* feat(workspace-cleanup): keep workspaces a paired client has unsaved edits in
* fix(workspace-cleanup): keep each window's unsaved-editor report and send them in order
* fix(github): keep the selected workspace's PR on its one-minute refresh
Follow-up reads inherited the priority of whatever entry ran last. A workspace
selected while its cached PR was still fresh never got a foreground entry, so
its follow-ups competed with every visible card for the shared list budget and
were read about every five minutes with 25 visible open PRs. Previously
selected workspaces kept priority 80 forever. Follow-ups now take foreground
priority only while their workspace is selected.
* feat(source-control): say when a merged workspace is safe to delete
Source Control's committed list is a merge-base diff against the base ref. A
squash or rebase merge copies the commits onto the base with new IDs, so the
list looks the same after the merge as before it. A notice now states whether
the merged review contains everything: HEAD is the review's head (or the
provider confirmed it is contained) and nothing is uncommitted.
* fix(sidebar): keep PR status visible on sleeping workspaces
The sleeping moon replaced the review glyph, hiding whether a sleeping
workspace's PR was open, failing or merged. Review status now keeps the lane;
the dimmed row already shows sleep and the accessible label names both.
Co-authored-by: Tobias Diez <code@tobiasdiez.com>
Co-authored-by: alpenshtokk <wawesandways@gmail.com>
* fix(source-control): wait for a git status read before judging a merged workspace
Before the first status read the change count is zero and HEAD may be stale,
so the notice could call a dirty workspace safe to delete. Use the HEAD from
the latest successful status read and say nothing until one completes.
* chore(i18n): ship the merged-review plural copy in the runtime English catalog
* fix: satisfy design-system, cast and flaky-test gates for the merged notice
Use the callout-important token instead of raw purple, let Button own its
type size, build the fixture worktree with the shared factory, pin the test
clock and give the long simulated-time tests a reason for their timeout.
* fix(github): align selected-refresh regression with feature ownership
* refactor(github): give refresh queue aliases their concrete owner
* fix(source-control): preserve cleanup warnings and reserve refresh fairness
* fix(source-control): describe missing merged-head proof as unverified
---------
Co-authored-by: Tobias Diez <code@tobiasdiez.com>
Co-authored-by: alpenshtokk <wawesandways@gmail.com>
Scan merge proofs run with lazy fetches and transports off, so partial
clones never download blobs; direct WSL reads forward those variables.
Base and coordinator refs resolve once per repo per scan, and the squash
search lists target commits once and only patch-ids commits that touch
the branch's exact file set.
* feat(workspace-cleanup): run cleanup rules in the background on the owning host
* fix(workspace-cleanup): keep workspaces whose own orca.yaml has an archive script
A workspace a worker start created gets the 'orchestration-worker' reason once every dispatch on it released its terminal and its work landed in the default branch, a merged review or the coordinator's branch. Commits the coordinator's branch already holds no longer count as unpushed.
* feat(workspace-cleanup): classify finished automation run workspaces
A new-per-run automation leaves one workspace behind per run. Cleanup
scans now tag such a workspace 'automation-run' once its run is final
and the automation has released the run's terminal, so the
automationRuns rule can remove it behind the usual safety blockers.
* refactor(workspace-cleanup): keep the scan file under its line cap
* Scope workspace deletion shortcuts to selected sidebar rows
* Complete shortcut event typing and web bridge cleanup
* Use stable fixture rows and typed input elements
* Align sidebar deletion and navigation with eligible focused rows
* Pass explicit absent repo for selected row host lookup
* Describe selected options in keyboard test fixture
* Bound sidebar selection and deletion work and preserve host ownership
* Keep navigation imports within the file size limit
* Separate navigation type imports consistently
* Match selected workspace repositories by explicit host
* feat(computer): type into a named text field without focusing it on Linux and macOS
* fix(computer): report unconfirmed Linux text readback and cover element typing
* test(computer): send an element with benchmark text only for targeted typing
* test(computer): pick gedit's editor, not its hidden overview map, in Linux e2e
* docs(computer): confirm focus before the unindexed type-text fallback
* fix(computer): report unreadable element text readback as readback_unsupported
* Give terminal agents a lasting Orca Agent ID
Terminal agents now get an Orca ID (orca_session_id:<agent>_<uuid>) derived
from their own conversation id on the host that runs them, so it stays the
same across worktree sleep/wake, app restart and a hand-run resume in any
pane. The terminal pane menu gets Copy Agent ID beside Copy Session ID, and
worktree ps --json agent rows carry orcaSessionId. Messaging a terminal
agent by that ID is refused with a message naming its pane until terminal
delivery lands.
