* fix(terminal): return IME composition ownership to xterm
* fix(mobile): derive terminal input from native replacement ranges
* test(mobile): record iOS Japanese IME traces
* fix(mobile): preserve native IME replacement ranges
* fix(xterm): flush queued application input after IME commit
* test(terminal): pin Korean intermediate commit
* test: pin Windows IME shortcut ownership
* test: replay IBus number candidate commit
* fix: preserve native macOS input-method punctuation
* refactor(terminal): remove stale mac focus override
* fix(mobile): preserve soft keyboard deletion ranges
* fix: keep IME-owned palette chords in renderer
* fix: stop carried IME shortcuts at renderer owner
* fix: preserve carried IME shortcut dispatch
* fix: narrow main-owned shortcut actions
* test(mobile): pin Japanese IME replacement traces
* test(terminal): retain paired native IME trace
* fix(chat): preserve browser IME composition ownership
* fix(chat): retain macOS IME confirm gesture
* fix(chat): expire unmatched IME confirm carry
* fix(chat): isolate IME confirmation expiry
* fix(chat): retain active IME confirmation
* refactor(terminal): remove dead composition handler
* feat(ime): add shared Enter-ownership seams for CJK composition
The confirming Enter of a CJK composition arrives as two keydowns and the
orderings differ by platform: Windows/Linux redispatch the unmarked Enter/13
before keyup, macOS delivers keyup first. A guard reading only isComposing or
keyCode 229 misses the redispatch, so surfaces submitted on a confirm.
Adds useImeEnterGestureOwnership (carry token, next-frame expiry), a shared
ImeEnterGuardedForm for native implicit submission, and the cmdk seam covering
18 CommandInput surfaces at one site.
A chorded Enter arms the carry but is never swallowed — the reverse would eat a
user's deliberate Cmd/Ctrl+Enter. Both failure modes are pinned by
ime-enter-gesture-ownership-contract.test.ts.
Co-authored-by: Orca <help@stably.ai>
* refactor(terminal): consolidate native input listeners and parked-screen owner
Extracts the shared native-input listener installer and renames the parked-screen
detector for what it actually does, replacing per-call-site duplication. The
listener installer keeps a forgetOptionKeyLocationOnBlur flag so per-window
semantics are preserved rather than flattened.
Net deletion; no behaviour change intended.
Co-authored-by: Orca <help@stably.ai>
* test(terminal): pin recorded IME shapes as regression tests
Nine regression tests built from hashed affected-platform captures, each with a
paired ordinary negative and a discriminating mutation verified to take the file
from all-passing to exactly one failure.
Covers the Windows MS-Korean Shift family (#12179, #11878, #12151, #11946,
#12152) and the Korean TUI line-break rows (STA-3237, STA-3222, STA-3129).
STA-3237 pins the empirical 3-Shift / 2-active-composition / 2-newline ratio the
device run established — the third Shift produces nothing because Space has
already committed. That ratio is not derivable from a static capture.
Co-authored-by: Orca <help@stably.ai>
* fix(ime): guard Enter-commit surfaces against CJK confirm
Applies the Enter-ownership guards across the surfaces whose Enter commits
something: publishes, clones, pairs, installs, posts, or persists.
Tiered deliberately rather than uniformly. Irreversible and remote-effect sites
take the carry token, which also blocks the unmarked redispatch. Locally
reversible sites take the oracle check with a one-line comment naming the
residual, because a spurious commit there costs one undo.
Three numeric fields are left unguarded with the reason in-code: Chromium blanks
number inputs at compositionstart, so a confirm-Enter only ever reaches an
empty-draft reset. Measured with a CDP probe rather than assumed — a guard that
cannot fire is noise.
Co-authored-by: Orca <help@stably.ai>
* test(ime): teeth-check the Enter guards on every guarded surface
One suite per guarded surface, each verified by deleting the guard and
confirming the test fails. A green guard test without that check is unverified,
not verified.
Two shapes pass vacuously in happy-dom and are avoided here: native implicit
form submission never fires, and blur() is inert on an unfocused element. Both
made "the commit did not happen" assertions pass with the guard removed, so the
suites assert the guard's contract directly instead.
Co-authored-by: Orca <help@stably.ai>
* fix(mobile): keep iOS Korean commits whole through the live-input path
iOS Korean reports isComposing: false on every event, so it bypasses the
composition guard entirely. The strict owner rejected UIKit's transformed
post-change field and sent only the leading jamo — the reported symptom.
Prefers the authoritative same-event field text over the predicted text when the
supplied operation cannot produce it. Generic: no Korean special-case, no locale
classifier, no normalization. Adds the RN-target-keyed submit carry alongside it.
Co-authored-by: Orca <help@stably.ai>
* test(e2e): make IME capture harnesses fail loudly instead of silently
Four instruments recorded silence as success, so a void run scored as a clean
one:
- readTerminalImeBoundaryTrace returned an empty trace when the probe never
installed, making every "nothing leaked" negative pass vacuously
- summarizeLatencies([]) returned a perfect zero distribution that passed all
three latency thresholds
- the macOS Vietnamese spec pinned an input-source ID that does not exist, and
failed as though the operator had chosen the wrong source
- the expectedLineCount=1 prefix property was undocumented and one edit from
silently downgrading a PTY assertion
Input sources now resolve by enumeration and name the near-matches on failure.
Co-authored-by: Orca <help@stably.ai>
* test(terminal): cover Cangjie cancellation and fix a cross-namespace assertion
Adds #11951's recorded Cangjie cancel shape to the existing cancellation suite,
which covered Pinyin and Sogou but not Cangjie. One keystroke then Backspace
arriving as deleteContentBackward with data: null, so the stale preedit is the
only thing a fallback could replay.
Verified against the historical pre-6cd944c62b3 bundle: the positive fails with
['尸'] where [] is expected, while the ordinary negative stays green.
Also fixes the Vietnamese spec, which asserted a TIS-space input-source ID
against getKeyboardInputSourceId(). Those two Orca APIs report the same source
in different namespaces — TIS nests it under VietnameseIM, the app API does not.
The resolver stays as an installation precondition; the assertion matches the
leaf.
Co-authored-by: Orca <help@stably.ai>
* test(e2e): add a real-IME macOS arm for the Korean chord commit
The existing korean-ime-terminal-shift-enter-commit spec synthesizes composition
over CDP: Input.imeSetComposition sets the preedit directly and Input.insertText
performs the commit. Asserting the IME produced events you injected yourself is
circular, so that spec cannot certify real-IME behaviour.
This arm selects 2-Set Korean via TIS, reads it back live, and injects through
System Events key codes, so the OS owns the preedit, the commit instant, and
isComposing. PTY byte expectations are preserved verbatim.
Covers 2 of the original 4 cases by design. The other two are the Windows/Linux
redispatch-before-keyup ordering, which macOS cannot produce and which cannot be
selected -- the OS decides it. Reintroducing synthesis to "restore coverage"
would reintroduce the circularity.
Co-authored-by: Orca <help@stably.ai>
* test(e2e): assert the macOS chord arm at the PTY boundary, not the renderer
The byte expectations were transcribed from korean-ime-terminal-shift-enter-commit
:364/:383, which assert against onData -- a renderer boundary where the terminator
is CR. This spec reads the PTY child, where the tty has already converted CR to LF.
Names both forms per row rather than swapping the constant, so the conversion reads
as evidence that the capture reached past the renderer, as #11936 and #11951 record.
Ctrl+Enter's CSI-u sequence is unaffected and is identical at both boundaries.
Co-authored-by: Orca <help@stably.ai>
* test(e2e): measure composer-to-onData latency and stop dropping IME keystrokes
Two defects in the echo latency probe.
It hooked onWriteParsed and onRender but never onData, so it measured
key->parse->render echo rather than the composer-vs-onData delta the latency rows
need. Adds a third hook feeding its own sample set.
And `event.key.length !== 1` silently dropped IME keystrokes: Pinyin and Cangjie
keydowns arrive as key:'Process' (length 7). Replayed over the captured corpus,
the old filter accepted 580 of 4137 Chinese IME keydowns -- it was discarding 80%
of them. The new filter matches the shape the owner itself branches on.
Attribution charges each onData to the latest keydown rather than a FIFO head,
because composing jamo emit no onData at all and a queue would credit a whole
composition to its first keystroke. The consumer now asserts sample count before
any percentile, so a zero-sample run cannot render as a flawless distribution.
Co-authored-by: Orca <help@stably.ai>
* test(terminal): pin the WSL shifted-jamo newline shape for #11919
In Korean 2-set, Shift types ordinary letters -- the double consonants and the
compound vowels. Each such keystroke reaches Chromium as key='Process',
keyCode=229, shiftKey=true.
The v1.4.163 classifier matched exactly that pattern with no code guard, so it
called those keystrokes Enter, rewrote them to a synthetic Shift+Enter, and
injected a newline into the middle of the word -- with no Enter key pressed.
That is why the reporters said "no modifier key pressed": they had not chorded
Shift+Enter, but they had pressed Shift, to type the double consonant.
Asserts the row's own recorded capture: 40 immediate keydowns, exactly 3 of them
Shift-carrying inside a single syllable, and an onData stream with one newline
per Enter press and none mid-word. Two ordinary negatives keep it from being a
blanket mute -- the same session's non-IME keydowns still reach shortcut policy,
and an ordinary Shift+Enter still resolves through the real policy.
