Swap the worktree-name emoji picker from emojibase-data's `github` shortcode preset to `emojibase`, which carries both `flag_kr` and `south_korea` style flag names, and drop the hand-maintained `kr` entry that patched around the gap. Filter skin-tone aliases so they neither crowd the suggestion list nor clobber base-emoji branch names.
Search now matches anywhere in the shortcode, ranked exact > prefix > word-start > substring, so `:korea` surfaces both Koreas.
Emoji-derived branch names now prefer spelled-out aliases: flags use country names (japan, germany, south-korea) and cryptic stubs are skipped (thumbsdown over no, victory over v).
Remove one-off incident docs and committed test-results noise, move
dev/repro/bench tools under tests/tools, and relocate i18next config
into config/ so the GitHub root scrolls to the description faster.
* perf(mobile): avoid unchanged worktree catalog payloads
* fix(mobile): isolate catalog snapshots by limit
* review: reassert host truth on unchanged polls; content-address snapshots
Client — the `changed` gate meant an unchanged poll skipped setWorktrees /
setLastKnownWorktrees / setCachedWorktrees, so optimistic local edits
(togglePin, handleDeleteWorktree's failure re-add) and the #8498 cache guard
were no longer repaired while the host catalog was stable. The gate bought
nothing: setCachedWorktrees is an in-memory Map write and areWorktreeListsEqual
already ran every poll, so the steady state still short-circuits on array
identity. All wire savings are unaffected. admit() now just returns the
confirmed rows and HostScreen applies them exactly as it did pre-PR.
Also on the client:
- a stale response from a superseded client/host no longer clears the token the
current client/host just established
- discriminate on `worktrees` rather than on `'unchanged' in response`, so a
future catalog field named `unchanged` can't reclassify a full response
- useRef over useMemo for the snapshot client; React may discard memoized values
- hoist WORKTREE_PS_FULL_LIMIT so the truncates-at-200 rationale travels with it
Host — replace the per-limit snapshot cache with a content-addressed id (ETag
semantics). Ownership lives in the id, so concurrent clients, differing limits,
and runtime restarts are correct by construction; this drops the LRU, the
eviction policy, the per-runtime WeakMap, and the retention of up to 8 full
catalogs. The remaining cache is a pure memo: because ids derive from content,
dropping or thrashing it costs CPU and nothing else. Keeping the memo also
keeps the measured steady-state cost — hashing every poll instead measured
2.24ms vs 0.75ms for the compare on a 310KB catalog.
Verified: mobile 2784 passed / 3 skipped, src/main/runtime/rpc 1064 passed,
node + mobile typechecks, oxlint, oxfmt, max-lines ratchet.
* fix(runtime): isolate catalog snapshot memo
---------
Co-authored-by: Brennan Benson <79079362+brennanb2025@users.noreply.github.com>
* fix(preflight): route landing banner through the runtime-aware preflight slice
Landing called window.api.preflight.check directly, which always probes the
local client. The preflight slice is the only caller that consults
getActiveRuntimeTarget and forwards to preflight.check on the active runtime
environment, so while connected to a remote runtime the landing banner
reported the client machine's git/gh state instead of the server's.
Delegate to refreshPreflightStatus and derive the issue list from
state.preflightStatus. This also drops Landing's duplicate probe: the slice
dedupes concurrent and forced checks, so the mount/focus/poll paths now share
one in-flight request with the rest of the app.
* fix(preflight): refresh landing status across runtime sessions
* test(preflight): cover paired runtime session races
* test: make landing preflight oracle behavioral
* fix(preflight): scope runtime session invalidation
* test(preflight): cover headed runtime switching
* test(preflight): isolate runtime status toast
---------
Co-authored-by: Marty <marty@localhost>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* fix(remote): unthrottle host renderer while serving a paired client
A paired desktop host left in the background could not open or close
agent sessions for its remote/relay client: the action stalled and
eventually failed with the host-side "Timed out waiting for terminal
surface after creation" (10s) error, while an already-live terminal's
keystrokes stayed fast.
Root cause: creating/closing a session routes through the host
renderer's setTimeout-coalesced graph sync to publish the terminal
surface, but the host window runs with Electron background throttling
(the hidden-window default, reaffirmed on macOS). When the window is
backgrounded/occluded, those renderer timers are throttled to a crawl
and the surface publication misses the 10s deadline. Live keystrokes are
unaffected because PTY I/O flows through the main process, never the
renderer.
