* refactor(main): organize agent session and tab lifecycle owners
* fix(ci): carry moved backend test and mobile routing metadata
* test: retire copy assertions in headless notification test
* feat(agent): one agent picker card with agent tabs on top
Dropdown pickers show agent icon tabs, then search, the agent's models with More models, one effort and Fast row, and an Open in Chat | CLI switch. The + menu opens the same card without tabs, followed by the prompt and launch buttons. Settings and automation forms keep Permissions and Arguments as fields below the picker.
* fix(agent): address picker review findings
- Release the dropdown pane's picked setup when it closes, so a reopened
picker starts from what the Permissions/Arguments fields saved.
- Skip the trigger summary with a trigger label and memoize it on the draft.
- Remove Arguments chips from the text being typed.
- List Blank Terminal and other non-agent rows in search and reach them
with ArrowDown.
- A query naming an agent lists the agent first and picks it without a model.
- Agent tabs: roving tabindex, arrow keys, Enter/Space, Shift+Tab from search.
- Modern automations show one Arguments field.
* fix(agent): keep automation extra arguments supplemental and retranslate the trigger summary
- Restore the automation editor's supplemental Extra agent arguments field as main had it,
so saved extras stay on top of inherited host arguments.
- Recompute the memoized trigger summary when the UI language changes.
* fix(agent): keep the picker's effort and Fast row on one line
Raise the agent dropdown's minimum width to 26rem, which fits Codex's widest
measured effort row plus Fast (404px), and cap it at the available width
with 8px collision padding so it still fits narrow windows.
* fix(agent): key the picker summary on UI language without a lint suppression
* fix(agent): name the model in the closed picker trigger
Before a host's listing loads, resolve the trigger's model label from the cached host listing,
then the release's catalog, then Claude's id-based listing name, and the id only last.
* fix(agent): name live-listed models in the closed trigger and fit the text-recipe effort row
- The closed trigger reads the agent's last listing an open picker showed, so models only the
live listing names (Codex gpt-6-astra) keep their label and effort.
- Text actions label their inherit-effort choice Default, with the inherited level as its tooltip,
so the row fits one line.
* fix(agent): name a closed trigger's model from the user's pick, and keep the inherit-effort name
- Replace the per-host listing cache with a bounded record of picked models, written only at
pick time and keyed by agent, host and model; mounted closed triggers re-render when it changes.
- The text action's Default effort keeps "Same as default · <level>" as its accessible name.
* perf(agent): re-render only the pickers showing a newly picked model
Each picker subscribes to its own agent, host and model's picked row; an unchanged write is
skipped, and the subscription is hoisted to module scope.
* fix(agent-picker): address CodeRabbit review threads
- Dropdown pane derives from the controlled value when there is no form source, so a kept setup no longer shows a stale choice
- Model search drops setup rows that repeat a listed model
- Agent-only picker search matches extra choices (e.g. Custom command)
* fix(agent): label the chat/CLI switch "Open as"
The CLI overview called artifact publishing opt-in, contradicting the CLI
reference and Settings docs, which say the gate is on by default for new
profiles while older profiles keep their previous setting. Align the
overview wording and drop the stale opt-in phrasing from the browser docs.
Fixes#43272
Without --worktree, a reuse start resolved the caller's workspace and was refused with terminal_worktree_mismatch whenever the terminal lived elsewhere. An explicit --worktree keeps today's behaviour, including the refusal.
Clicking a legend entry in the Claude, Codex, and Muse daily usage charts
now hides or shows that token type, and the bars rescale to the visible
types across the ten days shown. At least one type always stays visible.
The Claude and Codex charts share a new UsageDailyTokenChart component.
Codex and Muse reasoning tokens are no longer double counted in the daily
bars, and the Output tooltip keeps the full output count.
Fixes#43222
Co-authored-by: Neil <neil@stably.ai>
* feat(cli): list and read agents by their Orca session ID with orca agent list/show/read
Live terminal and chat agents are listed, shown and read by Orca session ID on the selected host; sleeping native agents can be shown and read by ID. Chat reads walk the chat's own bounded history pages back to the latest /clear.
* fix(cli): agent show names a live agent and its last reply as agent list does; help says sleeping terminal agents aren't listed yet
* Bound agent reads and scoped workspace inventories
* fix(cli): agent read carries the conversation's state, so a running chat turn reads as working
agent read now reports the same state agent list/show report, from the same status
rows, in text (a leading state line) and --json (state). A turn still running no
longer reads like a finished chat with no reply.