Co-authored-by: Orca <help@stably.ai>
* test(terminal): pin the composition commit lag that made Korean type one behind
macOS Korean 2-Set commits syllable N only when the first jamo of N+1 arrives, so
compositionend and compositionstart land in the same task. A composition-start
handler cancelled the pending finalizer that was the only path to triggerDataEvent
and ended the session without emitting bytes, so every committed syllable reached
onData exactly one syllable late and the backlog cleared only at a Space or Enter.
Types continuously with no Enter and no Space -- either would flush the backlog and
hide it -- and samples onData at every syllable boundary. Paired with a
length-matched ASCII arm that stays green throughout, so the positive is a fact
about composition rather than about timing in general.
Bisected to a single call site across five builds: pristine, 1.4.155 and 1.4.162
pass, 1.4.163 fails, removing the one call repairs it, restoring it fails
identically. That window is exactly the reporter's "started immediately after
updating".
Co-authored-by: Orca <help@stably.ai>
* test(mobile): cover the send-queue abort that silently drops queued keystrokes
One failed send in use-terminal-live-input-commit aborts every keystroke
queued behind it, with the error swallowed by .catch(() => false). The
existing test resolves(true) on every send, so the failure branch was
uncovered.
Four arms: the abort itself, an ordinary negative on the healthy path, a
throwing sender, and a liveness control proving the queue recovers once
the chain settles. Deleting the abort takes 4 passed to 3 failed, with the
ordinary negative correctly surviving.
Scope is stated in the docblock: this is a transport send-queue abort,
reachable only via a real disconnect or RPC error. REQUEST_TIMEOUT_MS is
30s, so latency alone cannot reach the branch — consistent with #7094's
symptom class, not proven to be its cause.
* test(terminal): pin that daemon snapshot/restore cannot disturb a composition
Two independent reporters attributed broken Korean composition to the
always-on PTY daemon repainting terminal state over the preedit. The
attribution is wrong on ancestry — the daemon shipped three months before
the version both call good — but the boundary was never actually tested.
Runs the real applyMainBufferSnapshot choreography against a live
composition, including the full 2J/3J/H wipe plus the resize and
alt-screen branches. textarea.value, selectionStart/End,
compositionView.textContent and .active all survive byte-identical, and
interleaving a restore between every jamo of 문제 still commits 문제 at
onData. Also pins that the uncommitted preedit is absent from the captured
snapshot: it lives in the textarea, never the buffer, so a restore has
nothing stale to echo back.
Injecting one textarea.value = '' into the restore fails exactly the three
restore-boundary tests.
* test(terminal): pin that Cmd tears down a composition where Ctrl and Shift do not
xterm's composition keydown exempts only keyCode 16/17/18 (Shift/Ctrl/Alt)
plus 20/229. macOS Meta — 91/93/224 — is absent, so a Cmd press mid-composition
takes _finalizeComposition(false): the overlay goes dark and never recovers,
because compositionstart is not re-fired. The user composes the rest of the
word blind. Linux and Windows users press Ctrl and are exempt.
xterm already has a Meta-aware modifier predicate in wasModifierKeyOnlyEvent,
so this is an internal inconsistency rather than a deliberate choice.
Owns no reported row and is version-neutral: 5/5 on both 1.4.162 and 1.4.163.
The branch is unexercised in all 328 recorded traces, so this is a hazard pin,
not a regression guard. Only the teardown is asserted; the likely duplicated
commit needs a compositionend the IME kept alive across the Cmd, which no
capture contains.
Deleting the exemption fails exactly the three paired negatives; adding Meta
to it fails exactly the two Cmd arms.
* test(native-chat): characterize preedit loss when a question card replaces the composer
An AskUserQuestion card fully replaces the composer by design, but the
in-flight composition goes with it: the composer unmounts before
compositionend reaches it, so the preedit is never committed to the draft.
The committed text survives only because the draft is cached and restored
via defaultValue. Node identity changes, value 'abc' is preserved, the 가
is gone.
Drives the real NativeChatView -> SessionGate -> InteractiveCard ->
questionActive swap -> Composer -> ComposerField, flipped by writing the
same store field an AskUserQuestion hook event writes. Flipping
questionActive to false fails exactly this test and nothing else across
639 native-chat tests, so the path was entirely unguarded.
CHARACTERIZATION TEST: it asserts the loss. Fixing the defect — committing
the preedit before the swap, or keeping the composer mounted — will make
this file fail. Update the expectations to the new contract rather than
working around them.
Owns no reported row. #12118/STA-3219 flicker is keyed to token counters,
which provably do not remount, and a question card arrives once per
question.
* test(terminal): pin the duplicated commit when Meta interrupts a composition
_finalizeComposition(false) sends textarea.value.substring(start, end) but
cannot clear the IME-owned textarea, so a later compositionend re-sends the
same range. Meta reaches that path because CompositionHelper exempts only
Shift/Ctrl/Alt; xterm's own wasModifierKeyOnlyEvent covers Meta four ways,
so the omission is an internal inconsistency rather than a choice.
Companion to the modifier-exemption guard, which deliberately pins only the
overlay teardown. This pins the data consequence.
HAZARD PIN: owns no reported row. The trigger is unverified on hardware —
no capture in the corpus contains a Meta-during-composition gesture, and
whether macOS keeps the composition alive across it is unmeasured. The
duplication follows from the code given that sequence; whether users reach
the sequence is the open half.
An earlier premise that Space (keyCode 32) reaches this path was refuted by
a corpus scan: 0 of 731 evidence files carry a keyCode-32 Space while
composing, against 171 at 229, and 229 returns early.
* test(terminal): characterize the syllable lost when the textarea blurs mid-composition
CoreBrowserTerminal._handleTextAreaBlur clears the helper textarea
unconditionally — "Text can safely be removed on blur" — while
CompositionHelper._finalizeComposition reads the committed text back out of
that same value from a deferred timeout. By the time it runs the value is
empty, the substring is '', and triggerDataEvent never sees the syllable.
xterm checks composition state in _syncTextArea and omits the same check
here.
Six cases. Blurring mid-composition loses the syllable in every ordering,
including compositionend-before-blur, which is Chromium's real order — so
it is not an ordering artifact. A bare textarea.blur() with no Orca code
loses it too, which places the owner upstream: Orca's unguarded release on
outside pointerdown is one trigger, not the cause. Committing 한 then
blurring mid-가 yields ['한'] where ['한','가'] is correct: one syllable
gone, surrounding text intact.
Teeth checked by inverting — adding an Orca-side composition guard flips
exactly the three cases that route through the release path and leaves the
bare-blur and no-blur cases green, which is the scope split: a fix in
regular-terminal-focus-ownership alone would not close this.
HAZARD PIN, but unlike the others this one has a real production injector —
clicking outside the terminal mid-composition. Owns no reported row. The
shape matches #9738's report; the injector does not, and a shape match with
a mismatched injector is not an owner.
* test(terminal): say which arm the STA-3237 fixture came from
The recorded keydowns are wave 4's A-shift-unmarked-only — the arm that
emits no PTY bytes. Nothing in the file said so, so two readers concluded
the row's events fail the owner's predicate and that STA-3237 and STA-3222
were different defects. They share an owner; the arm that fires is
Process/229+Shift, absent from this bubble-phase trace because the owner
claims it in the capture phase.
Also corrects "code-blind": the v1.4.163 policy emits \x1b\r only for a
shift-only key:'Enter', and a jamo keydown reaches that branch solely via
the isTerminalImeProcessEnter rewrite. The mock is deliberately wider so
the ownership guard stays under test if that rewrite moves.
Comments only — no assertion, fixture value, or mock behaviour changed.
* test(e2e): track the input-source selector the macOS specs shell out to
Five tracked macOS IME specs ran `swift .tmp/select-input-source.swift`, a
file that is gitignored and existed only on one machine. Anyone else
checking out the repo — or the same machine after .tmp is cleaned — could
not run them, and they are the capture drivers for the macOS rows that are
blocked waiting for exactly those runs.
Moves it to tests/e2e/ beside its callers. The chord spec now resolves it
from __dirname rather than reaching two levels up into .tmp.
* test(terminal): pin the CJK repaint decision against the reporter's own output
#12164 comment 1 and #5921 report agent output with double-width glyphs
rendering duplicated character-by-character while ASCII in the same line
stays clean. No IME, no composition, no keystroke — the user never types
the CJK.
Segmenting all three verbatim samples into maximal same-risk-class runs
gives 33 runs and zero violations of "this run is corrupted iff the
production detector flags it": 17 wide runs all corrupted, 16 narrow runs
all byte-identical. The paired negative is co-located in the same line
rather than in a separate run — the reporter supplied it without knowing.
Doubling is asserted as present, not uniform: 자바스크립트 and 시스템 each
leave a jamo undoubled, which is a repaint-region boundary artifact rather
than a per-character transform.
The discriminating arm is in the test rather than a source mutation:
be3f30e2f8 (#6890) elects a repaint for all 17 corrupted runs when the
agent types nothing, and reverting its disjunct elects none. Both
predicates agree once the user has recently typed, which is the pre-#6890
condition.
Samples inlined with per-sample SHA-256 because .tmp is gitignored and
cannot back a landed test.
* test(terminal): pin macOS period substitution landing after the composition
#11504's reporter published a DOM trace showing insertText ". " arriving
149ms after compositionend, when two spaces are typed with a CJK input
source and NSAutomaticPeriodSubstitutionEnabled is on. This replays that
trace against a real Terminal and asserts what reaches onData — bytes to
the PTY, not anything visual.
The owner is stock upstream CoreBrowserTerminal._inputEvent, not an Orca
module, confirmed at the resolved install and in the shipped bundle Vite
loads rather than in the TypeScript source.
