The window's runtime calls never carried the orchestration contract
version, so the card's Allow and Deny were refused as an obsolete client.
The renderer ships with its main process, so desktop IPC states the current
contract. An open card whose agent's own countdown has passed, with no
closing message (the app restarted while it was open), is no longer shown.
A worker's open permission prompt appears as a card at the end of its coordinator's chat, drawn
from the escalation payload as soon as it is stored. The person allows or denies it there; the
card's buttons answer the worker's own prompt over the desktop app's IPC, so only a person can
allow. Once answered, the card turns into a receipt saying who answered.
The transcript read is async, reads only its tail, and skips any path that
is not a regular file, so a FIFO or device cannot stall the main process.
A closed prompt still awaiting its result gets one last read when the lead's
turn stops, so a result written late still names the close.
The screen could show an older prompt's close, so who answered is now read
from the tool_result Claude's transcript records for that tool call: Orca's
own deny text credits Orca, the user's No and the 2:00 safety-check block
name themselves, and a run names an allow. Hook events cannot say this: a
deny or a No fires no PostToolUse. The screen only says whether the dialog
is still up before a write, and a PostToolUse seen mid-answer is kept.
An answer is written only while a caught-up screen still shows the dialog
in the same process run, and claimed only once Claude shows it applied: the
hook's deny line for a deny, no other close for an allow. A Yes, a No or the
2:00 auto-block that wins the race is refused and named from the screen.
A hook that hangs up is not an answer: with the dialog still up the
coordinator is told a person must answer it in the worker's terminal.
A PermissionRequest entry the user made async stays async: it prints {}
first and never waits for a decision.
A worker pane's PermissionRequest hook now waits for Orca's answer, so the coordinator can deny
it with a reason (the worker continues) and a person can allow it. One record per waiting hook,
keyed by Claude's prompt and tool call; it is answerable only while the hook waits and the dialog
is on screen, and anything else gets `{}` as before.
The approval escalation now says the worker is waiting for permission, tells a chat
coordinator the user has been shown Allow/Deny (a terminal coordinator: a person must
answer it in Orca or the worker), asks it to carry on with other work and not message
the worker until the answer arrives, and offers deny only as the way out when the
prompt is holding up the task. The guide and CLI docs say the same.
An await before the insert let a worker settle meanwhile, so the mail could land on a settled
Dispatch after its follow-up scan. The closer also removes only the approval map it read.
- A same-option answer counts as Orca's only when Orca recorded it; a retry returns the same
receipt with no second reason mail.
- An unreadable chat is not a closed prompt: the answer refuses, and the closer keeps watching.
- A closing line tells a reason only when it went to the worker as mail.
- The terminal held receipt prints the delivery gate's own refusal; the chat one prints what the
pointer's admission and queue gate would do (not attached, refused, or behind an open prompt).
A chat worker that stops at a permission prompt now escalates with the prompt
itself as structured data (approval id, state, worker, summary, allowed
actions). The coordinator can deny it with a reason through
`orchestration deny`; the worker carries on and gets the reason as mail.
Only a person can allow a prompt: the answer method decides "person" from
the transport, never from params. The stop report keys one escalation per
approval and sends one closing line when it is answered, and no longer tells
the coordinator that `orchestration send` answers a prompt.
`orchestration send` now warns when the delivery gate would hold the mail
right now (an open prompt, a blocked status, typed text), and says delivery
state is unknown for a worker on a paired server.
* fix(native-chat): answer Claude/Codex folder-trust dialog from a chat card
* fix(native-chat): show a chat card for Claude/Codex startup menus
* fix(native-chat): keep startup-menu cards holding Send; drop hand-typed identify retries
Startup-menu cards now hold the composer like the folder-trust card, so collapsing one can't free a Send whose Return would pick the highlighted choice.
The hand-typed agent probe retries and agentless hold are left out; Orca-launched panes already know their agent, and hand-typed native Claude identification predates this fix.
* fix(native-chat): read the whole screen for a tall startup dialog's heading
* fix(native-chat): drop the unreachable cwd prompt from startup-menu reasons
* fix(native-chat): show startup-menu warning text and give its choices no default
A startup-menu card now carries the dialog's body (Claude's bypass, API key, MCP and import warnings) under its heading, and its choices render as equals on desktop and mobile so Orca never nudges a consent choice. Long mobile option labels wrap inside their button.
---------
Co-authored-by: Kooooojun <edisonhuang1122@gmail.com>
Kimi Code's --yolo is now Ask When Needed; --auto is its Never Ask mode.
Kimi refuses --yolo (or -y, --yes, --auto-approve) together with --auto,
so a Kimi approval flag the user put in Arguments keeps control.