* Fix agent conversation read consistency and limits
* fix(cli): agent read derives its state from the history it returns
agent read no longer samples live status. A chat's state comes from the
journal snapshot that produced its entries: an open turn or an accepted
send with no turn yet reads working; done needs an explicitly successful
latest turn whose reply is the newest entry shown. A terminal agent's
state comes from the provider turn markers in the same transcript bytes:
a prompt with no end after it reads working, done needs an explicit end
(Claude stop_reason, Codex task_complete) with the reply as the newest
entry. Anything else omits state. The journal lane wait added for the
live state is removed.
* Tighten agent read snapshot evidence
* Keep agent read done for finished turns
A dead owner now only voids working. Codex completion is proved by the turn's own
decoded reply rather than row adjacency, and Claude's by its end row's own reply,
without needing the prompt inside the read window.
* Adapt agent commands to reorganized shared contracts
* Warn on incomplete agent lists, read sleeping transcript paths, and retry unanswered provider id lookups
* Follow main's moved shared modules
* Bound live transcript recovery to conversation panes
* Show sleeping terminal agents where agent list lists them
agent list lists a finished terminal agent while its pane sleeps, since main keeps
that status row. agent show now counts the same sleeping place, and the help says
which sleeping agents are listed. Messaging still ignores sleeping panes.
* Clarify sleeping agent discovery and messaging help
* Say sleeping terminal agents can't be messaged and note the WSL read exception
* Follow the moved CLI argument boundary after rebasing
* Show plain Orca session IDs in agent command help
* Expect plain UUID Orca session IDs in the transcript source test
* Follow main's moved shared modules after rebasing
* Name Orca session IDs <id> in agent help, since older IDs keep their agent prefix
* Follow main's moved agent-session modules after rebasing
Drop the address field's 3px focus ring for a hairline border, size the
toolbar buttons like other pane headers (24px, 14px muted icons), and
replace the Annotate split button with one icon. Grab and Draw move to
the icon's right-click menu and the top of the ⋯ menu.
* feat(workspace-cleanup): delete chat history of deleted workspaces
Adds an off-by-default host setting. When on, each cleanup run deletes agent
transcripts whose Orca workspace folder no longer exists, through AI Vault's
validated delete, and reports the count and space in the run's activity.
* fix(workspace-cleanup): keep chat history of folders Orca never made a workspace
Skip workspace parents that are or hold a project checkout, protect folders
holding a still-listed workspace in any spelling, re-check each folder right
before deleting, batch scope paths, count companion bytes, sweep chats at most
every 6 hours on hourly runs, and refresh the report when the switch changes.
* fix(workspace-cleanup): treat only nested Orca workspace folders as deletable
Look for deleted workspaces only in <root>/<repo> of nested layouts, protect
renamed workspaces' earlier paths, folder workspaces and case variants, use the
recorded cwd only, re-check the parent and the setting before each deletion.
* fix(workspace-cleanup): protect flat workspaces and per-host rows from the chat sweep
Drop parents inside a still-listed workspace, also read this host's own
worktree rows, record a run only when a rule or a due chat sweep has work,
and cache per-folder protection with precomputed matchers.
* fix(workspace-cleanup): find deleted workspace folders by filesystem identity
Walk up from the recorded cwd to the deepest existing folder and require it to
be, by device and inode, an Orca <root>/<repo> folder without .git; key
protected paths the same way, so symlinks, case and Unicode spellings are the
OS's call. A failed history scan no longer fails the run, and an open session
history list refreshes after chats are deleted.
* fix(workspace-cleanup): close the remaining chat sweep gaps
Walk physical ancestors of a symlinked parent, skip a gone folder with a
lookalike sibling (parsers trim cwd), re-read listed paths after the scan,
never touch WSL paths, use bigint file identity, and report the next run
that actually sweeps when only chat history is on.
* fix(workspace-cleanup): keep chats resumed elsewhere and follow dangling listed links
Before deleting, re-read each transcript: skip it if it was written in the last
hour or its last recorded cwd is outside the gone folder (a chat resumed at a
live workspace's root). Listed symlinks protect their targets, and parents are
re-checked against the listed paths read after the scan.
* fix(workspace-cleanup): sweep only Claude chats, whose every message records its cwd
Keep a transcript unless its latest recorded cwd, read back past large blobs,
is still in the gone folder; agents that record cwd once can't prove a chat
wasn't resumed elsewhere. Follow symlinks along listed paths, dangling ones too.