* Name terminal agents at row admission and fix review round 1
- Perf: a row's Agent ID is resolved once, when its provider key changes at
admission (legacy rows via the status adapter's prepareRow hook, tmux rows at
canonical commit) and carried on the row as runtime-only agentIdentity.
Snapshots, publications and internal notifications only copy it; no key
serialization, cache lookup or database access per walk. Drops the pane FIFO
cache.
- Refusal: the identity service keeps a reverse index id -> rows, kept exact by
the adapter's rowChanged hook; a terminal refusal reads only matching panes
and names only confirmed running ones (no resume seeds, exited or
unconfirmed restored rows).
- Retry: a read-only identity file is lookup-only and never mints; any store
failure backs off all SQLite work for 30 s; a Copy click retries once.
- Menu: Copy Agent ID shows for a live pane only when its host stamped an ID,
and for a sleeping pane only when this app owns it on macOS/Linux.
- SSH: a sleeping record keeps its connection; a foreign (non-WSL) one resolves
no ID.
- Renderer keeps an existing ID only for the same agent, owner and provider key
including the Pi transcript locator.
- tmux-selected terminal rows carry their ID at every snapshot seam and in Copy
resolution; structured chat rows stay excluded.
* Keep Agent ID work out of notifications and reach Copy wherever a host can answer
- Internal status/provider notifications walk rows without Agent ID work:
canonical (tmux-selected) rows are decorated only for outputs that carry
the id (snapshots, live emit, ps, Copy). The tmux publish body is now
publishTmuxSelectedStatus so tests drive the production path.
- A row this host can name but has not yet (failed or pending link) is
published with orcaSessionId: null; older hosts omit the field. The
renderer offers Copy Agent ID on a live row whenever the field is present,
and the click retries through terminal.agentId.
- Hosts advertise terminal.agent-id.v1. A sleeping pane offers Copy Agent ID
when its owner is this app on macOS/Linux, or a remote owner that advertises
the capability on a known non-Windows platform, and its record names no
foreign connection.
- "Running in pane" places require that this runtime observed the row; a
saved status, Done included, is history.
* Guard the pane menu subscription and stop offering Copy where it can never succeed
- A settled read-only miss (no link in a newer build's identity file) no
longer publishes orcaSessionId; only a resolved id or a pending failed link
does, so Copy Agent ID is offered only where a click can succeed.
- Render contract for the per-pane menu selection: another pane's or host's
status change re-renders no unaffected pane; a pane's own stamp flips its
availability.
* Move terminal agent identity SQL into a worker thread
Main's blocking-SQLite ratchet forbids the synchronous identity store, so
(per the decider ruling) agent-identities.db is now owned by one lazy
worker-thread entry per hook-server owner, desktop and orcad alike.
- The schema, open/version handling, mint/link transaction and read-only
behaviour move unchanged into agent-identity-worker-entry.ts; main keeps
only the key type, path and an async client over WorkerThreadRequestQueue.
- Admission stays synchronous: a new key is published pending (null), one
request per exact key is shared by every row and click, and the committed
answer re-stamps the current rows still pending on that key (legacy and
tmux-selected) as identity-only replays. Replies are fenced by store
generation, key and current subscription; failures back off 30 s and a
Copy click retries once.
- terminal.agentId and the terminal-ID refusal are async: Copy waits for the
pending lookup and revalidates the pane; send/task routing resolves the
terminal recipient (awaiting the store's has() on an index miss) before
any synchronous routing or effects.
- The renderer treats the host's row naming a conversation as authoritative
for its id, so a settled miss clears a pending null, while ID-less local
updates keep it.
- Packaged for desktop and orcad; registered as a worker entry, no baseline
growth.
* Own identity demand per key, publish settled ids as projections, recheck authority
Implements the second decider ruling for the identity worker:
- One owner-level demand table replaces the per-key promise backlog and the
settled-row scan: each exact key a current row or caller needs has one
record (identity, places, waiters); the pump sends the worker one operation
at a time, only for keys still needed, so a healthy burst completes and
detached keys are never sent. Operation deadlines start at dispatch; a
caller's wait expiring never drops a row's demand.
- One failure gate for lookups and membership: after a failure nothing is
sent for 30 s except one coalesced Copy attempt; a single owner wake
retries while rows still wait, and cancels with the last demand or close.
