* Use Ctrl+K for search on Linux and Windows
Preserve existing search and terminal clear shortcuts for legacy profiles until reset.
* Preserve existing Windows and Linux search shortcuts on rollout
* Preserve saved shortcuts in new profiles
* Keep window shortcut tests independent of defaults
* perf(runtime): push worktree row changes; host list poll becomes a 10 s backstop
* fix(runtime): announce title-only agent status changes and gate the rows capability with the switch
* fix(mobile): keep event reads a floor apart while a read is out; announce a retired title clear
* test(runtime): assert the rows notice without counting events
* feat(agent-launch): simplify the new-tab agent launch menu
Drop Permissions and Arguments from the launch-mode setup panel; they are
agent defaults edited in Agent settings. Put effort levels (unlabeled) and
Fast mode on one row, and make the Launch chat/terminal buttons fill the
row at a larger size with icons. Select-mode pickers (automations, new
workspace, source control, continuation) keep both controls.
* fix(agent-launch): let current Agent settings win over launch-menu speed edits
A Fast or speed-tier edit copies the current Permissions and Arguments into the
launch draft. With no Arguments editor in the launch menu, a later settings
change was silently ignored. Launch drafts now drop Arguments stamped by older
settings; the speed choice still applies. Also stop pointing hosts without a
model list at the removed Arguments editor.
* fix(agent-launch): show Fast as a button in the launch menu
Matches the mock: an outlined Fast chip with the lightning icon in place of the
"Fast mode" switch. The chat composer keeps its switch.
* refactor(main): organize runtime implementation owners
* test: retire wall-clock CLI title recovery case
* refactor: route new folder mutation to runtime store owner
* test: import folder mutation store contract from its owner
* feat(chat): /btw asks a side question in a fork beside the chat
In a Chat UI chat that can fork, /btw <question> (or /side) forks the chat
through its newest finished response, sends the question there, and moves the
fork beside the chat the way dragging it does. The source chat is untouched;
the draft clears only once the question went out. Agents that cannot fork keep
their own /btw.
* fix(chat): type-guard the side chat command names
* fix(chat): side chats own the screen from Send and never place an unsent question
- Placement starts at Send, retires for good on any other selection in the
workspace or on leaving it, and looks only at the requesting workspace.
- A question that never went out leaves its fork in the background; an
unconfirmed one says to check the side chat before asking again; a failure
the new chat already reported is not repeated.
- The command menu shows "Available after the first reply" / "when this reply
finishes" while it is open, using the drawn turn order.
- File pills count as attachments; picking /btw completes the row.
* fix(chat): /btw reads the newest response the drawn transcript offers Fork on
The menu reason and Send now follow the transcript's own Fork rule as drawn,
reported when it changes, instead of rebuilding turn membership in the composer.
* fix(chat): side chat failures always reach the source; watch window starts when the fork opens
- A failed side question toasts in the source even when the fork recorded it,
since the fork stays in the background.
- The 60 s placement window starts once the fork exists, not at Send.
- The fork-point scan stops at a /clear and, for tip-only agents, at the
newest turn; split placement looks only in the requesting workspace.
* fix(chat): /btw forgets a fork point the transcript no longer offers
- Send and the menu reason refuse a reported response that is no longer
forkable, and an unmounted transcript reports nothing to fork.
- The fork-point scan skips subagent sections before its early stops.
* perf(chat): /btw steps over a running turn instead of walking its rows
The slot builder groups root responses by turn in its existing pass; the side
chat's fork point reads those groups newest first and skips a running turn
whole, so streamed replies add no per-row work.
* fix(chat): /btw looks past responses outside any turn and skips the scan when nothing is forkable
* perf(chat): find the /btw fork point while laying out the transcript
Per the review ruling: the slot builder keeps the newest row that offers Fork
in its existing pass, only when /btw listens, and returns one point; the
response groups and their second search are gone.
* test(chat): /btw fork point contract through the transcript and Send hooks
* refactor(chat): response rows and /btw share one Fork rule
* test(chat): /btw fork point ignores pending-only renders and an emptied conversation
* test(chat): /btw contract counts every eligibility lookup and covers an undrawn transcript
* fix(chat): /btw shows ahead of agent commands, sends on one Enter, and focuses the side chat
- Side-chat rows lead the slash menu, so a long agent catalog no longer
pushes them past the menu's row limit.