* fix(workspace-cleanup): read only a record's own cwd and check each chat right before removal
Parse whole JSONL records backwards and use their top-level cwd, not tool
arguments; run the setting, folder, idle and last-cwd checks right before each
transcript's removal; resolve relative listed links from their real folder.
* fix(workspace-cleanup): protect subagent transcripts and re-read listed rows per folder
Check a chat's companion transcripts with the same idle and last-cwd rules,
read Orca's listed rows right before each gone folder's deletion, keep '..'
in listed link targets for the OS to resolve, and say Claude Code in the copy.
* fix(workspace-cleanup): fold names fully, keep POSIX repo names, re-stat rows only on change
Fold case with upper then lower (final sigma), take the repo name with this
host's basename, re-read listed rows per folder only when the catalog changed,
and sweep chats even when rule evaluation fails.
* fix(workspace-cleanup): re-check the chat itself after walking its companions
A chat resumed while its subagent transcripts were being checked has a new size
or mtime, so it is kept.
* chore(workspace-cleanup): sync localization catalogs and drop a cast in the cwd reader
* fix(workspace-cleanup): re-check every subagent transcript after the companion walk
Record each companion's version and require the same set, unchanged, before
approving deletion; skip the symlink-based sweep test on Windows.
* fix(workspace-cleanup): protect chats by what listed paths pass through, checked per chat
Each listed path claims the (existing folder, name) pairs along it, so a gone
folder holding one stays protected whether that path exists right now or not;
claims are re-read before every removal, and an unfollowable symlink chain
stops the sweep.
* fix(workspace-cleanup): an unreadable folder or a non-directory leaves the sweep fail-closed or ignored
A stat error other than absence while reading listed paths marks the snapshot
incomplete; ENOTDIR counts as absence instead of stopping every sweep.
* fix(workspace-cleanup): check row protection last before each removal and allow exactly eight links
Run the slow transcript checks first so the folder and listed-row checks are
the freshest before the rm, and resolve the eighth link instead of treating
it as an unfollowable chain.
* chore(workspace-cleanup): follow the shared settings and execution-host module moves
* chore(workspace-cleanup): mark an unreadable identity as undefined, not a string literal
* fix(workspace-cleanup): re-read protection when a listed symlink is retargeted
Listed paths that go through a symlink record where they led; a different
target before a removal re-reads every claim.
* fix(native-chat): fold Claude's image-source row into the prompt it follows
* refactor(native-chat): group desktop image units with the shared folding rule
* fix(mobile): take the send boundary from image-folded rows
* refactor(native-chat): share the image unit prompt lookup
* ci(mobile): refuse a GitHub APK with any Play automatic-protection class
The guard only matched com.pairip.licensecheck. With automatic protection off no com.pairip class belongs in the APK, so match the whole package and catch the check under any other package name.
* ci(mobile): name the pairip code, not just its license check, in the guard's error
* fix(mobile): open the GitHub release the update check found, not Google Play
GitHub installs outside mainland China had Update open Google Play. The check reads GitHub, and Play production is published by hand after GitHub, so Play could lack that version and offer nothing to update.
* fix(mobile): send GitHub installs to GitHub on the update-required screen too
Drops the Play preference from #43080: Play production lags GitHub, so it
can lack the version the desktop needs.
* fix(mobile): keep the keyboard open when the session mic is pressed
On Android, pressing the session's mic while the keyboard was open took
focus from Live input. A tap moved it on the touch's compatibility
mousedown, and a long press focused the mic without any mousedown. The
keyboard closed, and in hold mode the mic's later blur ended React Native
Web's responder, so a held recording stopped about half a second in while
the finger was still down.
Give the input bar the keyboard-persisting taps the key row above it
already uses, and have keepTextFocusThroughTaps also cancel a touch's
pointerdown: then neither a tap nor a long press moves focus, and click
still fires. Mouse pointerdowns are left alone, because RN Web starts mouse
presses on the mousedown a canceled pointerdown suppresses.
On a Motorola razr fold (Android 16) the keyboard stays open through a
tap and a 2.5 s hold, and the held recording runs until release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GR9M8yCkioUvXqiwZKWWiz
* fix(mobile): keep the keyboard open when the chat composer mic is pressed
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Kelvin Amoaba <97001695+AmoabaKelvin@users.noreply.github.com>
* fix(orchestration): name a run's mail by its coordinator
Mail sent from a run's mailbox (run:<id>) was saved with an empty sender party, so its card showed "an agent" with a fallback icon. The sender now resolves to the run's coordinator chat or terminal, under the binding the message was sent from.
