agentsStates` "always empty", contradicting the
probe note in `codex-subagent-activity.ts` — it is empty on the MultiAgentV2
path that emits these items, and the V1 path does populate it;
- `tokensByThread` was documented "retained UNCONDITIONALLY" while
`handleTokenUsage` LRU-caps it 65 lines below;
- the sweep is not "the LAST event a group ever gets": neither `settleTurn` nor
`settleSession` removes the group, so a later `thread/tokenUsage/updated`
naming a swept child still writes it. The retry condition is right; only its
stated reason was wrong;
- the `subAgentActivity` classification is not reached "for every event — and
every one of them arrives twice". `handleSubagentItem` intercepts those items
before `items.handle`, so the live path never consults the catalog;
`restoreThread` replays them straight through, and is the real consumer.
Comment-only apart from the childless-group guard.
* fix(cli): stop `worker read` printing the subagent roster sentence twice
The producer ALWAYS writes a roster block beside a plain-text twin carrying the
same sentence, for clients that cannot draw the block. The renderer honours that
contract from one side — it draws the block and drops the twin. The CLI honoured
neither side: it printed the twin as prose AND rendered the block as
`[subagents] <same sentence>`, so a real roster message read
[system] Ran 2 subagents (1 failed)
[subagents] Ran 2 subagents (1 failed)
Take the mirror of the renderer's rule, which is the cleaner half for a text
client: the twin IS the sentence, so print it and drop the block it stands in
for. A block that arrives WITHOUT its twin — a shape the wire admits and no
producer writes — still stands in for itself, because dropping it
unconditionally would lose the roster entirely. Either way the sentence prints
exactly once, off the same `subagentGroupFallbackText` helper both sides use.
Unreachable through `readWorkerTranscript` today, whose provider rollout decoder
never emits a `subagent-group` block — but the formatter is the CLI's contract
for any transcript source, and the shape is already producible.
The test pinned a TWIN-LESS group, a body `codexSubagentGroupBody` never writes:
it asserted the exact double-print this fixes was correct output, and would have
blessed either behaviour. Rebuild the fixture as the producer's real two-block
row, with the sentence taken from the shared helper rather than hardcoded so it
cannot drift, and assert the sentence appears exactly once. The twin-less shape
keeps a test of its own, labelled as the wire-only fallback it is.
Also record why `settleTurn` keys on the RAW `turnId` while `groupFor` remaps
off-primary activity onto the primary's active turn. The asymmetry is
load-bearing, not an oversight: were `settleTurn` to remap, a child thread
ending its own turn would sweep the parent group and settle every still-working
sibling to `unverifiable`. The lookup missing is the intended no-op.
* fix(native-chat): add the subagent roster's localization keys and narrow its twin filters
The roster row called 16 `components.native-chat.subagents.*` keys that were
never added to the catalog, failing the localization gate. Synced en.json; the
English strings are the component's own inline fallbacks, so nothing renders
differently.
Also tightens the twin/block handoff on both readers. The renderer dropped
every text block once a roster was present, which is safe only because Codex
writes a roster as its own message — the block is provider-agnostic, so a lane
folding prose in beside one would have lost it on desktop while mobile kept it.
And both readers decided "the twin is already printing" by recomputing the
sentence and comparing bytes, which a roster from a newer build never matches:
its unknown state normalizes to `unverifiable` here, so the CLI printed the
roster twice with two different verdicts. Both now recognize a twin by shape.
* test(native-chat): pin the roster twin recognizer against prose
Both readers use it to decide the twin is already printing, so a false positive
eats a message's real prose and a false negative prints the roster twice.