- Enter on a fully typed /btw sends it instead of only completing the row.
- Placing the side chat also moves keyboard focus to its composer.
* fix(chat): picking /btw completes the row again, as the common pattern does
Sending a fully typed /btw from the menu bypassed a same-named skill's
release and sent on a mouse click; a pick now always leaves '/btw ' to
add a question to.
* fix(chat): on layout-owner builds the moved side chat keeps its composer focus
Main selects a tab it moved once its frame shows it there, so placement no
longer selects it first in the source pane; moveTabToNewPaneColumn reports
whether main or this window moved it, as moveTabToGroup does.
* fix(chat): side chat placement owns main-drawn moves through to the drawn result
Per the review ruling: the placement claim stays alive while main draws the
move, reads the side chat's own new pane and selection as its own, and selects
and focuses the side chat once the frame shows it moved, in place when main
refuses, and never after the user selected elsewhere. The owner command takes
an optional completion in place of its own moved-tab selection.
* fix(chat): selecting the side chat in the pane it is leaving still retires the claim
Only what main's move changes elsewhere (a new pane holding only the side
chat, another pane showing it) reads as the request's own; the selection
counter in the placement contract reads every pane's selection, and the move
helpers' completion is exercised through main's frame transport.
* fix(chat): a side chat main never moves is selected where Fork put it
When main neither draws nor answers the move before the watch ends, the
side chat is selected in place rather than left hidden; the owner command's
completion path (no moved-tab intent, hand-back asks first) is tested.
* fix(chat): a side chat this window moves keeps its focus request
The move's own changes count as the request's from before the move starts,
so a synchronous move no longer retires the claim before the side chat is
selected and focused.
* fix(chat): picking an unavailable /btw row does nothing, as the common pattern does
* refactor(native-chat): keep /btw hooks compilable by React Compiler
* fix(chat): say side chats aren't ready while a fork from the chat is opening
* test(chat): assert the opening-fork refusal without its copy
* fix(native-chat): share one latest-turn scan between /btw and transcript rows
* feat(runtime): confirm a structured chat send by clientMessageId past the first journal page
* fix(mobile): ask the send's own chat before handing it back
* fix(mobile): read send-recorded support when the deadline fires and settle rows that land mid-lookup
A Smart-sort save rewrote every listed worktree's rank whenever the order
changed, so every repository on the host got an order-only
worktreesChanged event. planWorktreeSortOrderUpdates now keeps the
longest run of rows whose ranks already descend in the requested order
and re-ranks only the rest. It falls back to the full re-rank when a gap
is too tight for distinct whole ranks, or when the planned ranks would
not descend strictly (float rounding at huge ranks).
When the requested order holds a row the write cannot save (WSL-move
fenced or unknown host), the write keeps the full re-rank, since that
row's unqualified rank may belong to a same-ID twin and must never
anchor a partial re-rank.
Replaces #37142.
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local>
Co-authored-by: BTForIT <223167996+BTForIT@users.noreply.github.com>
* feat(cli): set and rm folder workspaces through the folder workspace RPCs
worktree set and rm on folder:<id> (or a path:/name: selector that matches a
folder) now call folderWorkspace.update and folderWorkspace.delete after the
runtime confirms the folder id is unique. Flags with no folder meaning are
refused before any change, a missing method reports incompatible_runtime, and
worktree activate refuses folder workspaces as not supported yet.
* test: read mocked method names without a cast
* fix(runtime): refuse shared folder ids inside folder update and delete
The CLI's show preflight could not make the id-only folder RPCs safe: an older
show could select one of two same-id rows, and a twin added during teardown was
deleted too. folderWorkspace.update/delete now take requireUniqueId, which
refuses a shared id before any side effect, and delete removes only the row it
checked. Hosts advertise folder-workspace.unique-id-mutation.v1; the CLI
requires it and reports incompatible_runtime otherwise.
* fix(runtime): refuse a folder delete when a twin appears during teardown
Two rows that share a folder id share every record keyed by it, so delete
can't keep one row's sessions while removing the other. After the awaited
terminal teardown, delete checks the id again before any cleanup or removal
and refuses with the duplicate-id error; only the stopped terminals remain as
an effect. Removal by row identity is no longer needed and is dropped.