Co-authored-by: rumoii <123233666+rumoii@users.noreply.github.com>
The workspace watch never follows a symlinked folder, so a tab opened
through one (target outside the root) got no change events. Watch each
such tab's folder directly; main checks for the link with an lstat walk
and already rewrites the folder watch's events to the link spelling.
* fix(mobile): show Loading for a new Tasks search instead of the previous search's rows
A new filter or search with no kept rows kept the same provider's rows on screen until the host's read landed, so a slow host looked like search did nothing. Only a re-read of the same list now stands on its rows; desktop already shows its load skeleton here.
* test(mobile): re-record Tasks list goldens for the new-filter spinner
* fix(mobile): hold Tasks paging while a new filter loads
* test(mobile): type the new-search spinner test's GitHub row fully
* perf(emulator): stream iOS Simulator video as H.264
The iOS Simulator pane decoded serve-sim's MJPEG stream (~24 updates/s,
p95 latency 77 ms). serve-sim 0.1.47 already serves /stream.avcc, so iOS
sessions now decode it with WebCodecs (~59 updates/s, 26 ms).
- Main derives the AVCC URL from the session registry by device id; the
renderer never supplies a stream URL.
- A decode-queue backlog restarts the stream from a fresh keyframe; repeated
backlog falls back to JPEG.
- Any H.264 failure falls back to JPEG for that session, including a socket
silent longer than serve-sim's 5 fps idle floor allows. A late idle timer
after sleep gets a fresh window instead of failing.
* test(emulator): drop the wall-clock simulator video stall tests
They froze the event loop and raced real socket timers, which the repo's test
rules forbid; the JPEG fallback they guard is covered in the renderer test.
* fix(emulator): start the simulator video stall timer at the first byte
serve-sim writes nothing on /stream.avcc until capture starts, so a slow cold
start could trip the 3 s stall and pin the session to JPEG. The renderer's
first-frame timeout already covers the wait before the first byte.
* fix(emulator): re-arm the first-frame timeout when H.264 resyncs
A resync opens a new helper connection, and main's stall watch only starts at
its first byte, so a silent reconnect would freeze the canvas with no JPEG
fallback.
---------
Co-authored-by: Kien Le <122910950+kien-ship-it@users.noreply.github.com>
The catalog's Install link opened DeepSeek's site, which installs only
dsh. Orca launches the community @deepseek-harness-tui/dsh-tui, so users
either stayed undetected or guessed the dead unscoped dsh-tui package.
* fix(browser): forward Orca shortcuts from focused HTML document previews
Workspace-document preview guests never installed the guest shortcut
forwarder that browser pages and plugin frames use, so once a preview
took focus Ctrl+Tab and the other window shortcuts stopped reaching
Orca. Install it in attachWorkspaceDocGuestPolicies, bound to the
preview's own host, with browserPage: false so browser-page chords stay
with the document, and dispose it with the rest of the doc policy.
* test(browser): pin Ctrl+Tab forwarding from a focused HTML document preview
* fix(browser): activate a document preview's split when it is clicked
A press inside the preview never reaches the host DOM, so forwarded
shortcuts such as Close acted on whichever split was active before.
* refactor(browser): read a document preview's page id from its guest policy
---------
Co-authored-by: agustin2606 <agustinvarela2606@gmail.com>
A switch made from Orca Mobile or the CLI opened the 'restart running sessions?' dialog on the desktop, where the person switching can't see it. Under the 'Ask' setting these switches now restart the moved sessions the way 'Always restart' does: each waits for its turn to end and resumes its conversation on the new account.
The phone's stale-resize filter took the binary scrollback frame's seq as
its layout high-water mark, but that seq counts output bytes, not layout
changes. A Desktop/Phone toggle on a terminal with little output then
looked older than the mark and was dropped, so the phone never resized
and kept showing the old mode. Seed the mark from the subscribed frame's
layout seq and leave scrollback out of it.
Co-authored-by: zhangxaochen <3416677+zhangxaochen@users.noreply.github.com>
An old mobile app lists a desktop-created structured chat it cannot render
with the title 'Update to view', which doesn't say whether the phone or the
desktop needs the update. Name the phone app, and document the symptom and
recovery steps in the mobile troubleshooting docs.
* fix(daemon): stop the macOS logout watch writing a login record every two minutes
The watch ran /usr/bin/login as its session-health probe, and each run adds an
entry to the macOS login history. Check DNS health first and only run the login
probe when the resolver already looks dead, which retirement requires anyway.
* fix(daemon): keep probing login after a rejection until PAM accepts again
Gate on the last login verdict instead of the rejection streak, so a sleep reset
of the streak cannot leave a rejected verdict cached while DNS looks healthy.