Keep the authoritative renderer unthrottled while at least one remote
client is connected and restore the throttled power-saving default once
the last one disconnects. Connect/disconnect are driven from the shared
MobileSocketWiring onReady/onClose, so both direct-WS and cloud-relay
clients are covered; headless serve has no window and is a safe no-op.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* refactor(remote): tidy renderer-throttle comment and test per review
Address automated review nits on #11581:
- Trim the module-level rationale comment to the non-obvious contract,
matching the repo's concise-comment guideline.
- Drop the dead `detachedThrottle` variable from the reapply test; the
detached-target scenario is already covered by the lazy-resolution
test, so the case now asserts only what it exercises.
No behavior change.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* fix(remote): scope paired terminal publication throttling
Keep headed paired terminal creation and close renderer-owned so host inventory, input routing, ACK recovery, and cleanup retain the established lifecycle. Hold a reference-counted background-throttle lease only while the renderer publishes a paired operation, and epoch-fence async resolution so renderer reloads reject before any request or PTY spawn. Preserve headless main ownership and prevent paired clients from falling back to a local terminal.
* test(e2e): verify minimized host terminal repaint
* fix(remote): preserve paired terminal inventory through graph gaps
---------
Co-authored-by: fanyunqian.1 <fanyunqian.1@bytedance.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* [P1] fix(routing): keep the paired-HUB owner for the active workspace selection
#11346 made `activeWorkspaceExecutionHostId` authoritative for the active
workspace, but the short-circuits that read it never see the owner record, so a
project reached through a paired runtime HUB (`hostId: ssh:<target>` plus
`runtimeOwnerEnvironmentId: <hub>`) loses its transport and its repo:
- `resolveWorktreeOperationRouteResult` returned `runtimeEnvironmentId: null`
for any non-`runtime:` selection, bypassing the HUB recovery in
`resolveExactWorktreeRoute` — every owner-routed terminal/git/filesystem
operation on the active workspace lost the HUB that proxies the SSH target.
- `selectRepoByIdForActiveWorkspace` filtered `state.repos` by the selected host
with no fallback, but `withRepoHostOwnership` deliberately keeps the SSH host
on the worktree while the repo row stays HUB-owned — so the active repo read
as `null` and `useGitStatusPolling` stopped polling the whole workspace.
Route resolution now recovers the HUB owner from the owner rows on the selected
host, and drops it only when rival HUBs project the same host. The repo selector
opens exactly one hole in #11346's fail-closed rule — an `ssh:` selection over a
`runtime:` repo row — and every other host mismatch still returns `null`.
`worktree-owner-route.ts` is a pure extraction of the existing owner-route
helpers; no behavior moved with it.
* fix(routing): fail ambiguous paired-HUB repo selection closed
* feat(updater): name hourly releases by version, build number, time, and sha
Hourly releases were titled with their raw tag
(`v1.4.163-hourly.202607312054`), which reads as one opaque digit run and
does not say which commit it came from.
Title them `1.4.163 • 01 • 07-31 13:54 • e698241` instead, and show that
same string in the in-app build picker by having the picker render the
release's stored name rather than deriving its own label. Composing it in
one place means the two surfaces cannot drift.
The build number is monotonic across the channel. It is read as the
highest number already in use rather than as a count of releases: the
prune step trims to 72, so a count would roll backwards after three days
and reissue numbers. Drafts count toward it — unlike in the freshness
check, which asks whether a commit shipped, this asks whether a number is
free, and a stranded draft still holds one.
Times are Pacific while the tag's stamp stays UTC. The stamp is a sort
key and a local one would repeat an hour at every DST fall-back, making
two distinct builds compare equal; the title is only ever read.
* fix(updater): fail the hourly build when the release name is missing
The workflow checks out `ref: main`, but a workflow_dispatch runs the
workflow file from whatever branch was dispatched. A branch that edits
this step while main still carries the old script produces an empty name
and an untitled release — silent, and only visible once someone opens the
releases page. Verified by hitting exactly that on run 30665586904.
Co-authored-by: Orca <help@stably.ai>
---------
Co-authored-by: Orca <help@stably.ai>
* fix(codex): trust extended-length resume paths
* test(codex): cover both sides of extended-length resume folding
The provenance fix folds the extended drive spelling on the rollout path and
the trusted sessions root, and runs per entry in the legacy id scan. Only the
rollout-path side had coverage, so pin the rest:
- normal-form rollout under an extended-length trusted home (root side)
- compressed .zst sibling derived from an extended-length persisted path,
where a folded comparison copy could leak into the returned path
- legacy id-scan entries: extended-length accepted, device namespace refused
All three fail with the production change reverted.