* fix(cli): give only the repair step for a duplicated folder id
The generic ambiguity recovery told agents to pass folder:<id>, which can't
name one of two folders sharing that id.
Import MAX_BLOCKS/MAX_PROMPT_BYTES from the module that defines them and
move automationCanonicalLaunchSetup into automation-launch-setup-source
(re-exported) so mobile shared readers skip unrelated RPC schema modules.
Carries #39022 unchanged.
Co-authored-by: m4air <m4air@m4airs-MacBook-Air.local>
Co-authored-by: gentslava <14369958+gentslava@users.noreply.github.com>
* feat(cli): match folder workspaces by path: and name:, and keep same-id ps rows apart
Adds one runtime matcher for worktrees and folder workspaces. With the new
matchFolderWorkspaces opt-in (advertised as workspace-selector.folder-match.v1),
path: and name: also match folder workspaces; branch:, issue: and identity:
stay repo-backed. Several matches refuse with each choice's kind, host, path
and selector; a folder id shared by two folder workspaces refuses with the
repair step. The CLI resolves path:/name: through it, so terminal, file and
ports commands reach a folder workspace, and create/set parents record a
folder parent. Explicit ambiguous parents report LINEAGE_PARENT_AMBIGUOUS.
worktree ps now keeps rows that share an id on different hosts apart, with
their own host, identity, metadata and activity; activity whose host can't
be proven attaches to neither.
* test: let the cross-version ps harness pass the host-qualified summary lookup
* fix(cli): read each ps row's own host session; qualify folder selector help
* fix(runtime): refuse unscoped duplicate folder ids; hand lineage the matcher's typed result
Unscoped folder:<id> refuses a duplicated id even when only one copy's host
is derivable; a host-scoped lookup still picks that host's copy. A wsl: hook
stamp counts as local host evidence. The lineage controller receives the
worktree-or-folder candidate directly instead of re-deriving the kind.
* docs(cli): guide the ambiguity fallback through truncated ps and id: selectors
* fix(cli): report an ambiguous parent by its own choices; cover per-host sessions
* fix(cli): send parents to the runtime parent resolver as typed; translate WSL paths for folders too
Parent selectors skip the CLI's show-based folder lookup, so worktree.create
and worktree.set classify them with parent rules: a folder spanning hosts is
a valid parent, and ambiguity reports LINEAGE_PARENT_AMBIGUOUS. Caller-distro
Linux path translation now also matches folder workspaces.
* chore(rpc): import the computer-use recovery data directly instead of through an alias
* fix(runtime): report a missing set parent as LINEAGE_PARENT_NOT_FOUND, as create does
worktree set and worktree create now share one explicit-parent error
conversion: several matches give LINEAGE_PARENT_AMBIGUOUS and none give
LINEAGE_PARENT_NOT_FOUND, so a missing parent is no longer reported as a
bare selector_not_found that could mean the child.
* refactor(git): move porcelain status streaming into its own module
* perf(git): let status use the untracked cache and expand collapsed directories
-uall never matches the untracked cache's stored mode, so every poll rescanned the whole tree. Scan in the configured mode and list only the collapsed '? dir/' rows with ls-files, reapplying the entry cap so results match -uall exactly. Ignored reads and status.showUntrackedFiles=no keep -uall.
* refactor(git): let the settled diff cache and stamp serve multi-path reads
* perf(git): reuse status numstat passes while their stamped inputs are unchanged
Each poll re-ran up to two whole-worktree numstat diffs even when nothing changed. Cache each pass in a SettledDiffCache validated by the worktree diff stamp (HEAD, index, .gitmodules, plus every unstaged working file for the unstaged pass); racy writes, mutations and submodule rows still recompute.
* perf(git): defer status while index.lock is held
A status read during an agent's commit sees a half-written state and is immediately redone. Wait (up to 1s, abortable, inside the shared lease) for the lock in the worktree's own gitdir to go away; a lock that survives a full wait is remembered as stale so later reads skip waiting.
* perf(git): pick the untracked scan from config and fall back to -uall when expansion cannot win
Read status.showUntrackedFiles and core.untrackedCache in one config call, remember worktrees that need -uall so polls do not scan twice, and rescan with -uall when rows sit outside cwd, need more than one ls-files chunk, or list empty.