* docs(codex): restore the roster's evictionated trigger to its KNOWN LIMITATION
The previous rewrite dropped both triggers the old comment named and kept only
the restart one, but eviction is the reachable half: `groupFor` caps `groups` at
MAX_CODEX_SUBAGENT_GROUPS and drops the oldest-INSERTED entry (it returns an
existing group without re-inserting, so this is not LRU), which can evict a
still-live group in-process. The row identity is keyed on the group id alone, so
the next activity item rebuilds that row from one child — the same N-to-1
rewrite, with no restart, and with the sweep skipped so the children never latch
`unverifiable`. Also softens "every real turn id is freshly minted" to the
provider assumption it is: turn ids are read verbatim off provider frames and
nothing in this repo mints or asserts them.
* docs(codex): justify the subagent wire notes from the live probe alone
The roster and disposition comments explained themselves in terms of a
provider-internal path taxonomy rather than anything this repo can observe.
Restate them from the evidence Orca actually has: the live app-server probe
saw `agentsStates` arrive empty, so nothing reads it; and `collabAgentToolCall`
stays substantive because nothing guarantees a session reports subagent work as
`subAgentActivity` at all — one that only emits the collab tool call gets no
roster row, and suppressing that too would leave its fan-out blank.
Same behaviour, same tests; comments and one test name only.
* fix(native-chat): stop the roster's durable twin from claiming live subagents
The spawn-group row is written once and revised in place, but the row itself
is durable and replayed on every reconnect. Its plain-text twin — the only
thing a client that cannot draw the block ever sees — froze a live count into
that row: `Kicked off 4 subagents — 2 working`. The desktop renderer never
shows it, and reconciles the block's `working` to `unverifiable` outside the
live turn. A text-only reader does neither. When the writing process dies
mid-flight the turn-end sweep never runs, so the sentence keeps asserting two
running children forever, with nothing left that could re-check them. That is
the collapse `docs/reference/ssh-execution-boundary.md` forbids: loss of
contact reported as a live state.
Fix it at the source rather than per client: the durable sentence now states
only what survives its process — that the group was spawned, plus whatever
outcome had latched. `Kicked off` vs `Ran` stays, because it reports whether an
outcome was recorded at write time; saying `Ran` while children were in flight
would assert they exited, the same error inverted. The adverse count stays so a
failing fan-out still reads as failing. Reconciliation stays in the renderer,
where the block still needs it.
The twin recognizer keeps matching the legacy `— N working` shape: journals
already hold those sentences and their rows replay forever, so dropping the
branch would print every one of them twice, once as the block and once as prose
the reader meant to drop.
Also align the two functions that read `agentPath`. The root check compared the
raw string while the label normalized separators, so `/root/` was both the turn
itself and a child of it — a phantom row labelled `root` inflating the group by
one. Compare normalized segments instead, keeping `/morpheus` a child. And a
trailing segment with nothing visible in it survives the empty-segment filter
and would draw a nameless row, so it now reads as no label and falls back to the
placeholder.
* fix(codex): key the subagent label collision ordinal on what the row draws
`codexSubagentLabel` tested the trailing segment trimmed but returned it
untrimmed, and `claimLabel` keys its collision ordinal on that string. Two
children at `/root/read` and `/root/ read ` therefore both drew as `read`
with no ordinal — the one thing the ordinal exists to prevent. Return the
trimmed segment so labels that render identically collide.
Also correct the legacy-clause note on the twin recognizer. It claimed shipped
journals hold the old `— N working` sentence; the feature is unreleased, so the
only journals holding one are dev worktrees of this branch. The branch still
earns its place — those rows replay too, and it adds no false-positive surface
the bare shape does not already carry — but the stated reason was wrong.
* test(native-chat): retire the subagent-visibility guards now the roster renders
Two tests from the sibling item-coverage PR asserted that subagent items stay
on the generic gray row, explicitly gated on "until a real renderer exists".
This branch is that renderer, so both guards fire on merge — the handoff they
were written to mark rather than a regression.
They now pin the other side of it: subAgentActivity is suppressed because the
spawn-group roster renders it, and collabAgentToolCall deliberately stays
visible, since nothing guarantees a session reports subagent work as
subAgentActivity at all.