* docs(codex): state accurately what the resume path fold rejects
The helper's comment called the rejected spellings "arbitrary device
namespaces", but \?\UNC\ is the extended-length form of \server\share\ —
a network share, not a device. Reading it as a security decision hides that
UNC is simply not folded yet, so say what is actually excluded.
Also trim the two new test comments to one line to match the file.
* fix(codex): reject rollout alternate data streams
* fix(codex): preserve resume provenance guard
---------
Co-authored-by: OrcaWin <alpha-eng@stably.ai>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* fix(orchestration): gate methods route calls to the caller's Run with `f
Gates are Run-scoped state; every gate command now resolves the caller's active Run
(via pane binding or explicit --from flag) and authorizes within that Run's scope.
Settled adopted work no longer requires --takeover-legacy, and the legacy coordinator
fence respects both binding-based and attestation-based proof of authority.
* fix(orchestration): gate methods route calls to the caller's Run with at
Gate and run methods now verify that declared terminal handles match the caller's
attested identity, preventing spoofing of other coordinators. Extracted shared
`resolveRunScope` to enforce one authorization rule across all orchestration
mutations. Added comprehensive regression tests for #11745.
* fix(release): restore the SignPath composite action when cutting from an older ref
Co-authored-by: Orca <help@stably.ai>
* fix(startup): record a durable diagnostic before the bootstrap fatal-exit guard exits
Co-authored-by: Orca <help@stably.ai>
* fix(settings): make agent-skill Re-check rescan skill freshness
Co-authored-by: Orca <help@stably.ai>
* fix(startup): keep the bootstrap fatal diagnostic when the log override is unwritable
Create the parent directory an overridden ORCA_BOOTSTRAP_FATAL_LOG names and fall
back to the default location when that path still cannot be opened, so a missing
parent no longer costs the only account of the failure. Also pins the Re-check
freshness rescan to the completed install scan rather than the click.
Co-authored-by: Orca <help@stably.ai>
* refactor(settings): move the post-recheck surface sync out of the panel
Co-authored-by: Orca <help@stably.ai>
* fix(startup): retain diagnostics without node fs
* fix(skills): keep freshness scoped to the local runtime
* fix(settings): register freshness status translations
* fix(settings): scope and sequence skill freshness refreshes
* fix(settings): refresh freshness across runtime transitions
---------
Co-authored-by: Orca <help@stably.ai>
* fix(relay): stop the retired exit record from double-broadcasting pty.exit
Retiring the publication record at cancel or grace-expiry time routed the pending
exit through pty-handler's unconditional broadcast, so subscribers that already
received the exit from the record's legacy projection got a second copy (which
fans out to a whole-provider teardown on the consumer side).
- Track the legacy exit projection in a RelayPtySourceLegacyExitIndex keyed by
pty incarnation so it outlives the delivery record; the handler's fallback now
re-targets only source-owner clients when subscribers already have the exit.
- Cover the previously untested defense layers: B2 same-client re-attach healing,
B3 deferred retire-on-append-failure (retire plus the microtask deferral), and
publishPendingExit's `accepts(id) &&` sub-guard.
- Harden layer D's capacity fan-out (it sits outside the catch it depends on),
route cancelExact's bare-timer probe through snapshotIfKnown, and keep
appendDenied counting cancel-induced append rejections.
* test(relay): pin the retired-record exit dedup and stop its index leaking
The legacy-exit dedup layer shipped with zero coverage: deleting pty-handler's
`publishExitAfterRetire` wiring left the whole relay suite green, because the
send scheduler's partial-progress branch keeps a projected record alive so B1
handles the common case and the fallback is never reached.
- Cover `RelayPtySourceLegacyExitIndex` directly, pin pty-handler's fallback
(a retired record re-targets its own exit instead of broadcasting a duplicate
to subscribers that already have it, and still broadcasts when nothing was
projected), and drive the reachable retire path: B2's re-attach retires a
record whose projection only the index remembers.
- Forget the index entry when `exitPublicationSettled` prunes a healthy exit.
`remember()` only re-ran from the exit path, which B0 short-circuits after a
settled exit, so every source-mode PTY exit leaked one Map row for the
daemon's lifetime and would re-publish on any later fallback for that id.
* fix(relay): contain retired exit publication faults
* fix(relay): retain projected exits across owner faults
* fix(relay): retire faulted exit deliveries
* fix(checks): stop skipped and manual checks reporting as failures
Route every check-classification surface through one shared helper so
desktop renderer, desktop main and mobile agree on the same verdict.