Three mutations against that install, predictions written before the runs,
each failing exactly the arms predicted: dropping the composed/keyDownSeen
guard fails two, dropping Orca's intercept fails the one arm where the
payload arrives before the send window drains, and flipping || to && —
the candidate-fix shape — fails the arm that pins the defect itself.
CHARACTERIZATION: arm 1 asserts the broken behaviour and will fail the
moment #11504 is fixed. Update it to the new contract rather than working
around it.
composed is absent from every recorded bundle, so composed: true is the
spec-required value rather than a captured one; the test asserts it before
dispatching so a harness that dropped the field fails loudly.
* test(terminal): replay the recorded Windows Shift sessions through the IME guard
STA-3179 reports a Shift release sending Enter; #12171 reports delayed
Hangul plus doubled newlines. Both replay their own recorded Windows
MS-Korean keydowns through resolveTerminalKeyboardShortcutAction with the
shortcut policy mocked, so the assertions are about which events reach the
policy and what reaches terminal input.
STA-3179's held-Shift gesture yields exactly one newline, from the unmarked
Enter alone; its release arms nothing for the next composition, asserted
after a precondition check that the release really is keyups with shiftKey
already dropped; and an ordinary Shift press-and-release still routes every
keydown, which is the paired non-IME negative.
Teeth, verified by mutation: bypassing the isImeOwnedKeyboardEvent guard in
keyboard-handlers takes STA-3179 from 3 passed to 2 failed / 1 passed — the
survivor being the ordinary-session negative, which is correct, since a
non-IME session should not depend on that guard — and #12171 from 2 passed
to 2 failed. Source restored byte-identical.
Recorded shapes are inlined and the bundles cited in comments; nothing is
imported from .tmp, which is gitignored.
* test(native-chat): correct 61977d45177 — the preedit survives the question card
61977d45177 claimed a composed syllable vanishes silently when an
AskUserQuestion card replaces the composer, and characterized that loss.
The claim was false. Its premise was an artifact of the harness: the test
simulated a preedit with a silent textarea.value assignment and no input
event, which no IME does.
Real composition fires input with insertCompositionText on every keystroke
— the shape this repo already records in its own observed-event capture —
and React's change handler returns on input/change with no composition
gate, so onChange runs for each frame. The draft cache is written
synchronously inside the updater, so the preedit is already committed
before the card can arrive. Driven that way, it survives.
Renamed to match the contract that actually holds, and extended: Hangul
jamo-per-frame, Japanese kana accumulation followed by per-segment
conversion asserting the candidate the user was looking at survives, and a
pin on the mechanism itself — the draft cache holds the preedit while the
card is up.
Teeth: there is no fix to revert, so the mutation is the plausible wrong
one — gating onChange on isComposing(). That takes 4 passed to 3 failed,
with the English negative correctly surviving, since it has no composition
to gate.
Two consequences remain, recorded rather than fixed: the OS aborts the
composition when the field disappears, so a lone jamo returns as a
compatibility jamo the user cannot compose onto, and the remounted
composer is unfocused because the card owned focus.
* docs(native-chat): name the corrected commit and the degraded-jamo consequence
Records in the file itself that 61977d45177 is pushed and wrong, quoting
the two claims that are false, so a reader who finds it in git log reaches
the correction from the file that replaced it.
Also states the residual as a consequence rather than a curiosity: a lone
leading jamo returns as a standalone compatibility jamo (U+3131), which is
not a composable state — the user cannot resume the syllable, only delete
and retype. Preserved, but degraded into something unusable. That is the
note to find if a reporter ever describes exactly that.
The invariant these tests pin is not "the composer commits on unmount" but
"composition input events must reach React" — which is what a future IME
change would break, and is not visible from the swap site at all.
* test(terminal): replay the recorded macOS Telex commit boundaries
#6905 reports Vietnamese composed characters breaking in the terminal.
Replays the retained macOS built-in Simple Telex capture — recorded
selection and value set before each dispatch, since that is what the
commit range reads — and asserts what reaches onData: the first commit
alone, then through the real Enter, then the ASCII tail of the same run.
A code-point count would catch NFD normalisation.
ENGINE CAVEAT, stated first in the docblock: this is macOS built-in Simple
Telex, Telex only. The reporter's three named engines cannot run on the
platform they declared, and which macOS Vietnamese engine they used is
unconfirmed. This file certifies no engine, and does not imply VNI.
The owner is upstream's — CompositionHelper._finalizeComposition's
waitForPropagation branch — so the arms are copies under .tmp aliased by a
scratch config, with node_modules verified unchanged by shasum after every
run. Collapsing the range end onto its start fails all three; collapsing
the start to zero re-emits the first word into the second commit, which is
the reporter's "duplicated" direction. Different failure sets, so the
mutants are distinguishable rather than merely detectable, and the ASCII
assertion passes under both.
Falsifiability here is by mutation, not by a defective build: #6905 does
not reproduce at HEAD, so this has never been watched going red on a real
reproduction.
* docs(terminal): lead the #6905 test with its engine caveat
Comment-only. Moves the caveat above the source line so a reader meets what
the file does NOT establish before what it does — the capture is macOS
built-in Simple Telex, the reporter's named engines cannot run on the
platform they declared, and which engine they used is what gates this row.
Co-authored-by: Orca <help@stably.ai>
* docs(terminal): record that the swallow eats a keystroke after Japanese conversion
This pin framed the swallowed insertText around Cmd interrupting a
composition. A differential through Japanese multi-segment conversion shows
it is broader: type a segment, convert, then press `a`, and the `a` is
lost. No modifier, no exotic gesture. Korean surfaced it first only because
2-Set composes on nearly every keystroke.
Also records why it cannot simply be fixed. The suppression de-duplicates
IMEs that deliver their commit a task after compositionend, which a sibling
test pins; this swallow is that dedup's false positive, and the two events
differ only in payload, so no flag-timing change separates them. Both a
smaller redesign and a content-aware variant were built and measured — the
first duplicates on IBus, the second costs a reported row's test and is
blocked while the patch cannot be regenerated.
The Japanese arrays behind this are authored, not observed: no Japanese DOM
composition trace exists in the corpus.
* docs(terminal): a Japanese capture does exist — correcting dbeecb11bee
That commit said no Japanese DOM composition trace exists in the corpus.
False. One does, filed under the Linux bundles rather than the bundle named
for Japanese: 30 DOM events, two にほんご->日本語 conversions, with full
selection state per event. It is retained byte-identically in three further
bundles — one capture copied four times, not four observations, checked by
hash rather than by counting files.
The claim came from checking the bundle named for Japanese, finding nothing,
and generalising to the corpus without querying the rest of it.
Replaying it emits 日本語日本語 under both sequencing extremes on all four
arms, matching its own recorded onData. So "repeated conversion is
undisturbed" is now captured rather than authored. It carries no
post-compositionend insertText, so it cannot speak to the swallow: the
a-after-conversion figure stays authored and unobserved.
Also rewords the paragraph opener. It claimed to broaden a Cmd framing, but
hazard 2 was never Cmd-framed — the lines above already say Cmd does not
reach it. The real gap was that hazard 2 named no trigger at all, which
reads as exotic when it is ordinary.
* build(xterm): land the patch regeneration harness
The five dependency patches under config/patches/ shipped with no tracked
way to regenerate any of them. The xterm one is the hard case: it is derived
from an upstream build, so no fix could be made without rebuilding, and the
tooling to rebuild lived only in one machine's scratch directory. That
blocked a measured fix for a live keystroke-loss bug, and the EditContext
reduction an OSS survey identified as the only real one available.
Adds the regenerator, the upstream pin, the hand-written source patch the
bundle hunks derive from, tests, docs, and a PR job that verifies the
shipped patches still match the pinned build. The job caches the shallow
clone keyed on the manifest, so a cold run is minutes and a warm one under
one. Round-trip verified: regenerating from a clean checkout reproduces the
shipped patch byte-for-byte.
Marks the emitted patch -diff -text. pnpm hashes it byte-for-byte, so a
CRLF checkout would break install on Windows, and its minified bundle lines
make a diff nobody can read — review the source patch instead.
Also rejects unknown flags. --check was the fallback for any unrecognised
argument, so a typo, or --help, silently triggered a full upstream build
instead of what the caller asked for.
* fix(xterm): stop swallowing a keystroke typed after an IME commit
Type a Japanese segment, convert it, then press a key one macrotask later
and that key was lost. No modifier, nothing exotic — every user who keeps
typing straight after converting. Korean surfaced it first only because
2-Set composes on nearly every keystroke.
handleCompositionInput discarded the payload unconditionally in the window
after the deferred send: _isSendingComposition stays true for one macrotask
after the timer cleared _pendingCompositionStart, and the branch substituted
'' for whatever arrived. The suppression is not itself wrong — it
de-duplicates IMEs that deliver their commit an event-loop turn late, which
terminal-stock-composition.test.ts pins. It just could not tell a duplicate
from new input, because the two events are identical apart from payload.
Now it compares against _sentComposition, the text the deferred send
actually emitted, and discards only a match. A flag-timing redesign was
measured first and rejected: it fixed this and duplicated on IBus, because
no timing change can separate events that differ only in content.
Edited in config/patches/xterm-src/ and regenerated through the harness, so
the emitted patch and the lockfile hash are derived, not hand-written.