* refactor(git): keep numstat cache limits beside the file diff limits
* refactor(git): stamp host settings files and resolve the common gitdir
* perf(git): read untracked-scan config from a stamped snapshot, with a rollback switch
Cache 'git config --list --show-origin' per worktree while every config file it named (plus repo, worktree and global candidates) keeps its stamp, so steady-state polls spawn no config read. WSL and UNC checkouts keep -uall: their config cannot be stamped from the host and network mtimes cannot vouch for the untracked cache. ORCA_GIT_STATUS_UNTRACKED_EXPANSION=0 restores the plain -uall scan.
* fix(git): stamp config and attributes before reusing status numstat
Numstat output also depends on Git config (including includes) and on .gitattributes, info/attributes and core.attributesFile. Stamp all of them, plus every .gitattributes that can apply to a changed path, and skip reuse under WSL. ORCA_GIT_STATUS_NUMSTAT_REUSE=0 recomputes every pass.
* fix(git): stamp include targets and onbranch HEAD in the config snapshot, and share one read
Include and includeIf targets (and HEAD for onbranch conditions) are now stamped, so a branch switch or a filled-in include re-reads config. Concurrent cold reads coalesce, both numstat passes share one config read, folders without their own gitdir skip the snapshot, aborted reads keep the cached entry, and Windows HOME follows Git's HOMEDRIVE/HOMEPATH order.
* perf(git): expand untracked directories only when the untracked cache is opted in
Git's default core.untrackedCache=keep only reuses a cache someone already built, and fresh repos have none, so status plus ls-files cost more than one -uall scan. Expand only for core.untrackedCache=true or feature.manyFiles=true without an override.
* fix(git): bound config stamps on dead mounts and share commondir resolution
Config file stats run under the same 2s deadline as local-git-config-signature, coalesced per file set, so a hung home or network mount skips reuse instead of stalling status. readHeadComponent now uses resolveGitCommonDir.
* test(git): pin the -uall truncation contract of untracked directory expansion
* fix(git): treat an unreadable commondir as unstampable config
main's resolveGitCommonDir rethrows non-missing errors; the config snapshot must fall back to -uall rather than reject the status read.
* feat(claude): sync the terminal theme into Claude Code
* fix(claude): retry missed theme reloads, keep settings key order, search entry
* fix(claude): persist theme ownership before settings writes, skip deliberate themes, restore only on toggle
* fix(claude): gate theme sync on Claude Code 2.1.119, follow Orca's Claude config dirs, keep restores owed
* refactor(theme): point markdown Monaco rules at the shared contrast helpers
* fix(claude): lock theme ownership edits across Orca processes, retry missed theme reloads with backoff
* test(e2e): opt-in real Claude Code theme sync journey; retry theme reloads for up to five minutes
* fix(claude): keep theme takeover recoverable when settings writes or the lock fail; steadier e2e journey
* fix(claude): serialize theme publication with takeover and restore, per-file reload retries; record the real-Claude e2e in the isolated home
* fix(claude): retry an owed restore publish, stop off-state retries, reread settings right before writing
* fix(claude): republish theme files under the cross-process lock
* refactor(theme): point chat selection contrast at the shared helpers
* feat(claude): default theme sync on for new profiles; offer it once to existing users at their next Claude launch
* fix(claude): offer theme sync to every existing profile, confirm only after the save, honest offer copy
* fix(claude): right-align the theme sync offer's two actions, primary last
* perf(ssh): let orcad coalesce output longer for background-visible terminal streams
Adds capability-gated SetOutputCadence (opcode 20) and an initial subscribe
cadence. Background streams batch over 100ms instead of 5ms (64KB cap kept,
bytes unchanged); switching to focused, or output carrying a reply query the
view answers, flushes immediately.
* perf(ssh): report focused vs visible-background cadence for remote panes
The remote multiplexer advertises outputCadence and sends cadence changes only
after the host echoes it. Panes report focused only when visible and the
active split, resyncing on bind, visibility, focus and tab-activity changes.
