* fix(computer): fail set-value when its read-back shows a different value
A value_mismatch came back as ok with exit 0, so callers that check only
the exit status took an ignored or misplaced write as success. The runtime
now turns it into a verification_failed error at the sidecar boundary,
which covers every provider and every caller of the RPC.
Part of #33511
* fix(computer): pair macOS AX windows with their CGWindowID
AppKit and WebKit windows don't answer AXWindowNumber, so matching a
capture to its accessibility window fell back to frames. Two windows with
the same frame then paired the screenshot of one with the tree and title
of whichever AX listed first, under the first one's id. Read the id with
_AXUIElementGetWindow, as yabai and Hammerspoon do, and keep AXWindowNumber
as the fallback.
Fixes#27252
* fix(computer): fail fast when a macOS app stops answering accessibility
Every request runs behind one helper lock, and a hung app cost a timeout
per AX read, so one get-app-state on it held every agent for up to the 60 s
RPC timeout. Pin the AX messaging timeout at 1.5 s, refuse an app that
doesn't answer a root probe, and stop a tree walk at the first read that
waits out the timeout, each with an action_timeout error.
A press or write that the app received but didn't answer now fails the
same way, saying it may have taken effect. Before, the click fell through
to AXConfirm and then a synthetic click, so a button whose handler blocked
ran twice, and type-text typed the text again as keys.
* fix(computer): report macOS accessibility actions that changed nothing
AXPress and other accessibility actions came back as plain unasserted
success even when the app ignored them, as Chromium does for a role=button
whose handler sits on a child. After such an action the helper now compares
the target's tree, frame and on-screen windows with the snapshot it took
just before. If they still match after re-reading for about a second, the
result says unverified/suspected_noop; if they change late, the result
carries the settled state instead.
The new reason is additive; older CLIs print it like any other reason.
Refs #31758
* fix(computer): propagate late macOS observation failures
* fix(computer): select existing text before replacement
* fix(computer): share native macOS window identity
* fix(editor): honor Plain text code block highlighting
* fix(editor): trim Plain text highlighting tests and drop redundant guard
---------
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
The N/10N case asserted exact start/peak/permit counts across concurrent work under fake timers and failed intermittently on CI (expected 3 to be 4). The phone dispatch/retirement and shared-context tests in the same file still cover the user-facing behavior.
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* fix(settings): permission menus match the chat permission picker
The per-agent permission menu (Settings > Agents and Source Control actions)
and the 'Agents without Approve for me' toggle rendered their own plain-text
items. They now reuse the chat picker's items: the same icons, Full access
warning colour, per-agent descriptions and width, with the agent's own
settings after a separator.
* Unify legacy permission menus and verify selection behavior
* Fix permission selection test typing and remove obsolete locale keys
A project checked out on two hosts kept one header, placed in the host
section of its anchor repo, so the other host's rows sat under the
previous project. Project headers now carry, for each host that owns one
of their rows, that project's own registration on that host and its
counts, and host sectioning gives every such host its own copy, the way
Pinned and status lanes already split. A header never borrows another
host's or another project's registration, a hidden host that only
anchors the header no longer gets a section, and a view that ends up
showing one host still names that host's registration.
Every header inside a host section now names that section's host, and a
copy an earlier section already drew is marked repeated. When a repo id
is registered on several hosts, a copy's create button opens the
composer on its own host; header drag keys on the project's anchor and
skips repeated copies, so reordering still moves the project once; the
phone keys each copy per section while collapsing it as one. The desktop
list also resolves each row's repo by host, as the feed already does, so
same-id registrations on two hosts keep their own.
Host sectioning also stops carrying an ended project group into the next
ungrouped project, and host badge counts no longer add a project group's
project count when the group is expanded.
Fixes#35741
* perf: gate store sweeps, dedupe plugin/catalog reads, index WSL git owners; convert nested WSL projects together
- STA-12863: skip binary draft cleanup on store writes that leave tabs, buffer index and boxes untouched
- STA-12852: native chat repo lookup re-resolves only when its file-link inputs change
- STA-12864: auto-update reads the install lockfile once per pass instead of once per plugin
- STA-12865: monorepo image fill runs per entry inside extraction and hands over bytes it already read
- STA-12867: Windows/WSL git router builds a repo->project owner map once per rebuild
- STA-12871: converting a project also converts nested projects so their saved-session identities move with them
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(plugins): retry the lockfile read after a failed auto-update read
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(wsl): expand nested conversion by the rewrite's own roots and drive roots; skip auto-update on a catalog switch
Addresses review on #36103: nested projects inside a saved worktree or under a drive root
were missed, and a source switch mid-pass could install from the old catalog.
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(computer): leave Linux and Windows windows in the background for accessibility actions
Element clicks no longer restore and activate the target window when an
AT-SPI action or UIA Invoke/SelectionItem/Toggle pattern handles them; only
the pointer fallback activates. On Linux, type-text and paste-text insert
into the focused editable field through AT-SPI EditableText without needing
window focus, and fall back to key synthesis or the clipboard otherwise.
Refs #12412
* fix(computer): refuse pointer input after failed activation
* fix(computer): stop after unconfirmed text writes
* fix(computer): insert the full UTF-8 text on Linux
* test(computer): focus each Linux text-entry trial
* fix(computer): verify Linux text mutations before deletion
AT-SPI can acknowledge writes without changing text. Read back
mutation counts and caret state before reporting success.