- The retirement barrier outlives a client: a replaced worker must exit
before another starts; a different store path is a different authority.
- Settlement re-stamps the stored row and publishes a projection only: window
readers get the row without its hook event (no agent-state record) and
session.tabs is woken by a row touch; no enriched subscriber, plugin event
or freshness heartbeat runs again.
- Copy and terminal membership check the owner generation after their waits.
A known chat is decided by the chat store alone; send, peer mail, reply and
assignee admission resolve the terminal recipient before reading the facts
they act on, recheck the sender after the wait, and group send reselects
its audience and mailboxes after each wait before writing.
* Scope identity answers to their store authority, expire Copy grants with callers, index provider aliases
- One synchronized authority token before Copy/reverse fast paths, compared after their awaits;
answers carry their authority and an ended store turns rows, tmux subjects and the reverse
index pending together. Status path changes sync the authority outside admission.
- A failed operation reschedules the pump so a queued click gets its attempt; grants belong to
live waiters and a click joining a running attempt earns none.
- Provider-native ids resolve through a commit-maintained alias index at the record store.
- Group fan-out no longer classifies authoritative candidates; @idle rereads statuses with
current candidates before the final synchronous selection.
- Hook-server unit fixtures use an in-memory identity client; a real worker load/stop failure
check covers the worker boundary.
* Pick a shared provider alias's owner among its owners only
The alias index keeps the store's listing order (readable, then unreadable) as derived ranks, updated from the draft's appended keys on commit, so a duplicate alias no longer walks the history per lookup.
* Ask a remote pane's host for its Agent ID by the host's own pane key
Copy Agent ID on a web-terminal surface sent the client's mirrored pane key, which the execution host does not know, so it answered null.
* Keep focus where the user moved, recover a missed identity re-stamp, ack peer-mail replays first
Copy Agent ID refocuses its pane only when focus is unclaimed or still in the pane; a carried pending row takes its key's settled answer; an already-delivered peer-mail retry is acknowledged before recipient preparation.
* fix(automations): follow-ups to the automation detail redesign
- Show the latest occurrence as a folded skipped run's last-run time. The
detail health band, list rows with loaded history, and the projected list
summary all read the retained run's first-occurrence timestamps, so a host
that kept refusing runs looked like it last ran days ago.
- Translate the new detail strings (Repeats, Missed-run grace, Execution,
Not run yet, the usage detail, and the prompt copy labels) in every
supported locale, reusing existing localized wording, and give the usage
detail CLDR plural forms again.
- Assert the prompt preview's eight-line boundary by behavior in the e2e
spec instead of the line-clamp CSS value.
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(automations): zh copy-prompt wording
Use 提示词 in the new zh copy-prompt label so it matches the panel heading.
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* feat(native-chat): run Kimi Code in Chat over its ACP server
Kimi chats launch `kimi acp` with Kimi's own sign-in and KIMI_CODE_HOME.
Orca sets Kimi's own permission mode at every start: Ask is Kimi's
default, Approve for me is its Ask When Needed mode, and Full access is
its Never Ask mode. Without that, a default_permission_mode in Kimi's
config.toml would apply while the session reports default.
A Kimi dialect reads Bash results as output and exit code, drops the
streamed arguments a running call shows as text, names the command or
path on approval cards, and takes questions as forms so every question
reaches the chat.
* fix(native-chat): import Kimi session history helpers from their new home
* fix(computer-use): scroll, expand and set sliders on Windows without raising the window
* fix(computer-use): wait for Windows scroll and slider writes to settle before verifying
* fix(computer-use): reject NaN for Windows slider set-value
* test(computer-use): wait on the Windows fixture state with vi.waitFor and a schema
* test(computer-use): check activation per action against a holder window in the Windows fixture
* test(computer-use): log fixture window events for activation failures
* fix(computer-use): report foreground when a Windows app raises itself for a UI Automation action
WPF activates its own window for Scroll, ExpandCollapse, RangeValue and SelectionItem calls from any UIA client, so the runtime now compares the foreground window before and after a pattern action instead of assuming background. The Windows e2e checks delivery against the fixture's own activation count, with a holder window in another process.