* test(daemon): keep coverage for not retiring once DNS recovers
* test(daemon): drop the duplicate retirement replay from the resolver gate test
* fix(agent-status): skipping a Codex async question clears it on SSH panes too
An SSH pane's Codex hooks are normalized by orcad on the host, so the
desktop's own hook server had no row to apply the skip keystroke to. The
renderer now sends a remote pane's skip to the owning runtime through a new
terminal.inferCodexQuestionSkip RPC, which applies it to the host's tracker.
* test(e2e): skipping a Codex async question clears a managed SSH terminal's row
The Source control empty state named the base by its fully qualified ref ('…no changes ahead of refs/remotes/origin/master'). Show it the way people know it ('origin/master') with a small shortRefLabel helper.
* fix(cursor): read the default marker in cursor-agent model rows
* feat(agent-launch): list a terminal-only agent's own models in the launch picker
* fix(agent-launch): cache Cursor listings and keep the seed for fallback lists
* fix(agent-launch): keep the host catalog for agents that have structured chats
* refactor(agent-launch): resolve the picker's discovery runtime once per render
* style(new-workspace): icon-only Create From tabs and shorter Create button
* style(new-workspace): let Tabs primitives own source-tab spacing
* chore(i18n): add Create button label to English catalog
* fix(session-history): index Codex rollouts archived as .jsonl.zst
Codex compresses rollouts idle for 7+ days to rollout-*.jsonl.zst. The
session scanner only discovered .jsonl files, so every Codex chat older
than about a week disappeared from Agent Session History. Discover .zst
rollouts, decode them through a shared zstd stream wrapper above
openTranscriptReadStream (native, WSL UNC, and SSH alike), never resume
a compressed archive from an offset, and drop a .jsonl.zst candidate
when its plain .jsonl twin is listed, matching Codex's own rule.
* fix(session-history): admit only rollout-named zst archives and close raw zstd streams early
Review follow-up on #40593: a filePredicate limits compressed discovery
to rollout-*.jsonl.zst so unrelated .zst files are never decompressed,
and zstdTranscriptStream links the raw and decoded streams with pipeline
so an early close of the decoded stream releases the raw stream's fd.
* fix(session-history): cover archived Codex rollouts by compressed size and collapse .zst twins in every listing
Archived rollouts keep the bounded resumable reader but never resume, and report their compressed stat size as the read cursor, so search coverage checks match. The .zst suffix is dropped from Codex alias keys with the plain file preferred, which also dedupes scoped walks and rollouts archived in only one home.
* fix(session-history): prefer the live rollout over an archive in another Codex home
---------
Co-authored-by: Kelvin Amoaba <97001695+AmoabaKelvin@users.noreply.github.com>
* fix(native-chat): open and reveal folder links instead of doing nothing (#40675)
* fix(native-chat): reveal folders in their owning worktree and localize the no-route toast (#40675 review)
- Activate a folder's owning worktree before revealing it: the explorer only
reveals inside the worktree it shows, so a paired host folder in another
worktree stayed hidden.
- Reject a paired folder outside the worktree root (e.g. a chat scratch
folder): the explorer drops those reveals, so report the unavailable route
instead of returning silently.
- Add an `unavailable` FileOpenFailure verdict with its own localized toast,
replacing the English error string that other locales would show.
* fix(native-chat): report a folder the OS could not open (#40675 review)
`openFilePath` returning false or rejecting left a local folder click with no
feedback. The directory route now reports the unavailable path in both cases,
with a test covering a refused local folder open.
* fix(native-chat): reveal SSH folders in the explorer and report a failed Shift-open of a folder
Folders below an SSH workspace root still ended in a 'can't open' toast;
the workspace explorer reads through any host, so it shows them. A
Shift-open of a local folder that the OS refused returned silently.
The trailing-separator hint now also hides the download row.
* fix(native-chat): match the owner's host and explorer spelling when revealing a folder
A same-ID workspace on another host counted as already selected, a
superseded folder click could still toast after a newer one, and an
SSH reveal used the chat's separators instead of the explorer's.
* fix(file-explorer): reveal a target spelled with other separators, and skip duplicate folder rows and toasts
The explorer built a reveal's ancestors in its root's spelling but
looked the target up as given, so a forward-slash path under a Windows
root selected nothing. A local folder link no longer repeats its open
as 'Open with default app', and a blocked folder workspace no longer
toasts twice.
---------
Co-authored-by: Pr1p <zwq19980411@gmail.com>
* fix(speech): keep the old model when the new one can't be moved in
Deleting the old folder before the rename lost both copies when the rename
failed, e.g. a Windows antivirus lock.
* fix(speech): keep an interrupted swap's backup and let delete wait for it
Retry renames through Windows file locks, and keep .previous when it is the
only old copy left.