Git merged both files without conflict; only running the suite surfaced this.
* fix(native-chat): let a subagent swept at turn end still report what it did
The turn-end sweep marks still-running children `unverifiable`, and the
producer latched on any state that was not `working` — so `unverifiable`
latched too. A subagent that outlived its turn then reported `completed`, the
latch refused it, and a child that finished successfully read as one we never
saw finish, permanently.
One predicate was doing two jobs. `isTerminalSubagentState` is right for
counting — `unverifiable` is not working — and wrong for latching, because
`unverifiable` records that we stopped being able to see the child, not what
it did. Split them: a child's own verdict latches, the sweep's guess does not.
The reverse stays refused. Nothing returns to `working` once we have given up
on it, so a straggler progress tick cannot re-light a settled row.
Neither the latch nor the sweep was wrong alone, and both were tested; the
defect lived only in their interaction, and only when a subagent outlives its
turn — which the probe that drove this design never produced, because the
parent it captured waited on its child.
* fix: drop the @pnpm/exe lockfile drift a merge staged
`git add -A` swept up the pnpm-lock.yaml mutation that every pnpm invocation
leaves in this repo. Nineteen lines, thirteen of them @pnpm/exe, and it fails
sixteen unrelated CI checks — native smoke, typecheck, packaging, xterm patch
sync — none of which name the lockfile.
* fix(native-chat): restore the item fall-through an inline dropped
Inlining the subagent routing helper lost its null check: the roster returning
null means it did not claim the item, and the translator must keep looking.
Returning unconditionally once any thread item parsed swallowed every ordinary
item — twelve settlement tests, none of them about subagents.
* fix(orchestration): rebind the subagent block arm to the renamed bound state
Main renamed clipMetadata's second parameter from a warnings set to a
TranscriptBoundState. The subagent-group arm still passed `warnings`, and git
merged both sides without a conflict because the lines never overlapped — the
rename and the new arm are in different hunks. Typecheck was the only thing
that could catch it, and did.
* fix(codex): publish the turn tail for a subagent item the roster claims
Main's #19055 added a `subAgentActivity` arm to the provider activity table,
which is reached only through `publishActivity`. The roster's admission returned
above that call, so every `subAgentActivity` item bypassed it and a fan-out that
reports nothing else left the turn tail stuck on the previous frame's text.
`publishActivity` already no-ops on a refused admission and on a non-primary
thread, so routing the roster's admission through it is safe.
Also corrects a docstring the frames extraction copy-pasted onto
`settleOversizedNotification`.
* fix(native-chat): bound the subagent roster on every boundary that carries it
The spawn-group arm was the one collection in the worker-transcript payload with
no cap, and the one block type mobile's `sanitizeBlock` forwarded verbatim. The
producer's `MAX_CODEX_SUBAGENTS_PER_GROUP` does not reach either boundary: the
journal schema declares no maximum on `agents`, and a remote host may run a build
with a different cap. Both transports now cap the roster and bound `id`, `label`
and the open `state` string; `label` and `id` also take the standard inline bound
on the journal write path, where every other provider string already does.
A token count is now persisted onto its entry at write time. `write` rebuilt
`tokens` from the LRU-capped thread map on every write, so an eviction silently
retracted a count the durable row had already shown.
Adds the first coverage of the three roster caps, including the group eviction
that rewrites a row from N children down to one.
* fix(native-chat): keep the roster drawn beside tool calls and its clock honest
The roster-only escape is keyed on `blocks.length === 0`, so a spawn group
sharing its message with tool-call blocks fell through to the settled-turn guard,
which returned bare null and took the roster with it — the exact regression the
escape above was written to avoid, after the message row had already counted the
group as renderable. Unreachable for Codex today; the block type is deliberately
provider-agnostic, so it is live for the Claude lane.