* refactor(orchestration): answers from a run reuse the shared run sender
* fix(orchestration): a run's mixed-age mail batch resolves from its oldest message
A Session History row from the user's own OpenCode store sent no account request, so the host fell back to the selected managed account, where OpenCode 2's --session silently creates an empty session with the same id. The resume now names the System store, and daemon spawns keep that choice for the private-server check.
The xterm helper's focusin now rebuilds the macOS input context when focus arrives from any text field, or after the page lost focus (a browser guest or another app) since the last terminal arrival. Refreshes for one helper within a frame share one refocus.
Fixes#23864
Co-authored-by: Fanzzzd <fanzzzd@outlook.com>
* fix(browser): stop refusing agent input while Orca's window is briefly busy
* docs(browser): say which focus refusals are safe to retry
* refactor(browser): drop the focus refusal phase that cannot happen
* fix(watchers): keep home-folder watches out of other apps' macOS data
A folder workspace rooted at the home directory subscribed to all of
~/Library. Parcel stats every changed path, and a stat inside another
app's container or group container makes macOS ask whether Orca may
access data from other apps, again for each write that app makes.
Exclude ~/Library from macOS watch roots that contain it, and only the
Containers and Group Containers stores when ~/Library itself is the root.
Ordinary project roots keep the same options.
* fix(watchers): exclude protected Library stores against the canonical watched root
Review follow-up. Decide the exclusion in the watcher supervisor, after the
root is resolved to the identity Parcel actually subscribes, and resolve
each home's Library the same way so symlinked or case-variant homes match.
Pass the paths as Parcel ignorePaths so a home containing glob characters
is still excluded literally, and protect both HOME and the account's own
home when a host runs with HOME pointed at a separate profile.
Callers go back to their previous options; every watcher gets this through
the supervisor.
* fix(watchers): fall back to no exclusion when home or realpath lookups fail
Read HOME, the account home and realpath inside the guarded lookup, as
root resolution does, so hosts or harnesses without them keep the
previous watch options.
* fix(watchers): decide Library exclusions in the watcher process
Resolve the protected Library paths where Parcel subscribes, in the
watcher child, so the realpath lookups stay off Electron main and the
supervisor is unchanged. Tests write fixture revisions instead of
wall-clock values.
* fix(watchers): apply Library exclusions where the watcher receives a subscribe
Keeps the existing Parcel options cast untouched.
* fix(watchers): count only real parent steps as outside the watched root
A home named like '..profile' inside the watched root is still a
descendant, so its Library must be excluded too.
* test(watchers): make simulated macOS paths portable
Tools like Expo, Claude and Node's spawn run BROWSER as a program, so the
"orca open-url --url %s" template failed with ENOENT. Point BROWSER at the
Orca CLI launcher this host already puts on PATH, and let the CLI treat a
lone http(s) URL as open-url.
Co-authored-by: fancive <10972140+fancive@users.noreply.github.com>
The copy took --standalone from the opencode2 row's launch command whenever the
history row said opencode2. That label never tells which OpenCode the opencode
binary is: the chat tab route always named opencode2, and OpenCode 2 copies 1.x
sessions into session_v2. OpenCode 1 rejects the flag and prints its usage.
The copy now opens with its own CLI's launch settings. prepareOpenCodePtyLaunch
already adds --standalone for OpenCode 2 when a private server is needed.
Fixes#42223
* feat(mobile): add chat text size, font and line spacing settings
Pinching in a chat now saves the size for every chat instead of resetting when you leave.
* fix(mobile): measure the screen on the first chat text read, not at import
On Windows, path.relative() returns an absolute path when the plugin root
and the file are on different drives or UNC roots. The plugin test hooks
and panel-check stack locator only treated '..'-prefixed results as
outside the plugin, so Orca's bundled testing runtime was resolved as a
plugin file and panel issues pointed at Orca frames.
Add a shared pathInsidePluginRoot helper (also backing isInsidePluginRoot)
that rejects empty, '..', '..<sep>' and absolute relative paths, use it at
both call sites, regenerate orca-test-hooks.mjs, and add win32-mocked
regression tests.