- GitLab `manual` jobs and pipelines are neutral again, not action_required/failure
- `skipped` counts as passed everywhere, including mobile
- a neutral check no longer demotes a summary that has passing checks
* fix(checks): move the check-classification parity test into the renderer project
The parity table lived in src/shared but imported a renderer module, and both
config/tsconfig.node.json and config/tsconfig.cli.json are composite projects
that include src/shared without that renderer path, so `pnpm typecheck` failed
with TS6307 on two of its three projects. Only the web project spans both trees.
Co-authored-by: Orca <help@stably.ai>
* fix(checks): stop the Tasks-grid pill contradicting its own verdict
The checks pill's label, tone and icon all read one ProviderCheckSummary, but
getChecksLabel short-circuited on the raw `neutral` counter while the tone and
icon key off `state`. After the classification fix a PR with 19 success + 1
neutral renders an emerald CheckCircle2 pill that reads "1 unresolved", and
mobile's own label (which keys off `state`) reads "19/20 passed" for the same
summary.
Move the label into src/shared/provider-check-summary.ts so desktop and mobile
cannot fork it again, and key it off `state`.
Also covers deriveWorkItemCheckSummary, the desktop-main producer of the summary
that reaches the Tasks grid and the relay-paired mobile client. It was rewritten
here with no test at all; the parity table stands in derivePRCheckStatusFromRollup,
which is a different normalizer. The new main-process test drives getWorkItem with
a real statusCheckRollup fixture, pinning the StatusContext `state` fallback that
would otherwise be deletable with the whole suite still green.
Co-authored-by: Orca <help@stably.ai>
* fix(gitlab): route the pipeline job-array rollup through the shared check classifier
The array path in derivePipelineStatus kept its own copy of the rollup rules, so
manual-only read green and one unrecognized job status demoted a passing pipeline
to neutral — both disagreeing with every other check surface.
Also retry the packaged-CLI smoke temp cleanup on Windows: the copied Orca.exe can
still be locked by AV/indexers after every assertion passed, failing the package job.
Co-authored-by: Orca <help@stably.ai>
* fix(gitlab): stop the skipped pipeline string diverging from the Checks tab
- classifyPipelineString now counts a skipped pipeline as passing, matching
the per-check classifier; canceled stays neutral and is pinned as an
explicit, sign-off-pending divergence.
- Pin the production string path (head_pipeline.status) in the parity table
and note that the job-array branch has no production caller yet.
- Count skipped checks in the Checks panel's passing header so it agrees
with the checks pill.
- Correct the packaged-CLI smoke retry comment: the EBUSY is the smoke's own
just-exited Electron process, not AV/indexers.
Co-authored-by: Orca <help@stably.ai>
* fix(checks): finish cross-surface check parity and back out the skipped MR-card flip
Review follow-ups on the check-classification PR.
- PullRequestPage and GitHubItemDialog kept private copies of getCheckCounts /
getChecksSummaryLabel that still counted only `success` as passing, so a
2-success/3-skipped PR read "2 passing · 3 skipped" there and "5 passing" in
the sidebar. Both copies move to pr-check-counts.ts, which routes the passing
bucket through classifyCheckOutcome; action_required keeps its own amber
bucket. The summary icon now keys off passing count, so an all-neutral PR
stops painting a green tick above "0 of N checks passing".
- The sidebar checks header and triage strip still called
`{status: completed, conclusion: null}` pending, contradicting the grey
"Unresolved checks" pill. Both now read summarizeProviderChecks and render an
unresolved chip/strip instead of an amber spinner that can never resolve.
- classifyPipelineString('skipped') is reverted to neutral. That flip painted
MR cards green for pipelines that never ran, on the only GitLab path with
production callers, and contradicted the same function's deferral of
`canceled`. Both tone changes stay deferred, pinned by one test.
- classifyPipelineString('manual') resolves to pending rather than neutral: a
blocked pipeline is outstanding, and neutral let the worktree card fall
through to its emerald `open` default while GitLab still refuses the merge.
- TaskPage's checks pill helpers move to task-page-checks-pill.ts so the
"1 unresolved on a green pill" fix is actually pinned by a test.
- smoke-packaged-cli no longer lets an EBUSY cleanup replace the real failure.