The commit-overlap pin's swallow arm now asserts the repaired contract —
the value its own comment already named as correct and as what stock
beta.287 emits. #11504's arm at :184 flips too; it never covered that
report, as its own prior note recorded, and the reporter's +149ms arm is
untouched and still asserting the defect. Provenance hashes in three test
docblocks are updated, since regenerating changes the patch hash and with it
the resolved install directory.
* docs(terminal): re-measure the #6905 mutation citations against the new bundle
Regenerating the patch moved the resolved install, so this docblock's
patch_hash, line count, two line numbers and three mutation outcomes all
described a bundle that no longer exists. The deferred branch is one the
fix writes into, so the outcomes could not be re-pointed on reasoning.
Line numbers read off both files by diffing anchors rather than derived by
arithmetic: 201 to 205, 159 to 163. Outcomes re-run through the retained
rig, which re-resolves through the module loader and re-derives each arm
from a unique minified anchor: pristine 3 passed, m1 3 failed, m2 2 failed,
m3 3 passed — identical to the old bundle. Guard controls in both
directions exit 1, so the counts are falsifiable.
Comment-only; the assertions and expectations are unchanged.
* fix(xterm): size the preedit overlay to the cells its text will occupy
updateCompositionElements computed the overlay's left edge from the grid
but never its width, so the preedit rendered at the font's natural advance
while the committed text takes two cells per wide glyph. Measured in
Chromium 150: 가나다라 drew 48.45px as a preedit and 69.20px once committed
— the same characters, same font, 30% narrower, and drifting further with
each syllable. Every macOS mono font carrying Hangul measured 0.49–0.72 of
two cells; never 1.0.
Deriving the width from wcwidth and the cell measure moves Korean, Japanese
and Chinese to 1.000 and leaves ASCII at 1.000, which it already was:
한 12.125 -> 17.297 (17.30 expected)
가나다라 48.453 -> 69.188 (69.20)
안녕하세요 60.563 -> 86.500 (86.50)
日本語 42.000 -> 51.906 (51.90)
abcdefgh 69.234 -> 69.203 (69.20, unchanged)
Edited in config/patches/xterm-src/ and regenerated through the harness, so
the emitted patch and lockfile hash are derived rather than hand-written.
The unit test asserts the arithmetic, which is what CI can run. The pixel
consequence was measured on macOS with SF Mono in an Electron harness, not
on the Windows font stack STA-3232 reports from — so this demonstrates the
mechanism and does not stand as that row's platform evidence.
* test(e2e): pin the macOS Korean preedit as visible only while composing
#11914 reports the composing text invisible until Space. Its c3 was recorded
as unobtainable, and the reason on file was wrong: the boundary IS
assertable, but not in happy-dom, which reports display:block in BOTH the
active and inactive states and zeros for every rect. A test there passes
with the defect present.
Captured on real hardware instead: hidden and 0x0 before, .active with
display:block, a 15.84x16 rect and checkVisibility() true while composing
그, hidden again after. 39 DOM events, 2 composition starts, onData
["한","그","\r"].
Two mechanism findings are carried in the setup because both are invisible
in the result and fatal if removed. The input source must be selected AFTER
the app takes focus — focusing resets it to ABC. And the IME must be warmed
until an observed keyCode 229; typed cold it emits raw QWERTY (g k s r m)
with no composition at all, which is indistinguishable from an IME that is
not installed. Two runs were voided on exactly that signature before the
warm-up was found.
The has229 and compositionStarts assertions exist to make such a run fail
loudly rather than pass as a clean negative.
Gated on darwin plus ORCA_E2E_NATIVE_MACOS_KOREAN, like its siblings. The
final spec form has not itself been executed — the machine became
unavailable — so it carries the probe's measured values as literals rather
than a run of its own.
* docs(e2e): correct 19a8d133db7 — the Korean preedit spec has been executed
That commit said the landed form had never run and carried the probe's
values as literals. It has now run on real hardware: 1 passed, 9.1s, rc=0,
with the capture and log sealed under a verified hash manifest.
The teeth check was also run rather than reasoned about, and it changes
which assertion matters. Forcing the active overlay to max-width:0 with
overflow:hidden — invisible on screen — leaves the active class, the
textContent, display:block AND checkVisibility() all passing. Only
during.rect.width fails. checkVisibility() is not sufficient against this
defect; the bounding rect is the single load-bearing assertion, which the
docblock already said and this run confirms.
An earlier teeth attempt injected the CSS mid-run and tripped the
hasActiveClass poll instead, failing at the wrong assertion. It is
inconclusive and excluded from the seal rather than counted.
* test(terminal): add #12171's ordinary-English arm from a real Windows capture
c4 was recorded as unmet and the ledger sourced its control to
evidence/windows-current/, which holds 12 captures and not one English one.
The arm here comes from windows-9803-final instead — same probe, same host
geometry, same injector, en-US with no IME, replayed keydown for keydown.
Two limits are stated in the file rather than left for a reader to find. It
is a different bundle and a different run about 3.6 hours later, so it is
not a same-run arm. And it is #9803's range-active MUTANT arm: ordinary
English stays byte-exact even with that saved-range mutation live, which is
why it reads as a negative rather than as a baseline.
Bundle cited by directory with its file SHA-256; MANIFEST.sha256 verifies
21/21, rc=0. Nothing imported from .tmp.
* docs(terminal): correct #12164's grounds — the cited comments say no such thing
The rejection of #12164 from this file's family was recorded as resting on its
comment 1 (output doubling) and comment 2 (filed against 1.4.163). Checked
against the API: the issue has exactly two comments, neither of which says
that, and the string 1.4.163 appears nowhere in the thread.
The conclusion survives on better grounds. The issue BODY's repro is "Run any
CLI agent (Codex, AGY, Claude, etc.) that outputs Korean text into the Orca
terminal" — untyped output, no keystrokes, no composition — so excluding
CompositionHelper is right, and the input-path hunt was looking in the wrong
place. The body is also LLM-authored (it still contains a literal
"## 5. GitHub Submission Draft (Ready to Post)") and its Root Cause section
blames a CJK IME preedit buffer its own repro never engages, so it should not
be read as observation.
Comment-only; suite unchanged at 5/5.
* test(native-chat): make composition frames carry isComposing, not just inputType
This suite's comment claimed "Gating onChange on `isComposing` breaks here."
It did not. composeFrame() fired `input` with `inputType` but never set
`isComposing`, so a gate on `isComposing` passed all four tests untouched —
the suite asserted a discriminator it did not exercise.
Composition frames now carry both, so neither gate is exempt. Verified by
pointing the mutant at it: with an `isComposing` gate on the composer's
onChange, this suite now fails 3 of 4 (it passed 4 of 4 before), and the
ordinary-English arm correctly survives, since a composition gate should not
touch it. Production code is unchanged and stays gate-free; the mutation was
applied, measured, and reverted.
Found while excluding NativeChatView's question-card remount as the owner of
#12118 / STA-3219: the remount is real, but the preedit survives it precisely
because this write path has no composition gate.
* fix(mobile): ship the patched xterm build, matching desktop
mobile pinned @xterm/xterm 6.1.0-beta.285 while the patch is keyed to
6.1.0-beta.287, so mobile shipped stock xterm and neither IME defect fix
reached it: the swallowed keystroke after an IME commit (9506039de72) and
the preedit sized to the font rather than the grid (e04e0c88da5).
Bumps the three xterm packages to the desktop versions and adds the patch
to mobile's own pnpm.patchedDependencies. No copy of the patch: pnpm
accepts the parent-relative path and records it in the lockfile against
hash 8d63166272e9040a…, byte-identical to what desktop resolves, so the
two stay in step by construction rather than by a drift check.
The workspace separation is untouched — root pnpm-workspace.yaml still
declares `packages: []` and mobile keeps its own lockfile, which is what
keeps the root's patches from failing as ERR_PNPM_UNUSED_PATCH.
Verified in the generated webview bundle rather than at the install:
alignPreeditToGrid 0->2, sentComposition 0->3, pendingInput 0->11, and the
stock-only _handleAnyTextareaChanges 2->0 and dataAlreadySent 4->0. pnpm
applies patches during linking before postinstall regenerates the bundle,
confirmed by a revert/reinstall/re-apply cycle in both directions.
Mobile suite 2971 passed, 3 skipped — identical before and after. Bundle
+1,514 B (+0.24%). Lockfile churn is xterm-only; --frozen-lockfile passes.
mobile/src/ime/ime-submit-carry.ts is NOT made redundant and is untouched:
it handles iOS firing onSubmitEditing on a React Native native TextInput
after unmarking a composition, which is outside the WebView entirely.
Known divergence left alone: desktop also patches @xterm/addon-webgl and
mobile now runs that version unpatched. That patch is glyph/texture-atlas
rendering with nothing IME-related, so it affects neither fix.
* test(terminal): pin the preedit overlay against already-committed cells
STA-3132 (arm A), STA-3170 and STA-3232 report a Korean preedit painted on
top of text already on screen. Builds v1.4.163-v1.4.166 cancel the pending
finalizer in compositionstart, so a committed syllable reaches onData one
syllable late and buffer.x is stale — the overlay lands on the cell the
flushed syllable is about to occupy.
Replays a recorded hardware trace rather than an authored one: the ordered
DOM event stream captured on Windows + MS Korean (wave5-r2 evidence, 64
events), echoing onData back as PTY output.
The load-bearing assertion is deliberately not the obvious one. Comparing
overlay style.left against cursorX is tautological — left is computed from
buffer.x. This counts committed syllables from the compositionend events
the IME fired, so the two sides are independently derived.