* perf(ssh): resync output cadence after splits and update cadence test fixtures
* perf(ssh): add orcad output-cadence kill switch and shorten background window to 50ms
ORCA_DISABLE_TERMINAL_OUTPUT_CADENCE=1 makes orcad neither echo the capability
nor apply the background window. Only callers that retune cadence advertise it,
so other terminal subscribers keep their wire shape.
* perf(terminal): send idle remote keystrokes without the 8ms debounce
Remote terminal input batching was a pure trailing debounce, so every
isolated keystroke waited 8ms before leaving the renderer. Send input at
once when the batcher is idle and coalesce only what arrives inside the
following window. Recovery-hold replay keeps the trailing path so a
synchronous flush cannot re-enter the still-buffering hold.
* perf(ssh): disable Nagle on loopback sockets carrying orcad RPC
The SSH forward listener, the stdio bridge listener, and the remote
bridge's connect to orcad relay WebSocket RPC (keystrokes and echo).
ws sets TCP_NODELAY on its own sockets, but these relay sockets did not,
so small writes could wait behind an unacknowledged segment.
* perf(terminal): decode terminal frame payloads as views, not copies
Every message buffer is fresh and consumers decode the payload
synchronously or retain it read-only; the only in-place rewrite touches
header bytes.
* fix(terminal): send idle remote keys at once only to input-acking hosts
On hosts without input acks, an immediately sent key travels as a stream
frame while a following accepted write (Ctrl+C, paste) goes as a separate
terminal.send RPC on another socket, so the host could apply them out of
order. Keep the trailing batch there so both still leave as one RPC.
Adds a branch picker to the Source Control header for switching or creating a branch in place. Switching is disabled while an agent is working in the workspace, and results show as toasts.
* refactor(agent-session): split subscribe params and read failure text into their own modules
Both host files sit at the 300-line budget, so the append negotiation that
follows needs room. Pure moves: SubscribeParams keeps its exact schema (now
z.object + sessionId instead of OptionsParams.extend, which is equivalent) and
is re-exported from the params module; stream failure text moves beside the
other read-refusal classification.
* perf(agent-session): stream running tool output heads as appends
A running tool call re-sends its whole output head (up to 16KB) on every
revision, and keeps re-sending the saturated head after it stops growing.
Streams that also ask for toolOutputAppend (host advertises
agent-session.tool-output-append.v1) now get only the head's new text under the
same shared sender/reader rules and self-healing refusal as reply appends.
- The live read negotiates the exact subscribe capabilities object instead of
a boolean; mobile asks for tool output appends when advertised.
- Hosts with only text-append.v1 drop the unknown key, so clients ask only
when the new capability is advertised and both ends track the same bases.
- Text-only streams keep the 8-base limit already-shipped readers evict at.
- A host that refuses the capabilities param (an older build after a
downgrade) makes the next open go without appends instead of failing forever.
* perf(native-chat): ask remote hosts for streamed text appends on desktop
Desktop chats on a paired or SSH-managed orcad received every streamed reply
checkpoint whole: about 9x the reply's bytes at the Claude/Codex checkpoint
cadence, and O(N^2) for ACP/JSONL providers that rewrite each 60ms window.
The desktop read now asks environment hosts for reply and tool-output appends
when they advertise them, reusing the shared reader that rebuilds whole items
before the reducer. Local IPC reads are unchanged and still subscribe in the
same turn. The support read runs once per chat run, overlapping hydration.
* refactor(remote-runtime): share compressed frame inflate across client platforms
Move the length checks and decode of paired E2EE compressed frames into
the shared module behind a small gzip/UTF-8 codec, so the desktop can
read them with node zlib while mobile keeps its native/fflate path.
* perf(remote-runtime): read compressed frames from paired hosts on desktop and CLI
Desktop and CLI sockets to orcad/orca serve now advertise
transport.compressed-frames.v1, so a v2 host gzips large text replies and
terminal output before sealing them, as it already does for phones.
- Every Node client transport (one-shot requests, request connection,
shared control, subscriptions) opens frames through one sealed-frame
reader that inflates compressed frames on v2 channels, synchronously so
frame order and v2 counters stay in arrival order.
- A compressed binary frame may claim up to the 8 MB binary outbound cap
(terminal frames can exceed the 4 MB JSON cap), for phones too.