* fix(speech): remove the migration source's model backup on delete
* test(speech): assert the kept model, not the backup folder name
* refactor(speech): remove each model folder's leftovers in one loop on delete
* fix(speech): check model folders with async stat during the swap
* fix(azure-devops): recognize vs-ssh.visualstudio.com SSH remotes
Organizations on the older visualstudio.com URLs clone over SSH with
<org>@vs-ssh.visualstudio.com:v3/<org>/<project>/<repo>. The repository
parser only accepted ssh.dev.azure.com, so these remotes parsed to null
and the repo got no Azure DevOps pull request or status features.
Accept vs-ssh.visualstudio.com in both the SCP and ssh:// forms, and
resolve it to the same reference the org's https://<org>.visualstudio.com
remote already gets.
Fixes #ISSUE
* fix(azure-devops): validate vs-ssh org names and fix renderer PR links
A vs-ssh remote's org becomes the API hostname that receives the Azure
DevOps PAT, so reject anything but a plain org name. The renderer's
manual PR link parser also missed vs-ssh remotes and built a link on
vs-ssh.visualstudio.com; resolve it to the org's visualstudio.com repo.
---------
Co-authored-by: Kelvin Amoaba <97001695+AmoabaKelvin@users.noreply.github.com>
* fix(orchestration): a worker report during an Orca restart is not rejected as a changed identity
* refactor(orchestration): share the startup bounded wait; skip long-poll checks
* chore(orchestration): drop casts in the restart identity wait
* fix(orchestration): only long-poll checks skip the restart identity wait
* fix(orchestration): ordinary mail never waits for the restart identity restore
Diagnostics only. Separates sleeping hosts (long gap) from awake hosts whose
connection was dropped (short gap) after silence closes. Stores one small
per-host key, cleared on the next activation and by erase.
* fix(bitbucket): connect access tokens and API tokens without account access
Bitbucket answers GET /2.0/user with 403 for a credential it accepts but
that has no account access: every repository, project or workspace access
token, and API tokens without read:user:bitbucket. Orca treated that 403 as
a rejected credential, so those tokens could never connect. Only 401 now
means rejected; 403 connects without an account name.
* test(bitbucket): type the 403 fetch stub without a cast
* fix(bitbucket): accept a /user 403 only when Bitbucket sent it
* fix(session-history): discover OMP profile session dirs
OMP stores per-profile transcripts under <config home>/profiles/<name>/agent/sessions
(XDG_DATA_HOME/omp/profiles/<name>/sessions on POSIX with XDG set). The app process
usually lacks OMP_PROFILE/PI_PROFILE, so profile-run sessions were missing from
Agent Session History, orca search, and native chat.
Enumerate existing profile session dirs in ompProfileSessionsDirs() (same env +
safety rules as resolveOmpSessionsDir; PI_CONFIG_DIR-aware; suppressed under the
legacy OMP_CODING_AGENT_DIR override) and include them in ompSessionsRootDirs()
(discovery + path allowlist) and the omp branch of the native chat resolver.
An explicit ompSessionsDir override still pins one location.
main-sync-fs baseline for omp-session-root.ts 1 -> 3: one readdir of the small
fixed profiles dir plus one existsSync candidate check per profile (single call
site), the same small-dir probe style the file already uses - not a per-file
sweep. Flagged for reviewer sign-off per the ratchet's rule.
* fix(session-history): enumerate XDG-only OMP profile dirs
ompProfileSessionsDirs only listed profile names from the local config home, so a profile that exists solely under $XDG_DATA_HOME/omp/profiles (e.g. OMP run with a different PI_CONFIG_DIR) was never discovered. Enumerate both roots; per-profile candidate checks and output dedup are unchanged.
* docs(session-history): document profile dir helpers for coverage
* fix(session-history): use non-blocking profile enumeration in the native chat ID lookup
* refactor(session-history): share one OMP root list for discovery and chat lookup
Drop the duplicated async profile scan: native chat ID lookup now reuses
ompSessionsRootDirs, which already runs the small sync profile probe on the
main process for row deletion. Reuse normalizedProfile for directory names
and trim tests to the behavior that fails on revert.
* fix(session-history): prefer a profile's XDG sessions dir in OMP ID lookup
---------
Co-authored-by: ducts <ducts@fpt.com>
Co-authored-by: Kelvin Amoaba <97001695+AmoabaKelvin@users.noreply.github.com>
* feat(automations): redesign the automation detail overview
Group the overview into a run-health band (last run, next run, spend), Schedule and Execution property panels, and a Prompt panel with copy and a fading show-more. Run Now is now the primary action next to Edit; Pause and Delete move into an overflow menu with Delete separated as destructive.
* test(e2e): align prompt disclosure spec with the 8-line clamp
The redesign raised the collapsed prompt clamp from 4 to 8 lines and caps the
prompt at 80ch, so the resize prompt no longer overflowed at 820px. Lengthen it
so it wraps to ~11 lines at 820px and ~6 at 1400px, and assert line-clamp 8.
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>