The elapsed clock also froze at a sibling's timestamp on a partial sweep: in a
group where one child completed and another is unaccounted for, the ended turn
left `working === 0` with the completed child's `settledAt`, and the row showed
that child's duration as the group's run length. No clock is drawn while any
child is `unverifiable` with no terminal timestamp.
* perf(native-chat): bound the roster's provider strings without digesting them
`boundInlineText` computes a sha256 and a Buffer BEFORE it checks the length,
so the roster paid two digests per child on every write even when nothing was
truncated — and `write()` runs on every claimed activity item (each delivered
twice) and again from `handleTokenUsage`, which streams. A same-process A/B over
a 64-child group: 76.5 us/write before, 2.0 us/write after (plain, unbounded row
is 1.2 us).
The cap changes with the mechanism. 16 KB is the tool-output bound; both readers
of this row already clip the same fields to 512, so the producer was admitting
~2 MB per durable roster row for consumers to throw ~97% of away. One
`MAX_SUBAGENT_FIELD_CHARS` now serves the producer and both readers, and the
marker is an ellipsis rather than the tool-output truncation sentence — `id` is
the roster key and the renderer's React key.
Also raises the orchestration arm's per-group bound from 20 to the producer's
64, matching the mobile arm: a 21-64 child group is routinely producible here,
so that arm clipped children and warned while its sibling clipped none. The
slice and warning stay as the transport's own defence against a remote host with
a larger cap.
* fix(orchestration): suppress one roster block per twin, not all of them
`hasTwin` was a single boolean over the whole message, so a message carrying two
`subagent-group` blocks and one plain-text twin printed one sentence and dropped
the second roster with no marker. Count the twins and claim one per group
instead. Not reachable from this branch's producer, which writes one group per
journal item, but the surrounding reasoning is explicitly about wire shapes the
producer never writes and this is the adjacent one it missed.
* fix(native-chat): loop-3 fixes to the Codex subagent worklog
Five defects loop 2's own fixes introduced.
Twin claiming was order-blind: the count-based claim silenced whichever
roster block came first, so a lone twin belonging to a LATER group erased
an earlier group's roster and printed the later sentence twice. Exact-text
claims are now settled for every group before any leftover twin is claimed
by position; the positional fallback stays for a newer build's frozen twin,
which can never equal a recomputed sentence.
`boundSubagentField` sliced UTF-16 units and could leave a lone high
surrogate in a durable row, and the clip removed exactly the tail that told
two children apart — `id` is the renderer's React key and `claimLabel`
writes its repeat ordinal at the end. It now backs off a split pair and
reserves the child index inside the bound, so both readers' re-clip cannot
cut the disambiguator off again.
`MAX_SUBAGENT_FIELD_CHARS`'s doc claimed a `groupId` bound the producer
never applies; the doc now says so and why. The worker-transcript metadata
cap is a separate literal again: it governs message ids, turn ids, tool-call
names and image urls, so a roster-motivated change must not move it.
* fix(native-chat): never infer a lost subagent from a turn boundary
QA drove a real Codex session with three live `spawn_agent` children and sent
a mid-turn correction. The roster row immediately read "Ran 3 subagents /
3 unverifiable" with no clock, while all three were still running — they
reported `completed` 57-87s after that turn ended.
Both sites rested on the same false premise: that a turn ending means no
event will ever settle a child. Children outlive their turn and keep
reporting into the same group.
- Renderer: drop `reconcileSubagentRoster`. Nothing plumbed to the component
distinguishes a row written by a dead host from a turn that merely ended —
journal render items carry no epoch, and a new epoch deletes the rows of the
one it supersedes — so the row now draws the state the journal recorded.
Under-claiming beats over-claiming.
- Main: stop sweeping on `turn/completed`. That sweep wrote `unverifiable`
into the DURABLE journal, which mobile reads with no reconciliation.
`turn/completed` is Codex's only turn-end notification, so an abort cannot
be told apart from a clean finish; the safe default is not to sweep.
`settleSession` — the provider actually being gone — is unchanged and is now
the only sweep. `unverifiable` stays non-latching so a late verdict still lands.