Discriminated by a historical re-add across seven real bundles, since the
owner is deletion-shaped: pristine beta287, v1.4.155 and v1.4.162 pass;
v1.4.163 fails; v1.4.163 with that single call removed passes; the byte
identical baseline restored fails again; head passes. Every failing arm
fails only this case — the ordinary negative stays green in all seven.
The negative asserts its own category rather than claiming it: zero
composition events, zero isComposing, zero keyCode 229, exactly 16 events,
paired against the Korean arm's 4 starts / 3 ends / 11 updates / 64 events.
Scope: cell indices, not pixels. happy-dom has no layout, so the recorded
8x16 cell metrics are supplied to the render service. This makes no claim
about pixels visually overlapping; that is affected-OS confirmation and
stays open. Covers the overlap arm only — STA-3132's auto-line-break arm
and STA-3232's half-line-capacity and a11y arms are untouched.
* test(e2e): matrix macOS period substitution against the OS preference
#11504 reports macOS inserting ". " after a Hangul Space commit. This
sweeps six arms across both states of NSAutomaticPeriodSubstitutionEnabled,
reading the preference live per run rather than asserting a literal.
Two results worth having on record.
The reporter's stated trigger did not reproduce. Their words are "There is
no second press at all. One space is enough", but korean-single-space emits
zero insertText with the preference on or off. So does word-space-word-space.
Their timing does reproduce, with different content. korean-double-space and
korean-longer-word-double-space emit a delayed insertText at +122.5-122.7ms
after compositionend — squarely the reported +149ms — but the payload is a
space, never ". ". Consistent with the double-space rule seeing two slots
under ABC and only one under Korean, where the IME commit consumes the first.
The substitution itself is real and preference-bound: latin-double-space
gives "ab . " with the preference on and "ab " with it off, on one build
with the preference as the sole variable, reproduced across two runs.
That also refutes a claim in PR #11506, which states the substitution "is
enforced outside the renderer and never reproduces in dev builds, so changes
here must be verified against a packaged app". It reproduced in the dev build
twice and did not reproduce on the signed packaged app. That claim should not
be used as a verification gate.
Gated @headful behind ORCA_E2E_NATIVE_MACOS_PERIOD, same shape as the Korean
preedit spec, so it does not run in ordinary CI. Evidence is onData and DOM
only — the PTY-child reader aborted and no packaged-app arm was stable.
* test(terminal): actually enforce the recorded jamo progression
The preedit assertion compared sample.overlayText against sample.overlayText
— the same expression on both sides. A lane proved it by mutation: corrupting
seven of the eight recorded preedit values left the suite fully green. So the
docblock's ㄱ→가→간→나→낟→다→달→라, which the matrix also cites as this row's
recorded shape, was cited and unenforced.
The first attempt at a fix was insufficient and is worth recording. Threading
stroke.preedit through to the expectation still passed on a corrupted fixture,
because that value both drives the rig and was the expectation — corrupting it
moved both sides together. Same tautology, one level down.
The expectation is now an independent literal. Verified by mutation rather
than by reading: corrupting two recorded values fails one arm; restoring them
passes 3/3.
overlayCell was never affected — it is compared against a count derived from
the compositionend events, not from the buffer, and remains the load-bearing
assertion for the overlap.
* fix(e2e): select the selectable input source, not the first match
TISCreateInputSourceList can return several entries for one input source
id. A third-party IME publishes a non-selectable parent alongside the
selectable mode, and taking sources.first can return the parent — after
which TISSelectInputSource fails with paramErr (-50) while the caller
reports success from the enable step.
Found with Qingg (com.aodaren.inputmethod.Qingg), which exposes exactly
that pair under one id. Its mode id equals the bundle id, so filtering by
name would not have helped; selectability is the discriminator.
Now filters on kTISPropertyInputSourceIsSelectCapable and falls back to
the old behaviour when nothing advertises it, so single-entry sources are
unaffected. Also enables every entry for the id rather than only the one
being selected: selecting a mode whose parent is still disabled fails the
same way.
Compile-checked, and selecting com.apple.keylayout.ABC still exits 0.
Unrelated to the enable path: on macOS 26.5.2 third-party IMEs are gated
behind a consent sheet in System Settings. TISEnableInputSource returns
noErr immediately regardless, and the enable only lands if that sheet is
answered while the requesting process is still alive.
* test(terminal): discriminate #12171 against the real shortcut policy
The prior candidate mutation for this row was correctly refused: its suite
mocked shortcut policy so Process/229 became actionable, while the real
resolveTerminalShortcutAction has no Process branch — so the kill measured
the mock. This does not mock it.
Replays a capture of this row's own gesture (d, l, Shift+T, e, k, Space,
Enter under MS Korean, committing 있다) taken on Orca 1.4.164, through the
real useTerminalKeyboardShortcuts hook, capturing bytes at terminal.input.
The earlier capture could not discriminate at all because it recorded no
shiftKey; this one records it on 10 of 10 keydowns with code populated.
One physical Shift+T produces two shifted Process/229 keydowns. Under the
pre-#12265 classifier each synthesizes {key:'Enter', shiftKey:true}, which
the real policy resolves to sendInput '\x1b\r' — twice, giving 1b0d1b0d,
the two escapes the known-bad ed96881b0d1b0d contains.
Mutation is the retained pre-12265-process-shift.patch applied to HEAD, not
an authored one: patch -p1 applies clean and diffs identical to the mutant
copy. Arms are copies; shared source hashes the same before and after.
The English arm stays clean under both modules, so the mutation
discriminates by language rather than by harness — and a real Shift+Enter
through the same rig yields exactly ['\x1b\r'] in every arm, so a silent
pristine result means the code is quiet rather than the harness dead.
Scope: 1b0d1b0d is measured at the renderer boundary. The capture recorded
no PTY bytes — window.api.pty is frozen on shipped builds and the onData
channel needs a build-time flag — so this shows the renderer producing the
two escapes that payload contains, not a re-observation of the payload.
* docs(native-chat): narrow this file's disclaimer to what is now true
It said "THIS OWNS NO REPORTED ROW". Half of that is stale: the remount site
is now the attributed owner of #12118 and STA-3219. On real Windows TSF the
questionActive swap aborts a live composition — the old node gets only a
blur and no compositionend, the text returns as committed, and the next jamo
yields 아ㄴ rather than 안.
The other half holds. This file pins the opposite property, that the text
survives, which is the half those reporters already agree with. Mutation
shows the gap rather than asserting it: deleting the unmount entirely leaves
three of four tests green, because every substantive assertion is
after.value === … and a composer that never unmounts keeps its value.
Also records why the abort cannot be asserted here. The DOM exposes no
observable separating committed text from a live preedit — value is the same
string either way, there is no EditContext, and the only composing-ness
state is a per-instance ref discarded with the node. A test pinning "no
compositionend fires" would be an anti-guard: red the day it is fixed.
The cadence objection is kept, since it is now the open question rather than
the reason for exclusion.
* refactor(terminal): drop the unread isComposing field from XtermBypassEvent
Added by #6396 for terminal IME candidate handling that this branch has since
removed. No production or test code reads it, and the policy is safe without it:
during composition `key` is 'Process', so the non-ASCII printable checks that
would care never match.
Co-authored-by: Orca <help@stably.ai>
* fix(native-chat): keep the composer mounted through an in-flight IME composition
A question card replaced the composer outright
(`{questionActive ? null : <NativeChatComposer/>}`). Unmounting the field
mid-composition aborts the composition in the OS: the node is detached before
`compositionend` can fire, the preedit returns as committed text, and a resumed
Hangul syllable degrades — 아 then ㄴ yields `아ㄴ`, never `안`.
Confirmed in rasterised pixels on Windows with a real MS Korean IME, at both
v1.4.171 and the reporter-era v1.4.164 (the swap block is byte-identical
across them): the preedit underline present before the swap, the composer
visibly absent during it, and the same glyph back afterwards WITHOUT the
underline — committed, not composing.
The swap is now deferred while a composition is in flight, which is what
editors that survive IME do: ProseMirror gates DOM work on `view.composing`,
CodeMirror protects the composing subtree from redraws. Hiding instead of
unmounting does not work — `display:none` and `visibility:hidden` both blur the
focused element and abort the composition the same way.
The hold releases on `compositionend`, which browsers also fire on blur, so
clicking into the card's own answer input yields the input region immediately;
with nothing composing the card still replaces the composer at once, so no
stray "Send a message" appears beside a question.
The existing characterization test flips to a regression guard: it pinned the
node being destroyed, which was the defect. Node identity is the load-bearing
assertion — value-only checks are trivially satisfied by a composer that never
unmounts and cannot tell a held composition from a destroyed one.
The typing-redirect handler moves to its own hook. That is not cosmetic: both
touched files sat at the 400-line cap, and `max-lines` suppressions are
forbidden, so the room had to come from a real extraction.
* fix(macos): opt Orca out of AppKit automatic period substitution
macOS "Add period with double-space" (`NSAutomaticPeriodSubstitutionEnabled`,
on by default) is applied by AppKit's text input system. Native terminals never
join that system; Chromium text fields do, so xterm's helper textarea inherits
it and a double space arrives as `". "` — a period nobody typed, handed straight
to the PTY (#11504).
Chromium answers AppKit for quote and dash substitution and defaults both off,
but declares no period accessor at all, so AppKit applies that one without
asking. This user default is the only lever: there is no per-field or
per-webContents opt-out to prefer over it. Writing the key into Orca's own
defaults domain overrides the global value for this app alone and leaves the
user's system-wide setting untouched. It necessarily covers every Orca text
field, not only terminals — AppKit offers no narrower scope, and that tradeoff
is deliberate rather than accidental.