* test(computer-use): skip an unstarted holder when closing the Windows fixture
* fix(computer-use): bound Windows pattern scroll paging and tighten slider read-back
* fix(computer-use): cap Windows slider read-back tolerance and run the UIA e2e only on CI
* feat(dictation): optional AI cleanup of dictated text, with a correction word
Off by default. When on, a dictation's text goes through the user's agent
CLI (the AI text actions' runner, Haiku by default for Claude) before it
lands: the pill says "Cleaning up…" and the text is inserted once. Failure
or a 10 s timeout inserts it as heard with a warning; the pill's X skips
cleanup. A consent dialog asks before text first goes to an agent. The
correction word ("scratch that") survives filler removal for cleanup to act
on. The microphone test gets "Try cleanup", and phone dictation the computer
transcribes is cleaned on the host.
* fix(dictation): keep finals that arrive during cleanup and tighten cleanup edges
* fix(native-chat): let a chat continue on the account you switch to after a usage limit
A usage-limit reset belongs to the account that hit it. After a switch the
host kept holding the chat until the old account's reset, and desktop and
phone hid Continue until then, so Switch Account on the card did nothing.
The host now releases the hold once the chat would start on another
account, letting only that account's own reading hold it, and publishes the
earlier continue time that both clients read.
* fix(native-chat): save the account a usage limit hit, per cut
The fallback to the chat's founding account could report a switch that
never happened for a chat that had already moved, releasing it before the
new account's reset. Save the cut's account while the child that ran it is
live, use only that, and key the switch time by the cut it belongs to.
* fix(native-chat): record the account each chat runtime launched on
record.accountHome names only the founding account, so after a switch
nothing said which account a usage limit hit. Owner proof now records the
account each runtime fence launched under, every host cut carries its
turn's fence, and the hold reads the cut's account from that pair. This
replaces the sweep's saved copy, which a fast child exit could miss.
Spending the new account showed before a switch now holds the chat until
its own reset; only a reading taken after the switch releases it.
* fix(native-chat): restart the switch time when the chat moves to yet another account
A second switch for the same cut kept the first switch's time, so a reading
of the newer account taken before it was chosen could release the chat.
The agent already has full thread data from selectedCommentGroups. Report
only the host-resolvable thread count in the human-readable section to
avoid duplicating thread bodies in the JSON payload.
* fix(computer): verify macOS type-text and paste-text sent as keys by reading the field back
When text can't be written through AXValue (Chromium/Electron fields, fields
that refuse the write), type-text and paste-text send keys or Cmd+V and always
reported unverified. Now the helper predicts the focused field's value from its
value and selection before sending, then reads it back until it settles:
verified on a match, unverified (text_mismatch) otherwise. Secure fields and
fields without a readable value and selection stay unverified as before, and
the result never repeats the typed text or the field's contents.
* fix(computer): keep the caret put after a refused macOS text write and skip unprovable readbacks
A refused AX value write no longer moves the caret before falling back to keys, so the keys and the prediction both use the user's selection. Text with control characters, a value the field already holds, and a field still filling at the 2 s cap stay unverified rather than verified or text_mismatch. The quiet window starts after the first read answers.
* fix(computer): bound macOS text readback size and say a mismatch can mean keys went elsewhere
* fix(computer): treat a macOS text readback past the cap as undetermined
* fix(computer): refuse quit, log out and lock shortcuts unless explicitly allowed
* fix(computer): close Linux keysym bypasses in the system shortcut guard
* fix(computer): refuse Linux keycode, L1/L2 and VT-switch keysym bypasses
* fix(computer): catch control characters in Linux keycode tokens and point recovery at the stated effect
* feat(computer): show where snapshots were cut and read one element's subtree
* fix(computer): validate subtree targets against the fresh walk and keep scope for focused input
* fix(computer): return whole-window action results after subtree reads
* fix(computer): count kept records toward the macOS limit and index subtree reads against the latest cache
* fix(computer): index Linux/Windows subtree reads against the snapshot they started from
* fix(skills): stop calling another Orca build's skill copy out of date
Every Orca build on a host (the installed app, dev builds, the CLI, a paired
server) shares one skills folder, and each one auto-updates it to the text it
ships. A build only recognises SKILL.md text from itself or older releases, so
a copy a newer build (or a dev build) wrote looked like a user edit: the
nudge said "An installed Orca skill is out of date" and Update offered to
replace it with older text.
Whenever Orca writes a skill it now leaves a marker,
~/.agents/.orca-skill-digests/<skill>/<sha256>, on the disk of the host that
plans. A copy whose bytes match a marker plans as leave 'other-build': it
reads as current, is never nudged, and silent update leaves it alone. A user
edit changes the bytes, so it is still asked about. A paired server plans and
records on its own disk, so the client never compares hosts.