* fix(checks): stop completed unknown checks from spinning
---------
Co-authored-by: Orca <help@stably.ai>
* fix(daemon): keep audit-only evidence off the main thread and out of false negatives
Three audit-only daemon evidence defects:
- The Windows CIM probe emitted the "absent" marker whenever $p was empty, so a
non-terminating Get-CimInstance failure (Winmgmt down, corrupt WMI repository,
access denied) exited 0 with {"exists":false} and was recorded as proof the
daemon process was gone. The script now sets $ErrorActionPreference and reports
query failure explicitly; only a query that ran and found nothing yields
'missing', everything else stays 'unavailable'.
- probeMacosProcess read the process start time through execFileSync('ps') on the
Electron main thread. It now uses the async readMacosProcessStartedAtMs, matching
the async command-line sibling in the same probe.
- daemon_audit_eligibility fired on every successful inventory, so routine
listProcesses traffic could drain the shared 1,000-event per-session telemetry
ceiling. Repeated identical observations now collapse to one heartbeat per
5 minutes; a changed observation still emits immediately.
The evidence stays strictly audit-only — no lifecycle, routing or recovery path
consumes the observation.
Co-authored-by: Orca <help@stably.ai>
* fix(daemon): take the linux audit start-time read off the main thread
The linux branch of the audit-only evidence probe still bottomed out in
getProcessStartedAtMs, i.e. two readFileSync calls plus an
execFileSync('getconf', ['CLK_TCK']) spawn on the Electron main thread —
the same hazard the darwin half of this change removed. The `async`
wrapper hid it from the type signature and from the main-thread
regression test, which only covered darwin.
That path is reached whenever the pid record carries no start ticks or
boot id: legacy bare-integer pid files (the upgrade population this audit
exists to observe) and any host where /proc boot-id or start-tick reads
fail.
readLinuxProcessStartedAtMs now reads procfs asynchronously and caches
CLK_TCK for the process lifetime (it is fixed for the kernel's lifetime),
retrying only after a failed probe. The main-thread test now covers linux
with a ticks-less pid record.
Co-authored-by: Orca <help@stably.ai>
* fix(daemon): measure the audit heartbeat window on a monotonic clock
A backward wall-clock jump (NTP correction, VM resume) made elapsed time negative, which read as "still inside the window" and suppressed daemon_audit_eligibility heartbeats until wall time caught up. Rate-limit arithmetic now runs on performance.now() and treats any backward movement as eligible.
Co-authored-by: Orca <help@stably.ai>
* fix(daemon): guard the audit rate limiter and scope the CLK_TCK cache
The rate-limited tracker is the only production emitter, but its payload
serialization and clock read sat outside the guard that keeps audit
telemetry from affecting daemon availability — a throw there propagates
out of listProcesses' try and turns a successful inventory into a
reported inventory failure.
Also key the CLK_TCK cache by the runner that produced it: the value
belongs to the host executing getconf, not the module.
Co-authored-by: Orca <help@stably.ai>
* test(daemon): make audit CI synchronization deterministic
---------
Co-authored-by: Orca <help@stably.ai>
* fix(orchestration): route the legacy gate at the caller's own Run
The retained-legacy-coordinator gate treated an unnamed Run as the adopted
Run, so callers with no relation to it were fenced with legacy_read_only,
and the adopted Run's NULL coordinator made the owner escape hatch
unreachable.
Resolve the caller's bound Run first and keep the adopted Run only as the
unbound fallback, and treat an unclaimed adopted Run as free — the same
rule bindingMatches() already applies 100 lines down.
* refactor(orchestration): pass the open db handle into boundRunId
Co-authored-by: Orca <help@stably.ai>
---------
Co-authored-by: Orca <help@stably.ai>
The move of hasUnsafeWindowsBatchSyntax into src/shared/windows-batch-spawn.ts
silently added `(` and `)` to the cmd.exe denylist, so every .cmd shim or
argument path containing parentheses became unspawnable across nine call sites.
Parentheses only group commands and cannot chain one without a separator the
guard already rejects, so they are dropped again.
The rejected character set is now the single source for the user-facing error
strings, and `orca account add` translates the sentinel into a real message.
Co-authored-by: Orca <help@stably.ai>
The sidebar->board drop translation still read the mounted DOM cards, while
getCardDropTarget now indexes the lane's virtual layout, so a searched and
scrolled lane resolved the drop to the wrong full-lane position.
Co-authored-by: Orca <help@stably.ai>
Neil
Jinjing
OrcaWinandNeil
fsdwen
Rod BoevandOrcaWin
LailyandOrcaWin