Measured on the reporter's own build v1.4.161: with the preference ON, typing
a,b,space,space yields `onData ["a","b"," ",". "]`; with it OFF the same arm
yields two spaces.
Note the issue's causal model is wrong and this fix does not follow it. It
claims the substitution only fires with a CJK input source and never with ABC.
The measurement is the inverse — every Korean arm is clean and the ABC arm is
the one that fires — so the fix is not conditioned on input source.
NOT YET VERIFIED ON HARDWARE. The unit tests inject the writer, so they prove
the call is made on darwin and skipped elsewhere; they do not prove AppKit
honours an app-domain override for this key. That check is outstanding.
* fix(xterm): keep a live composition across a lone Cmd press on macOS
CompositionHelper.keydown exempted keyCodes 16/17/18 from tearing a composition
down, which covers Shift/Ctrl/Alt but not macOS Meta — 91/93 in Chromium, 224 in
Firefox. A lone Cmd press mid-composition therefore reached
_finalizeComposition(false), which dropped the preedit overlay's `active` class
and committed the live syllable early. macOS keeps the marked text alive across
that press, so no later compositionstart re-arms the overlay and the rest of the
word composes invisibly.
Measured on hardware (m4air, macOS 26.5.2, Apple M4, 2-Set Korean) with the Cmd
posted as a CGEventType.flagsChanged, which is what a physical modifier emits.
AppleScript `key code 55` posts nothing a browser can see — a bare `key code 56`
for Shift is equally silent — which is why no capture in the corpus ever reached
this branch. Three arms, same build otherwise: overlay live throughout with the
exemption, dark and prematurely committed without it, live again with it
restored. Evidence under
.tmp/ime-handoff/swarm-scratch/wave31-cmd-preedit/evidence/.
The fix cannot widen past a lone modifier: only a standalone press reports these
keyCodes, and a Cmd chord during composition is reported by Chromium as 229,
which was already exempt. Cmd+A still ends the composition, via the IME's own
compositionend. Ghostty draws the same line, returning early from flagsChanged
under hasMarkedText() for every modifier including Super.
Orca's terminal pane was never affected — shouldSuppressTerminalModifierKeyboardEvent
drops a standalone Meta keydown before xterm sees it, and deleting only 'Meta'
from that set is what flipped the hardware arm to broken. The popout preview
terminal and mobile's webview install no such guard and did reach the teardown.
terminal-ime-xterm-composition-commit-overlap.test.ts asked its fixer to update
the two Cmd arms to the values it named as correct; both now emit a single ['한'].
* test(native-chat): drop two byte-identical duplicate cases
`4632b86919d` copy-pasted two cases twice into the same describe block:
`retains carry across a same-frame non-Enter keyup before redispatch` and
`expires carry before a deliberate Enter after the next frame`. Each pair is
byte-identical — same title, same body — so the copies asserted nothing the
originals did not.
This is what has been failing `static analysis` on this branch since 2026-08-06:
`oxlint vitest(no-identical-title)` reports both under `--deny-warnings`, and
`verify` fails solely because it requires static analysis to pass. Every other
gate in `verify` was already green, including typecheck, xterm patch sync, the
full test shard set, and both package jobs.
12 cases still pass in the file.
* test(e2e): skip the WebGL arm when no WebGL renderer exists
The #12164 probe runs two arms, webgl and dom, and closes by asserting the
active renderer is the requested one. That assertion is right for the dom arm —
it is what proves the pane actually left WebGL, without which the arm is
meaningless — but headless CI has no GPU, xterm falls back to DOM silently, and
the webgl arm then fails.
The failure reads as a Korean rendering defect and is not one, so the webgl arm
now skips with the active renderer named. The dom arm keeps the assertion
unchanged.
This is the third of three checks that have been red on this branch since
2026-08-06. `static analysis` and `verify` were fixed in c51c6b5837e; the CI log
shows this job as 1 failed / 1 passed, the pass being the dom arm.
* chore(lint): drop five unused no-console disable directives
`check-changed-code-quality` reports unused eslint-disable directives as errors,
and these five sat above diagnostic `console.log` calls in IME test and spec
files where `no-console` is not enabled — so each suppressed nothing.
This is the second of the two static-analysis steps. `c51c6b5837e` fixed
"Enforce focused code-quality plugins" (duplicate test titles); this fixes
"Enforce changed-code quality". Both had been red on this branch since
2026-08-06, and I mistook the first for the whole job.
The diagnostic logs themselves are kept — they are what a failing IME arm prints
for a reader to inspect.
* test(e2e): cover #12164 under fractional device scale factor
Fractional display scaling was #12164's last unexplored branch, and the reason
is worth recording: earlier attempts were BLOCKED, correctly, because they
proposed mutating the Windows display scale on a remote physical machine with no
console recovery. `--force-device-scale-factor` reaches the same renderer state
per process, so nothing outside the Electron instance changes and there is
nothing to restore.
The hypothesis was specific: `프프로로젝젝트트` is what a half-pixel cell boundary
could produce on a 2-column glyph, and nothing else in the suite varies dpr.
Measured at 1.25 and 1.5, both under WebGL: ink extents 25/21/16 with identical
ink groups, matching the scale-1 run. No doubling.
The arm self-certifies before asserting — if the flag does not take, the test
fails rather than silently measuring at dpr 1. That matters here: the sibling
spec's WebGL arm went two days reporting a missing GPU as a Korean rendering
defect precisely because a silent fallback looked like a result.
* fix(terminal): match Mod+letter shortcuts by physical key, not IME-rewritten key
With a CJK input source active, macOS and Windows report the physical key through
`code` but rewrite `key` to the layout's character: Korean 2-Set turns Cmd+C into
`{ key: "ㅊ", code: "KeyC", metaKey: true }`. Every `key.toLowerCase() === 'c'`
match misses it, so the shortcut is not recognised and xterm encodes the chord as
PTY input instead — issue #13033 reports `ESC[12618;9u` and a terminal that jumps
to the bottom, because user input scrolls the viewport.
This is the same key-vs-code confusion that owned #12171, where a `Shift+T`
typing ㅆ was read as Enter for want of a `code` guard, so the fix is the same
shape: trust `code` when it is present, fall back to `key` and then the legacy
`keyCode` when it is not (Chromium omits `code` on synthetic and some keypress
events, and `keyCode` keeps its US value even when `key` is rewritten).
Applied to the four terminal-side sites, including the dashboard pop-out, which
#13033 called out specifically as having its own key handler:
pty-connection.ts Cmd/Ctrl+C copy guard
keyboard-handlers.ts Cmd+G search navigation
agent-interrupt-inference.ts interrupt inference
preview-terminal-key-handler.ts pop-out paste
Nine further `key.toLowerCase()` letter matches exist outside the terminal
(TaskPage, editor, GitHub composer, browser markup). They have the same defect
and are deliberately left for a separate change rather than widening this one.
An existing case, `matchSearchNavigate > returns null for wrong key`, overrode
only `key` and left `code: 'KeyG'`, so it began passing for the wrong reason. It
now overrides both — which is what "wrong key" means once matching is physical —
and a companion case pins the Korean-rewritten chord still matching.
#13033 was closed NOT_PLANNED; the reporter's event shapes drive the new test.
* fix(renderer): match every Mod+letter shortcut by physical key, not IME-rewritten key
Completes the previous commit. A CJK input source rewrites `event.key` while
`event.code` keeps the physical key, so `key.toLowerCase() === 'z'` and friends
silently stop matching — the shortcut is not recognised and the keystroke falls
through to whatever handles unclaimed input.
The helper moves to `@/lib/ime-latin-shortcut-key` first: it now serves the
editor, GitHub composer and browser markup, and importing terminal-pane
internals into those would be the wrong direction. `lib/` already hosts
`ime-composition-keyboard-event` for the same reason.
Nine remaining sites, all previously unreachable under Korean/Japanese/Chinese/
Vietnamese input:
TaskPage, ActivityPrototypePage, ProjectViewWrapper Cmd+F search
useMarkupKeyboardShortcuts Cmd+Z undo
GitHubMarkdownComposer, RichMarkdownLinkBubble,
rich-markdown-link-shortcut Cmd+K link
native-chat-shortcut Cmd+J
rich-markdown-key-handler Cmd+Shift+X
Six of the nine test `!== 'letter'` as early-return guards and three test
`=== 'letter'`; the negation is applied per site, since a blind substitution
would have inverted six of them.
Full suite: 4249 files pass. Three files fail locally and none is caused by this
change — the branch touches no file under `src/main/` or `src/relay/`, all four
failures reproduce on an unmodified tree or pass in isolation (the worktree
poller passes 21/21 alone, so it is full-suite parallelism), and all 16 CI test
shards are green.
* docs(ime): scope the IME composition rules to the terminal-pane directory
#11893 proposed adding these to the root `AGENTS.md`, which every agent loads on
every task regardless of what it is doing. They only bind keyboard handling, the
composer and the terminal input path, so they belong next to that code —
`tests/e2e/AGENTS.md` already establishes the nested convention here.
Kept from #11893: range-derived commits, guarding above the key dispatch,
the `attachCustomKeyEventHandler` / `CompositionHelper` interaction, no
normalization at commit, and the recorded-trace evidence bar.