Fixes#43195
Co-authored-by: alaya <21375302+AlayaElla@users.noreply.github.com>
Co-authored-by: Neil <neil@stably.ai>
OpenCode /models, /help and similar commands open a dialog the Chat
view cannot show, and later chat sends were typed into it. Flag the
commands that open a TUI screen (OpenCode, Codex pickers, OMP's
in-Terminal commands) and switch the terminal-backed chat to Terminal
once the command is submitted, on desktop and mobile.
* feat(mobile): add a stepped slider control
* feat(mobile): terminal font, size, line spacing and weight settings
Terminal settings get a Font section with a live preview: a font picker
(system fonts plus JetBrains Mono, Fira Code and Geist Mono shipped as
variable woff2), a point-size slider replacing the old percentage presets,
line spacing and weight. Pinch now snaps to whole points. Tablets start at
15 pt, Normal spacing and Regular weight.
The size keeps its storage key as points over 13; the font, spacing and
weight live in a new negotiated page storage key. Both terminal hosts read
them at start and on a new set-font-style message; a style change keeps
the grid and lets the cell-box refit resize the PTY.
* fix(mobile): resume an interrupted desktop page download instead of starting over
The phone held hash-verified page assets only in memory until the whole
build arrived, so a cut link discarded them and every later entry started
the build from byte 0. Each verified asset is now kept under the host's
partial/<buildId>, and the next fetch of that build reuses it.
* fix(mobile): keep nothing for a build with case-aliased asset paths
Each NavBar side shrank to whatever the centred title left, so a long title
pushed a side's second button under the title view. On the leading side the
title paints above the button and swallowed taps, so Hide sidebar did nothing
on wide layouts with long host names. Floor each side at two buttons.
* test(editor): keep a draft whose workspace disappeared while closed
* fix(editor): keep unsaved drafts of workspaces gone at startup and offer Reopen
* fix(editor): settle recovered stranded drafts by identity; keep them over a clean returning tab
* fix(editor): reopen stranded drafts as dirty untitled tabs in the active workspace
* fix(editor): stamp the recovered tab's empty baseline and retry stranded drafts on host changes
* refactor(session): move the editor session builder to its own file
* fix(editor): retry stranded drafts after any dropped trigger, including paired-server SSH changes
* Revert "refactor(session): move the editor session builder to its own file"
This reverts commit 47d3a79eb8.
* refactor(session): move the terminal session builders to their own file
* refactor(session): keep terminal builders in place; move only the tab sanitizer
* fix(editor): don't flag a recovered draft's own placeholder create as a change on disk
* fix(editor): retry stranded drafts when the active workspace's host changes
* fix(accounts): restart busy Claude sessions at once when the account is switched
A restart accepted after a Claude account switch waited for each busy
session's turn to end, so long turns, subagents and background shells
kept spending the old account until it hit its limit. Accepted restarts
now interrupt the session and resume it on the new account right away,
the same as Codex.
* fix(accounts): say in the restart setting that busy sessions stop their work
* fix(accounts): keep each tab's queued Claude restart selection local to that tab
* fix(accounts): run a queued account restart once its pane's terminal attaches
* fix(accounts): wake queued restarts from the leaf bind path too
* test: mark the restart mount-race test's stub casts as safe
* fix(accounts): run the queued-restart wake without a new React hook
* Wake queued account restarts when mounted ownership ends
* Move or report every running Claude terminal on an account switch
A switch only saw panes whose hook status row said Claude, so Agent Teams panes (an `Orca` foreground), panes with hooks off, and panes whose row was dismissed were skipped without a word. The scan now also finds Claude from a saved session record, the pane's own process read, or a one-pane tab's launch; resumes from the saved record when the row is gone; and, when nothing on the outgoing account moves, warns about running sessions Orca can't resume that may still be on it, with Review.
Tracked in STA-15061.
Co-authored-by: Claudio <claudiofernandopinto@gmail.com>
* Keep Claude account restart discovery aligned with saved sessions
* Restart WSL Claude panes only with live evidence
* Keep the original conversation pane during account restart
* Keep retrying account restarts in their conversation pane
* Place the restart fixture cast rationale on its assertion
---------
Co-authored-by: Claudio <claudiofernandopinto@gmail.com>
* fix(native-chat): send other agents' mail as a pointer and let the person approve a denied action
Another agent's orchestration message reached a structured chat's agent as
ordinary user input, so Codex's automatic reviewer counted a coordinator's
words as the user's authorization. The chat card still shows the whole
message, labeled with its sender; the agent is now sent Orca's existing
pointer at `check`, so the content arrives as tool output. Mail stays unread
until `check` takes it, and a new coordinator is pointed at mail the old one
never read.