* fix(computer): verify complete Linux text replacements
* fix(computer): decline unreadable Linux text selections
* fix(emulator): follow decoded Android video dimensions
* test(emulator): trim decoded video size test to the rotation case
* chore(emulator): note video stream effect's timer cleanup for react-doctor
---------
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* feat(computer): drive macOS apps in the background by default
Clicks no longer activate and raise the target first. Pointer and key
input go to the target window's pid through a new BackgroundInputTransport:
mouse and scroll events carry the window id in field 51 and a window-local
location, keys run inside a focus-without-raise key-window scope, and
chords carry no Unicode string so menu key equivalents fire. Scroll and
drag now land (#27306); typed text no longer doubles in Electron (#17942).
The HID tap and focus fence stay for --restore-window only. Results
report action.delivery, the handshake reports supports.input.background,
and a missing private symbol or a popup click fails with
background_unavailable instead of falling back to the foreground.
* test(computer): read typed key strings back instead of diffing live-state events
* fix(computer): keep Unicode characters in one key event
* fix(computer): preserve active macOS key windows
Refuse key-window changes in the active app and surface failed
focus records, including cleanup after partially delivered input.
* fix(computer): guard macOS click routes before delivery
* fix(computer): let paste events read the clipboard
* fix(computer): classify clicks in the requested macOS window
* fix(computer): fence macOS scroll and drag recipients
* fix(mobile): report task pages and review state accurately
* test(mobile): record task page error provenance
* test(mobile): record task page errors in reply matrices
* fix(github-project): make linked PR numbers clickable
* fix: open linked pull requests at their exact URLs
* test: assert complete project relation cell text
* fix: clarify linked pull request controls and overflow
* Fix linked pull request focus clipping and overflow label
* Keep linked pull request overflow outside clipped values
* fix(github-project): center plain project cells and say what +N means
Drop the clickable linked PR numbers: the numbers stay plain text as
before. Plain text cells (repository, parent issue, linked pull requests,
sub-issue progress) now center vertically like the other cells, and the
linked PR overflow reads "+N more" instead of a bare "+N".
* fix(github-project): keep the linked PR "+N more" visible when numbers truncate
* fix(browser): close client-hosted Find on full navigation
* test(browser): trim client-hosted Find navigation coverage
---------
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* fix(emulator): match the requested AVD while waiting for boot
* test(emulator): trim AVD boot identity test to the regression case
---------
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
A successful read of the default /mnt/ root stored nothing, so every confirmed running-distro
probe (e.g. the 2 s WSL transcript observer) launched another wsl.exe guest read per distro.
Any successful read is now remembered until the distro stops; a failed read still retries, and a
stop during a read leaves the restart to read again.
* perf(wsl): a held WSL spawn reads the running list once
The provider-level holds (daemon adapter, daemon-off local provider) accept a running-list read started within a 5 s lease, so they reuse the startup-target hold's probe instead of running wsl --list --running again. A noted stop still forces a fresh read; a read older than the lease probes again.
* perf(wsl): a held spawn's provider hold reuses only that spawn's own running read
Replaces the 5 s lease: the startup-target hold records a read anchor on the spawn's options and reads after it; the provider hold reads after the same anchor, so it reuses that spawn's read and nothing older. No anchor means a fresh read; a noted stop or wake still forces one.
* fix(wsl): cap how old a spawn's own anchored read may be at the provider hold
A daemon respawn or a same-session history lock can sit between the two holds, so the provider hold reads after max(anchor, now - 2 s).
The WSL rows kept for a transient running-list failure were stored per
scope key in a module-global map with no limit, so every visited scope's
rows stayed alive until a history delete or a restart. The map now keeps
the 4 most recently used scopes (carrying a scope over counts as use) and
spends no slot on scopes without WSL rows. Explicit deletes still clear it.
* fix(ci): drop duplicate grab-annotations type and split chat subscription perf test
* feat(browser): send compact annotation feedback; page details are opt-in
Annotations sent or copied to an agent now carry the page, element label,
intent, selector, source file, text and feedback. Tab id, viewport, bounds,
classes, nearby context, computed styles, DOM path and HTML move behind an
'Include page details' checkbox in the Send menu, which also drives Copy.
* refactor(browser): share the annotation Copy tooltip and simplify the details toggle
* refactor(browser): keep tray types with page types
---------
Co-authored-by: Neil <neil@stably.ai>
Fixes#34312. Takes over naufaldi/orca#1 by @naufaldi, with the main.css part dropped because #34932 already covers it.
- SshTargetCard and the reasoning chevron now use the scoped `[[class~='group/x']:has(:focus-visible)_&]` shape. Tailwind's `group-has-*` compiles to a non-subject `:has()` whose subject is `*`, and that rule made Blink restyle whole subtrees on every mutation. This change is the entire measured win: 19.4 → 1.6 ms, 12.5 → 3.6 ms and 8.0 → 2.4 ms per mutation.
- rich-markdown-editor.css drops a redundant `:has()` on the TOC header, as defence in depth.
- Adds a guard, check-has-restyle-scope.mjs. It compiles the renderer's Tailwind CSS, adds plain stylesheets and the listed vendor stylesheets, and fails on any non-subject `:has()` whose subject has no class, id or attribute. Three pdf.js rules are baselined after being measured as harmless. It runs in lint, check:changed and PR CI in about 1.5 s.
Co-authored-by: Naufaldi <naufaldi.rafif@gmail.com>
OrcaWinandOrcaWin
Kelvin Amoaba
Brennan Benson
Neil
Jinwoo Hong