Added from defects found since it was written:
- match shortcuts on `event.code`, not `event.key` (#12171, #13033)
- `keyCode === 229` means an IME owns the press
- do not unmount a field mid-composition, and hiding is not a fix because
`display:none` blurs and aborts it too (#12118, STA-3219, #11332)
The evidence bar now also names the mutation check, since a test that survives
deleting the code it guards is guarding nothing — a failure this effort hit more
than once.
* fix(terminal): gate Ctrl+Enter CSI-u on a negotiated pane, porting #12462
Found while scoping the rebase onto `main`: #12462 landed on 2026-08-06 and
fixes a real defect this branch does not carry. Ctrl+Enter emitted
`\x1b[13;5u` unconditionally, so a pane that never negotiated the kitty
keyboard protocol — local Windows ConPTY, plain shell — printed the escape
verbatim into the prompt.
This branch deletes `terminal-ime-deferred-newline.ts`, which is one of the
files #12462 touched, so a rebase resolving those conflicts by taking our side
wholesale would silently reintroduce the defect. Porting it forward now means
the fix survives the rebase however the conflicts are resolved.
Mirrors the Shift+Enter guard already here: local ConPTY falls back to the
legacy CR every emulator sends, and a negotiated pane keeps the chord, so the
fallback is scoped to panes that cannot receive CSI-u rather than to Windows.
NARROWER THAN #12462 BY ONE CONDITION, deliberately. `main` also allows CSI-u
via `hasCtrlEnterCsiUAuthority()` (trusted consumer evidence, #12329); that
helper and its plumbing do not exist on this branch. Omitting it is the
conservative direction — an authorised pane gets `\r` instead of the chord,
rather than an unnegotiated pane printing an escape — but it should be restored
when the two histories are reconciled.
Test covers both directions and is mutation-checked: forcing the gate open
fails it, so it cannot pass by construction.
* fix: reconcile two more fixtures main moved while the stack waited
Both caught by CI, not locally, and the reason the local run missed one is
worth recording:
1. `browser-toolbar-profile-dialogs.ime-enter.test.tsx` did not pass
`useNativeUserAgent` / `onUseNativeUserAgentChange`, which `main` added to
`BrowserToolbarProfileDialogsProps`.
Local `pnpm typecheck` reported 0 errors on the same commit CI failed. The
cause was a stale `config/*.tsbuildinfo` — tsc reused an incremental cache
from before the merge. Deleting it reproduced CI's error exactly. Any
"typecheck clean" during this merge should be treated as unverified unless
the cache was cleared first.
2. Localization keys for `SshDisconnectedDialog` were absent from `en.json`:
the merge took this branch's component alongside `main`'s catalog.
Regenerated with `pnpm run sync:localization-catalog` rather than hand-added.
* fix(mobile): regenerate the lockfile the merge resolved by taking one side
CI's `verify` failed with `ERR_PNPM_OUTDATED_LOCKFILE` on `mermaid (lockfile:
11.16.0, manifest: 11.16.1)`. The mismatch was in `mobile/`, not the root — the
root lockfile was consistent throughout, which is why inspecting it (and even
GitHub's merge ref) found nothing wrong.
Cause: during the merge I resolved `mobile/pnpm-lock.yaml` by taking this
branch's side wholesale rather than merging it, so it kept `mermaid 11.16.0`
while `mobile/package.json` came from `main` at `11.16.1`. Taking one side of a
lockfile is only safe when the corresponding manifest comes from the same side.
Regenerated with `pnpm install --lockfile-only`; `--frozen-lockfile` now passes
in `mobile/`. Verified the xterm patch entry survives intact — same hash
`4f1b42d268f3964d…` and the parent-relative path into `config/patches/`, which
is the desktop/mobile coupling that would silently break the mobile build.
Two earlier diagnoses of this failure were wrong and are worth recording: it was
not the root lockfile, and it was not a stale merge ref (a rerun reproduced it
exactly).
---------
Co-authored-by: Orca <help@stably.ai>
Terminals froze app-wide several times daily, needing a manual pkill. libuv
unlinks the pathname a server bound to when it closes, with no ownership check,
so a departing daemon deleted whichever socket then sat at the canonical path —
including a live replacement's. The replacement kept hosting PTYs no client
could reach.
#12709 fixed that mechanism; this replaces the shape around it. Two invariants:
only a daemon publishing itself onto the canonical endpoint may mutate that
entry, and only by replacing one it has itself just proven dead; and no actor
removes a name it did not create.
Publish binds a private name, takes the canonical one with an exclusive link,
and on EEXIST proves the incumbent dead by connecting before replacing it in a
single rename. Only 'connected' means occupied and only refused/missing prove
death — a timeout proves nothing and declines. Deletes the claim sweeper, the
reclaim tail of killStaleDaemon, and three unfenced unlinkSync(socketPath)
calls in the launcher.
Measured: rename exposed no gap across 6,525 darwin / 8,004 linux probes of a
live handover, where unlink-then-link gapped on 200 of 200.
Verified on all three platforms: full suite on macOS and Linux, and daemon
restart e2e on a real windows-2022 host. Contract in src/main/daemon/AGENTS.md.
* perf(windows): stop the capability poll respawning blocking wsl.exe probes
#11295 added a 30s renderer interval to `useWindowsTerminalCapabilities` whose
early-return only fires when WSL is available with at least one distro, so on the
common Windows host (no WSL) it re-ran a full capability read forever. Each read
IPCs four probes whose main-process handlers were synchronous `execFileSync` calls
to wsl.exe/pwsh.exe, blocking the Electron main event loop for up to 5s a time.
The un-latching intent is kept: a host that answers "no WSL" is still re-checked,
now on an exponential backoff (30s, +60s, +120s) that parks once the answer stops
moving, re-arms on window focus, is shared by all consumers of an owner key, and
stops entirely when the last consumer unmounts. The wsl/pwsh IPC handlers now use
async twins that share the existing caches and back off identically.
* fix(windows): classify async wsl/pwsh probe failures with the execFile error shape
The async twins feed `execFile` callback errors into classifiers written for
`execFileSync`: a non-zero exit lands on `error.code` as a number rather than
`error.status`, and a timeout is a SIGTERM kill rather than ETIMEDOUT.
So a Windows host without WSL (wsl.exe ships in System32, so it exits non-zero
instead of ENOENT) was cached as retryable, shrinking the shared window from
10min to 45s and making the still-sync callers re-pay their blocking spawn ~13x
more often; and a pwsh cold start past 5s cached "pwsh missing" for 30s,
demoting the user's PowerShell 7 preference — the exact case the ETIMEDOUT
branch exists to prevent.
Also drops a literal NUL byte from the new re-probe module's signature
separator, which made the file binary to git, and seeds `lastProbeAt` at
registration so focus churn right after mount cannot defer the first re-probe
indefinitely.
Co-authored-by: Orca <help@stably.ai>
* perf(windows): route relay host-capability probes through the async wsl/pwsh twins
A paired web/mobile client resolves `useWindowsTerminalCapabilities` to a local
target (TabBar's `isWebClient` gate, and `useSettingsNavigationMetadata` forces
`{kind:'local'}`), so the new re-probe arms there too. But `window.api.wsl/pwsh`
on a web client is not the ipc/app.ts channel — it is `host.wsl.*`/`host.pwsh.*`
over the runtime RPC, which still ran the sync probes and blocked the desktop
main event loop on `execFileSync('wsl.exe' | 'pwsh.exe')` for up to 5s per call.
Switch those handlers and the relay preflight capability probe to the async
twins added here; they share the same caches, dedupe and backoff, so remote
callers see no behavior change.
* fix(windows): harden async capability reprobes
* fix(windows): dedupe PowerShell shell probes
---------
Co-authored-by: Orca <help@stably.ai>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
Folder-project workspace ids (`repoId::/path::workspace:<uuid>`) were compared via the suffix-stripping `splitWorktreeIdForFilesystem`, so every workspace sharing one directory compared equal at ~35 runtime call sites — PTYs leaked between siblings and paired/mobile clients hung on "Loading terminal". Both identity helpers now use the suffix-preserving `splitWorktreeId`, `stopTerminalsForWorktree` routes through the shared helper, and `findResolvedWorktreeIdForPath` gains a `targetWorktreeId` tie-break.
Co-authored-by: dgk-dev <dgk-dev@users.noreply.github.com>
* Persist the Linear issue list view and per-workspace filters
Layout, grouping, ordering, columns, and attribute filters survive a restart.
Facet ids are workspace-scoped, so filters are kept per Linear workspace and the
active filter is *derived* from the selected workspace rather than reset by an
effect on switch — no ordering race can apply workspace A's facets to B, and an
unresolved or cross-workspace selection reads as unfiltered without erasing
anything.
A single shared catalog backs the renderer state, `TaskResumeState`, and the
strict `ui.set` schema, so a new view option cannot leave paired web/mobile/relay
clients rejecting the whole payload. Persisted values are normalized as untrusted
input: a corrupt preference or a single bad workspace entry is dropped without
taking the rest of the resume state with it.
Deriving the filter also removed the guard that used to make three neighbouring
behaviours safe, so they are re-scoped here:
- The primary-team facet reset now fires only on an in-workspace team change.
A workspace switch also changes the primary team, and clearing there wiped the
filter that had just been restored for the workspace being switched *to*.
- The list-read force check no longer fires on the session's first read, so a
restored filter serves warm cache instead of forcing a network round trip
behind a blocking spinner on every cold start.
- The filter dropdown derives "no single workspace" from `workspaceId` alone.
With an unresolved workspace it previously rendered the statically populated
priority section, whose clicks now have nowhere to be stored.