- ORCA_DISABLE_MOBILE_COMPRESSED_FRAMES=1 now also stops a desktop or CLI
from asking for compressed frames.
- Relayed phones and the browser client are unchanged.
* refactor(remote-runtime): simplify the mobile inflate codec and clarify capability comment
* fix(remote-runtime): name why a sealed frame could not be read
Text-only sockets report an authentic binary payload as an unexpected
binary frame again, and tell a decrypt failure apart from a corrupt
compressed frame, instead of calling every failure undecryptable.
* fix(terminal): use Regular weight by default on Windows
Cascadia Mono, the Windows default, has no Medium face, so a 500 request
rasterizes as SemiBold. New Windows profiles default to 400; Windows profiles
still on 500 move to 400 once and see one persistent toast with a live
preview and a switch to keep 400 or go back to 500. Skipped when the
terminal font renders both weights identically.
* chore: prune stale renderer layout-writer baseline entry
* fix(i18n): translate the Windows weight toast label reactively
Adds a Custom… option to the App Icon switcher. The picked PNG/JPEG is
header-checked before decoding (square, 64–2048 px, ≤10 MB), re-encoded as
PNG into userData, and applied to the macOS Dock and every window icon. It
persists across restarts and can be removed, which resets to Classic.
Closes#13316
* fix(terminal): never type a colour reply into the shell after the program that asked has exited
The PTY owner holds a colour-query reply made inside a shell command for up
to 20 ms and drops it if the command's OSC 133;D arrives, or (POSIX) another
process group takes the terminal; otherwise it answers. Command markers ride
the colour scan every emission already pays, so the hot path does no new
work until a query is pending. No daemon protocol change.
Closes#16633Closes#14631
* fix(terminal): track command markers in the existing colour scan, drop the live-group shortcut, and hold through claimed echoes
* fix(terminal): one verdict for every query held behind the first, contiguous ranges through a held echo, drop held replies at the final snapshot
* refactor(terminal): move the torn query/echo holds out of the ingress
* fix(terminal): a query completed after a verdict is judged afresh; the final snapshot judges held askers instead of dropping them
Shows CI result lines in the chat that opened or pushed to a pull request. They're hidden in the terminal overlay and on the phone until the phone app gets the setting.
* fix(agent-picker): search fallback models and show cached models while the list refreshes
The new-tab agent picker's search skipped any agent whose model catalog was
unknown, while that agent's model submenu still listed the built-in models,
so typing "gpt" found nothing even though Codex showed GPT models. Search now
covers the same models the submenu offers.
The picker also waited for a fresh model probe on every read, and Claude and
Codex catalogs go stale after two minutes, so most opens showed the built-in
list for 0.5-1.5s. Like the chat composer, it now reads the saved listing
first and shows it at once, then follows with the waiting read, which joins
the same host refresh (no extra probe). An unknown catalog still takes the
waiting read, as before.
* Fix model catalog refresh when reopening the agent picker
* fix(agent-picker): wait only for active catalog refreshes
* fix(mobile): keep held model lists through late preloads and running chats
* fix(mobile): let progress answers fill empty lists and decide recording per read
* fix(mobile): record a pre-send listing on rejoin and leave unheld retirements alone
* test(desktop): use a valid listed catalog in the preload race test
* test(mobile): type the catalog cache test's renderer
* fix(tasks): confirm a lost Tasks comment by author on GitLab and GitHub REST reads
Only GitHub GraphQL's viewerDidAuthor marked the viewer's comments, so on GitLab,
the GitHub REST fallbacks (including issues with more than 100 comments) and the
gh view fallbacks the phone kept an unconfirmed draft even after the comment
posted. Those reads now match the login of the account that made them, and an
explicit PR refresh reads comments uncached.
* fix(mobile): settle an unconfirmed comment only on one posted after the send
A refresh of an issue with more than 100 comments reads comments the first
load never showed, so an older identical comment by the same account looked
new and could clear the draft.