Codex automatic-review denials now show as a chat row with the reviewer's
reason and an "Approve once" button, which only this computer's own Orca
window may press; it uses Codex's thread/approveGuardianDeniedAction and
records the approval in the chat.
Tracked in STA-15406.
* fix(orchestration): wake any new mailbox owner after re-pointing stale mail
* fix(native-chat): settle check:changed for pointer mail and denial approval
Move the Codex approval and mobile review copy into their own modules, reuse the shared record guard, and update the tests that encoded whole-mail consumption to pointer delivery with mail read at check.
* style: format review notice copy
* fix(native-chat): hold accepted mail while custody is unreadable and show the whole denied action
* fix(orchestration): hold in-doubt mail while custody is unreadable and wake the new owner after a send-time handoff
* test(orchestration): keep crew reads off a chat host another test file left installed
* perf: reuse AI-vault title requests across a sync pass until an input changes
The title pass rebuilt every tab's conversation 2 + 2 x batches times per
20 s pass. Keep one request snapshot per sync and rebuild it only when a
conversation, tab, layout or owner input changed; request timing and
titles are unchanged.
* perf: look up saved titles once per pass and keep only request inputs
Index the request workspaces' saved titles once per pass instead of
searching a workspace's tabs per request, and keep only the store fields
the request list reads so an old snapshot doesn't pin unrelated data.
* fix: read a request's saved title from its own workspace
Two workspaces can hold the same tab id; key the per-pass lookup by
workspace and tab, keeping the first match like the per-workspace search.
* chore: build title request inputs from same-name copies
The layout-writer ratchet reads field-name string lists as computed-key
writes; spell the picked inputs out instead. Also apply formatting.
* chore: move title request inputs to their own file and read workspace records via selectors
* fix(agent-session): resend a chat's first snapshot when its capture is abandoned
A subscriber opened without a cursor gets a snapshot first. When live
authority moved during that read (a hand-off finishing, a start going
live), the capture was abandoned and the following delivery sent batches
from sequence 0 instead, so the client never received the conversation
and stayed empty. The subscriber now owes that snapshot until one is
emitted, and the next delivery sends it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* test(native-chat): pin the owed snapshot after an abandoned capture
The opening snapshot's authority changes mid-read, so its capture is
abandoned; the next delivery must send that snapshot before any other
frame. Without the owed-snapshot check the client's first frame is a reset.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* test(native-chat): a chat store from before revisions were kept keeps its tabs on upgrade
* fix(native-chat): an upgrade no longer reopens closed chats or copies unchanged ones
A chat store copied by a build from before revisions were kept reopened every
closed chat and copied every converted one on its first launch after #42783,
with nothing written by an older Orca.
- The original's receipt is compared by meaning: a receipt saved before WAL
identity was kept matches on the fields it has, so the newer shape alone is
no write.
- A chat with no revision here takes what this file shows as its baseline. A
converted chat is kept as a copy only when the original holds an item it
never saw; one not converted yet converts from the current original. Only a
chat this file doesn't know joins, and only when the original changed.
- So only a chat an older Orca started or positively continued gets a tab;
every other chat keeps its open or closed state.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(native-chat): tell a continued chat by its restored history, not raw rows
An older Orca saves a chat's turns as outcome records, and a rewind can retire
an alias, so a converted chat's raw journal rows are not what it shows. With no
revision to compare, a converted chat is now kept as a copy only when its
restored history, as conversion reads it, has an item the chat here never
held; nothing depends on save timestamps. An original whose identity can't be
certified records its baselines instead of rereading every launch.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(native-chat): an original as copied takes every chat as shown
With no write since the copy, a converted chat this store moved on can't be
mistaken for one an older Orca continued, and the pass restores nothing. The
test of a store from before revisions were kept now checks that its copy saw
the original's WAL, as in the reported profile.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(native-chat): keep deleted empty chats deleted and finish a partial pass
- On a store copied before revisions were kept, every chat with history was
cataloged at the copy, so only one with history is new when this store
doesn't know it; a deleted chat with no history stays deleted.
- A converted chat counts as continued also when the older Orca revised an
item past the revision it holds.
- A pass where one chat failed still records the baselines it read, so only
that chat is retried next launch.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* fix(native-chat): decide unknown chats by the copy's catalog alone
A chat with history always has a catalog row from the copy and one without
history has no revision, so the record lookup added nothing. The downgrade
test's original records its tab index, as every older Orca that saves tabs
does.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* style: format
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>