* test(native-chat): pin the roster at the seam the QA defect came from
The mid-turn correction opens a new turn, so the fan-out's row stops being
the current turn and the list hands the roster `activeTurnIsWorking={false}`.
Asserted through the list, not the component, because that prop is what
carried the wrong claim.
* fix(native-chat): settle a roster the dying host never got to sweep
`settleSession` only fires when the provider goes away while this process is
alive. If the host itself dies, nothing sweeps and nothing reconciles on
restore, so a `subagent-group` row persisted as `working` claimed live children
forever — the mirror of the defect the previous commit fixed, and the same
`ssh-execution-boundary.md` violation in the other direction.
Reconciled host-side, at journal open, not in the renderer: mobile shows only
the durable text twin and reconciles nothing, so a renderer-only fix would
leave it claiming live children indefinitely. Opening the journal is also the
one moment a host can honestly say the previous writer is gone.
- `staleSubagentRosterRevisions` rewrites every child still reading `working`
to `unverifiable` and regenerates the twin from the same summary, so the
block and the sentence cannot disagree.
- No terminal timestamp: the child stopped being observable at an unknown
moment, and stamping the reopen would report the downtime as its run length.
- Revises in place under the parsed identity, so a reopen upserts the row
rather than appending a duplicate, and a second reopen writes nothing.
- Skipped on a corrupt load: that journal is still owed a rebuild from provider
history, and content past the repair's free sequence retires the demand.
Reconciles journal ROWS, not roster state — the producer's in-process group map
is untouched, so the roster's known seeding limitation is unchanged, as is
`canReplaceSubagentState`: `unverifiable` still does not latch.
---------
Co-authored-by: Merge Sim <sim@local>
|
||
|
Brennan BensonandMerge Sim
|
bffdad9f05 |
fix(native-chat): make structured chat tabs renameable (#19153)
* fix(native-chat): let a structured chat tab be renamed Renaming a native chat tab accepted the text and silently did nothing: setTabCustomTitle only scanned terminal tabs and only bridged to unified tabs whose contentType was 'terminal', so the agent-session tab it was keyed to never matched. Any label that did land was then re-nulled by the next host snapshot, which preserved color/createdAt/isPinned but not customLabel. Also routes both placeholder sites through one helper so a Claude chat stops falling back to 'Codex Chat'. * test(native-chat): cover structured chat tab rename and label fallback * chore: drop the local @pnpm/exe lockfile artifact Swept in accidentally; running pnpm here adds @pnpm/exe to the root lockfile, which fails CI's frozen-lockfile guard. * fix(native-chat): reach the rename shortcut and tab color too Review found the first fix covered only the context-menu path. The tab.rename shortcut gated on activeTabType === 'terminal', so on a structured chat tab it stayed the silent no-op this branch set out to fix. setTabColor carried the identical terminal-only lookup one function below the one that was fixed. Both lookups now share one resolver instead of two copies. * fix(native-chat): stop unknown agents reading as Codex, cover the terminal path Review found the placeholder helper encoded "unknown means Codex": its signature accepts null/undefined and Tab.agentSessionAgent is the open AgentType, so the first caller passing a Tab would label gemini or grok as "Codex Chat". Routed through the shared agent-name table instead. Also adds the missing regression test that a terminal rename still resolves through its entityId now that both rename and color share one resolver, and a guard on a test that passed with the fix reverted. * fix(native-chat): degrade instead of throwing on a null tab title A stacked branch can publish title: null when a conversation name is cleared. The wire type says string, so this consumer trusted it and threw inside the store patch that applies the snapshot. Fall back to the placeholder — the producer bug is fixed separately, but a consumer of wire data should not crash on a contract violation. * fix(native-chat): rename the focused structured tab, not a background terminal * fix(native-chat): cycle terminals from the structured tab, not a stale terminal --------- Co-authored-by: Merge Sim <sim@local> |