* refactor(gitlab): share the glab viewer read and skip it on Checks polling
* test(mobile): pin the clock in the unconfirmed comment test
* fix(mobile): chat page follow-ups
- clear the touch state when the chat hides, so a held reply lands on return
- keep the chat pinch scale per chat
- record waiting display-math openers in the math plugin instead of a line regex
- share the details markup test and the orphan row id predicate
* fix(mobile): record math openers whose scan reaches the end, skip no-op pinch commits
* fix(mobile): fall back to a whole parse only for details section tags
* refactor(mobile): drop a repeated comment
* perf(mobile): open Jump back in and notifications inside the host document
* fix(mobile): gate open targets on page 3, keep notification suppression, fold double taps
* fix(mobile): report the shown session from host documents and only open this host's sessions
* test(mobile): ignore the page's location report in the agent-history notify check
* fix(mobile): let a Back before the paint gate settle the host stack
* refactor(mobile): build the route-param clear frame beside its name
* perf(mobile): let the model pill take picks before the chat attaches
* fix(mobile): lock the picker while a held pick is sent and keep the mode toggle for attach
* fix(mobile): report a held pick's refusal only for its own chat
* perf(mobile): route the last status.get callers through the shared status read
* refactor(mobile): drop the unused one-shot Codex reset capability read
* fix(mobile): open the chat transcript when the text-append check fails, and key connections on one epoch
* feat(native-chat): add /fork to fork a chat from its latest response
Typing /fork in a native chat now runs the same fork as "Fork from this
response", through the newest response still in context, instead of
passing the text to the agent. It is offered only where the host and
agent can fork, and answers with a composer notice while a response is
still running, before any response exists, while a fork is opening, or
when text follows the command.
* fix(native-chat): keep /fork off the transcript rows' fork surface
* fix(native-chat): read /fork's response from the hook's latest input
* Prepare queue rows to preserve accepted delivery intent
* Keep waiting native chat input in one ordered queue
* Keep a typed view of admitted queue records
* Keep the removed queue caption in the runtime catalog
* Remove the unused queue barrier caption
* Keep queue writes local and preserve accepted send provenance
* Read the sender from the queued admission request
* Keep continuation fixture clock within its operation window
* Remove deleted queue-barrier test from ratchet baseline
* Keep ratchet baseline change scoped to the deleted test
* Filter historical acceptance reads before loading payloads
* Instrument cached acceptance cursor once per fixture
* Restore original personal provenance after legacy card cleanup
* Complete legacy sender fixture metadata
* Narrow admitted history record types
* Reconstruct accepted personal proof after restoring original admissions
* Use the fixed clock for historical queue fixtures
* Consolidate queue source and type imports
* Preserve admission refusals and failed startup ownership
* Keep manual queue holds independent of later turn endings
* Remove obsolete mobile queue barrier expectations
* Keep journal format initialization independent of its decoder
* Assert returned-row recovery for released send clients
* Preserve Stop assertion across handoff-only publication
* fix(agents): workspaces created on a paired host use the host's launch settings
Creating a workspace with agent picks or a draft could launch the paired host's agent with desktop settings, or lose the picks during recovery. Let capable hosts compose those launches from their own settings and preserve the user's inputs through fallback creation. Older hosts retain their existing requests.
* fix(agents): workspaces created on a paired host use the host's launch settings
When a workspace is created on a paired computer, agent picks and Arguments can be replaced by this computer's defaults or dropped during recovery. Keep explicit choices with the submission and let the paired computer resolve its own launch settings.
Preserve older-host request behavior, disabled-agent draft substitution and prompt submission, and restore the composer if compatibility preparation fails.
* fix(agents): workspaces created on a paired host use the host's launch settings
Keep the submitted choices available when a background launch has no prepared startup carrier. Tighten the regression checks for restored composer names, legacy commands and prompt submission so recovery preserves the user's work.
* fix(agents): workspaces created on a paired host use the host's launch settings
Creating a workspace on a paired machine could ignore its agent settings or lose a draft while the terminal connected. Keep the selected launch inputs and the pending text together through fallback and reconnect, and never replay partially written input. Preserve older hosts' launch commands and ordinary new-tab prompts' editable behavior.
* fix(agents): workspaces created on a paired host use the host's launch settings
Keep a workspace's pending prompt attached to its original terminal while the pane reconnects. Stop automatic delivery before reporting a final failure, retain the recoverable text, and recognize a launch handed off to the paired host even before its tab appears. Avoid delaying creation on a redundant host status request.