* fix(skills): mark copies already at this build's digest so other builds see them as Orca's
A copy that already equals this build's bundle plans keep-only, so no write
ever recorded its marker, and copies written before the marker existed stayed
'edited' to any other build. Executing a plan now records the marker whenever
a location holds this build's bytes, and the silent update records it for
keep-only entries. The startup scan asks the host (this machine or a paired
server) once per session to run that keep-only update, so an already-current
build marks its copy on its first run. Preview stays read-only.
* chore(skills): drop unreachable other-build setup copy and document touched planners
'other-build' leaves are always replaceable, so the non-replaceable setup
attention lookup never selects one; it now shares the generic fallback and its
dedicated string is removed. Add short docstrings to the touched planning,
execution and copy functions.
* fix(skills): retry the startup marker when the host did not record it
A marking-only call kept its session mark even when the host never ran the
recording, so a failed or busy call left the copy unmarked for the rest of the
session. Release the mark when the call rejects, another update held the
target, or the host reports the skill busy or changed, so the next plan
refresh asks again. A host without silent update keeps its mark, avoiding a
call per scan.
* fix(skills): include Codex session's P2 fixes; warn only when a marker call may have written
Includes two P2 review fixes from a Codex session in this worktree:
- A newly selected agent links to a recognized other-build shared copy
instead of being left as shared-not-current, and execution re-checks that
shared copy so the dependent link is skipped if it changed after planning.
- Installed-skill views refresh after every silent-update call, since a
marking-only call re-plans on the host and can still update a copy.
A rejected marking-only call now shows the uncertain-outcome warning only
when the request reached a paired host and its reply was lost. A refusal
before sending, or the host's own error reply, cannot hide a write: the mark
is released for retry without a warning. Calls that started real updates
keep warning on any rejection. Restores the docstrings removed in that
session.
* fix(skills): stop retrying a failed skill marker on every plan refresh
Every marker call ends in a plan refresh, and a refresh re-scans. Freeing
the mark after a busy or failed call made each refresh ask again, so a host
that stayed busy or unreachable looped without end. Ask once per target,
skill and bytes per session; a later session, a manual update or any write
of the skill records the marker instead.
* fix(skills): recognise another build's copy under a skill's former name
A build from before a rename records its marker under the former name.
Check that marker for former-name copies too, so a renamed build no longer
calls such a copy edited and nudges about it. A user edit, a cloud receipt
or extra files still keep the copy as it was.
* fix(skills): warn about any failed skill update that may have reached the host
A paired host's timeout or lost reply also arrives as an RPC error, so an
RPC error is no proof that nothing was written. Only a refusal before
sending now stays quiet.
* fix(skills): send a silent skill update only to the host that was checked
Pin the update to the runtime and pairing that answered the capability
check, so a re-pair while the check awaits cannot send it to another host.
* fix(skills): record markers without updating, and never offer to replace another build's copy
A marker call now asks the host to record copies already at its bytes and
nothing else, so it can never write a skill and needs no warning or
re-plan; hosts that predate it reject the request and write nothing. It
is sent only in a scan that starts no update.
Another build's copy is no longer offered for Replace, which could
downgrade it, and its dialog line reads "Kept: written by another Orca
version".
* fix(skills): keep other-build setup current across host pairings
* fix(skills): preserve other-build prerequisites during migration
* feat(computer-use): add orca computer wait for text, enabled controls and window titles
* fix(computer-use): bound wait reads by the deadline and wait out missing Linux/Windows windows
* fix(computer-use): read disabled traits after role text with parentheses and bound the capability probe
* fix(computer): check macOS click targets are enabled, in view and on top before sending input
* fix(computer): scope macOS click cover checks to the target window
* fix(computer): scroll before trusting container hits and stop hit walks at the shared container
* refactor(computer): share the one-scroll rule in the macOS click check
* Keep model picks independent of answer metadata
* Remove unused answer model extraction
* Verify model selection stays separate from answer metadata
* Keep the agent-chosen model as a display default
* Keep agent-chosen model text out of input selections
* fix(agent-picker): stop listing Claude short-name rows beside the versioned model they run
* fix(agent-picker): hide a short name only on reported proof; keep hidden rows searchable
* fix(native-chat): keep a pending hidden model pick listed
Kelvin Amoaba
Brennan Benson
OrcaWinandOrcaWin
Neil