* fix(agents): workspaces created on a paired host use the host's launch settings
New workspaces on a paired computer could lose their first prompt when the terminal briefly disconnected. Delayed delivery could also continue after reporting failure, or leave terminal-only agents without an accessible copy of their prompt.
Keep pending prompts through attachment loss, stop content writes when automatic delivery ends, and offer the existing composer to recover unsent text. Preserve launch identity without inventing empty Arguments or treating missing metadata as a replacement.
* fix(agents): workspaces created on a paired host use the host's launch settings
A replacement pane can learn its terminal before the layout catches up after reconnecting. Re-registering the effective launch settings then dropped the original first-prompt identity.
Recover that identity from the existing verified terminal binding so later remounts keep the same pending prompt.
* fix(agents): workspaces created on a paired host use the host's launch settings
A terminal binding can identify the launch before its effective Arguments are known. Keep permission notifications working in that state without assuming a launch-settings recipe exists.
* fix(agents): workspaces created on a paired host use the host's launch settings
Keep an unsent prompt attached to its launch when a paired host briefly disappears, so reconnecting the same terminal can finish delivery. Clear the rest of the exited pane's status as before, and release the binding after delivery or a confirmed end.
Offer Copy prompt when automatic delivery finally fails, so the prompt can be pasted into the terminal that already opened. Keep creation-failure Reopen separate.
* fix(agents): preserve paired prompt recovery across transport updates
Resolve the consumed startup token through the pane's verified binding instead of reading renderer ownership during every remote attach. Keep the transport's host-authoritative exit and reconnect behavior and its fixture guards.
Settle pending prompt delivery when the host reconfirms a closed surface, including after it discards the terminal's exit record. Update the remount fixture for the Windows PTY model binding.
* fix(agents): merge duplicate imports flagged by type-aware lint
* test(agents): fix type assertions in paired launch fixtures
Type the repository rows, mutable agent choice and terminal tab used by
launch regressions. Reuse the session binding across its real installers
and document the partial-fixture adapters required by the runtime APIs.
Keep the RPC, remount and prompt-delivery assertions unchanged.
* fix(agents): route workspace terminals through their owning host
Creating a terminal for an existing workspace could send it to another focused server. Resolve its owning host when no explicit route was supplied, and keep the host-shaped draft regression within the shared startup contract so renderer checks do not pull in main-process code.
* test(agents): align launch fixtures with runtime dependencies
Retain real i18n exports while replacing translation output, hoist the terminal restore toast spy, and supply the explicit host environment used by paired terminal callers. Keep the existing behavior assertions unchanged.
* fix(agents): isolate prompt failure notice dependencies
Load the prompt failure notice only when a settled hosted launch needs it,
so workspace recovery consumers do not initialize toast and localized agent
catalogs just by importing delivery code. Keep prompt retirement and its
recoverable draft synchronous, and preserve the existing Copy prompt action.
Restore the ordinary new-tab failure toast's original invocation and verify
that Fix Checks retains its edited request through the activation handoff.
* test(agents): route legacy wire launches to their fixture host
Pass the paired environment explicitly in legacy terminal-creation fixtures,
matching real launch callers instead of relying on the focused host. Keep
all compatibility and accepted-creation assertions unchanged.
* fix(agents): preserve paired work-item completion without a startup plan
Blank-terminal work-item launches on paired hosts now keep their successful activation result. If the agent command cannot be built, show that error once instead of reporting an activation failure.
* refactor(agents): deliver paired creation prompts once like agent tabs
Drop the retained hosted-prompt lifecycle (pending map, rebind, deadline,
terminal retirement, token recovery). After the host creates the session the
desktop attempts the prompt once; failure shows the prompt-not-sent toast with
Copy prompt.
* fix(agents): silence prompt failures after closing a workspace terminal
Closing a paired workspace terminal before the agent was ready could report a prompt failure after that terminal was gone. Use shared launch completion and existing tab records to keep deliberate closes silent while preserving one Copy prompt notice after an interrupted delivery. Prepared launches now decide client delivery from the host request.
* fix(agents): report every failed paired prompt delivery
A paired host can retract its tab from inventory without the user cancelling the launch. Always report a failed prompt attempt with Copy prompt, using the same single-attempt completion as agent tabs.
* test(agents): name host-session fixture mock types