* Harden Linear view persistence against the failures review surfaced
Five issues, each found by a reviewer and reproduced before fixing:
- The filter dropdown's prune effect only ran when the user opened the popover,
because the filter was always empty at startup. Restoration makes it run on
mount, where `availableTeams` may still be the issue-scraped fallback rather
than the real fetch. Metadata complete for a *partial* team set passes every
R12 guard, so it pruned facets belonging to teams it simply hadn't seen — and
the write persisted, deleting them permanently. Gated on `teamsSettled`.
- `canonicalize` dedupes but enforces none of the transport bounds; only the
throwing parser does. So `serialize` could emit a 101-label filter that the
strict `ui.set` schema rejects, which drops the WHOLE taskResumeState — github,
jira and linear query included — on every subsequent write, since the renderer
resends the merged object each time. Added `boundLinearIssueAttributeFilter`
and a round-trip test built from serializer output rather than a literal, which
is the only kind that can catch renderer/schema drift.
- `linearIssueView` now carries `.catch(undefined)`: value tolerance stops at the
top level, so any future instance of the above is a cosmetic reset of the view
instead of silent loss of every other resume field.
- A workspace switch forced an uncached list read in both directions. The switch
is a later observation, so the null-baseline fix didn't cover it; the cache is
already workspace-keyed, making the force pure cost.
- Recency for the 20-workspace cap came from object key order, which is wrong
twice: re-filtering an existing workspace left it at the head (first evicted,
though just used), and an array-index-like key enumerates first regardless of
insertion, so a write could evict the very entry it added. Recency is now an
explicit ordered key list.
Also adds the nested parity assertion — the top-level one compares only
TaskResumeState's own keys, so a field added to LinearIssueViewResumeState stayed
invisible to it, which is exactly what `.strict()` rejects.
The wiring test was blind: deleting the hydration guard outright left all four
assertions green. The gate is now `shouldPersistLinearIssueView`, unit-tested
directly, and the file is renamed to the repo's `*-boundary.test.ts` convention
with an assertion that fails on that mutation.
* Log discarded Linear views and fix empty-filter serialization
- Schema now logs when linearIssueView is discarded, making validation failures visible
- Fixed serialization: filters that become empty after bounding are now omitted
- Added AssertNoExtraKeys type check for bidirectional schema/type parity
- Refactored view option catalogs to use canonical constants, preventing UI/schema drift
* Remove workspace persistence limits and LRU eviction
Stop capping persisted Linear workspace filters at 20 and evicting
least-recently-used workspaces. Simplify persistence to store all
workspace filters, gate persistence only on resume state application,
and remove tests that pinned implementation details. Users can now
persist filters for all their workspaces without arbitrary limits.
* add test for linear persistence
* Improve Linear filter test clarity and fix e2e overlay dismissal for CI
- Convert parameterized filter-pruning test to sequential assertions
- Fix dismissOverlayChrome to toggle overlay triggers instead of
force-clicking inert page elements in headless CI
* Prevent TaskPage from stealing Escape from Radix menus
- Add check to detect open Radix dropdown menus and popovers; return
early from Escape handler to respect their capture-phase ownership
- Update overlay dismissal in e2e tests to use keyboard.press('Escape'),
now that TaskPage no longer interferes
* The capture-phase Escape guard in TaskPage bailed out for open dropdown menus and popovers, but an open Radix Select matches none of those selectors: the shared SelectContent wrapper (src/renderer/src/components/ui/select.tsx:60) renders data-slot="select-content" and Radix gives its content role="listbox", not role="menu". So with a select open, the window-level capture handler ran first, called preventDefault() and closeTaskPage() — closing the whole task page instead of just the select. Added [data-slot="select-content"] to the guard, as suggested. I did not add [role="listbox"]; the reviewer explicitly notes it's too broad, and the data-slot selector covers every select rendered through the shared wrapper.
---------
Co-authored-by: m4air <m4air@MacBook-Air.localdomain>
Co-authored-by: m4air <m4air@m4airs-Air.localdomain>
Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
* fix(rate-limits): read Kimi usage credentials from the configured WSL runtime
Kimi's usage fetch always read the Windows host's ~/.kimi-code, so a Kimi CLI running inside WSL rotated only the WSL-side token and the status bar was stuck on 'Run Kimi to refresh'.
Resolve the Kimi home from the local-account runtime target (mirroring Codex's getDefaultWslDistro()/getWslHome() UNC pattern), pinned to host off Windows, and keep KIMI_CODE_HOME host-only.
Fixes#12370
Co-authored-by: Orca <help@stably.ai>
* fix(rate-limits): bound and offload the Kimi credentials read for WSL homes
Adopted from @cengiz-io's #12372: read credentials through
createAuthFilesystemOperation (async, per-path dedup, AbortSignal bound) so a
stopped distro degrades to an error instead of parking Electron main on a UNC
read. ENOTDIR joins ENOENT as "not signed in" to keep existsSync parity, and the
WSL runtime target is now probed with the async wsl.exe helpers.
Co-authored-by: Orca <help@stably.ai>
* test(rate-limits): build Kimi credential-path expectations with path.join
The host-home assertions hardcoded POSIX separators, so they only passed on
a POSIX runner — on a Windows dev machine `join` emits backslashes and all
four cases failed (three assertion mismatches plus a WSL-suite fixture whose
map key never matched the path the fetcher read).
---------
Co-authored-by: Orca <help@stably.ai>
Co-authored-by: OrcaWin <alpha-eng@stably.ai>
* feat(native-chat): add model and effort pickers for grok
Grok had no session-option catalog, so the native chat composer showed no
pills and every launch ran the CLI's own defaults with no way to change them.
Adds a `GROK_SESSION_OPTION_CATALOG` (model via `-m`/`/model`, reasoning
effort via `--reasoning-effort`/`/effort`) and the discovery plumbing behind
it. Grok's selectable ids depend on the signed-in account and on `[model.*]`
config, so the seed carries only `grok-4.5` and a runtime `grok models` probe
supplies the rest as authoritative — a retired id must be droppable, since
launching one is a fatal exit rather than a warning.
Because `grok models` publishes `Default model:` and marks the row
`(default)`, the picker can name the model a fresh session is actually
running: `defaultModelIsCliDefault` plus an untracked record means no `-m`
was ever emitted, so the CLI is on its own default. That default scopes the
effort row but is never written to persisted settings — that field is what
authorizes `-m` on every later launch, and adopting a model the user never
picked would pin today's default forever, fatally so on an account without
it. `grok --help` publishes no default for `--reasoning-effort`, so the
effort value stays unnamed until something sets it.
Known gap: that refusal to persist is also a limit. An option set while on
the CLI default is dispatched and honored in-session, but reaches no later
launch — it persists under the default's id with `model` left unset, and
both `resolveNativeChatSessionOptionDefaults` and
`resolveAgentSessionOptionLaunch` bail without that key. Picking a model
explicitly persists normally. Closing this means teaching both to resolve
options from the default model while still refusing to emit `-m`, which is
the launch-args path and wants its own review.
Known gap: the picker infers "no `-m` was emitted" from its own in-memory
record, so a model reaching argv from outside it — the user's own
`agentDefaultArgs`, or a renderer reload that drops the record while the
flagged PTY lives on — leaves the pill claiming the CLI default while
another model runs. No wrong model is persisted.
Extracts `hasFlag` and `labelFromModelId`, and splits the model-probe spec
out of the commit-message registry so discovery no longer implies an agent
can write commit messages.
Co-authored-by: Orca <help@stably.ai>
* docs(native-chat): note the invariant keeping modelIsCliDefault agent-safe
The flag is computed without checking the catalog, so it reads as unsafe for
the four agents with no CLI default. It is safe only because `persist` bails
unless `modelId` is truthy, which for those agents implies a tracked model.
Widening that guard would silently change persistence for every agent.
Co-authored-by: Orca <help@stably.ai>
* fix: retire persisted models on mount and handle -- terminator
- When a pane mounts after model discovery has already settled, it now checks
the cache and retires persisted models that are no longer available.
- CLI flag detection now respects the `--` option terminator, treating
everything after it as positional arguments rather than flags.
* Fix: persist grok session options under probe-confirmed defaults
Options set under the CLI default were silently lost on restart.
Distinguish seed guesses from probe-confirmed defaults by renaming
`modelIsCliDefault` to `modelIsUnverifiedDefault`. Once confirmed,
adopt the default as a persisted flag so options survive restarts.
* fix(native-chat): close the retired-model fatal-launch paths from counsel review
Counsel report C1/C2 (High), C3, P1, C4:
- Untrack a session model an authoritative discovery dropped and gate every
persist path, so option writes can never re-adopt a retired id (C1).
- Resolve launch defaults through the enrichment cache: a persisted model
missing from every settled probe no longer becomes a fatal `-m` (C2).
- Serialize retirement and picks on one settings write queue that re-reads
live state at apply time (C3).
- Stabilize onSwitchToTerminal so the session-option surface is not rebuilt
every TerminalPane render (P1), and cap the enrichment host map (C4).
Co-authored-by: Orca <help@stably.ai>
* Store agent in enrichment entry and extract token utilities
Refactor enrichment to store the agent field directly instead of
parsing it from a composite key, and extract CLI flag token filtering
into a shared utility. Use a dedicated function for tracked model ID
lookup. Improves code reuse and reduces parsing overhead.
* Rename modelIsUnverifiedDefault to adoptModelAsLaunchDefault
Move the model adoption gate into the core session-options module, where probe confirmation and discovered-model status are known. This ensures adoption decisions are gate-checked before persisting to avoid fatal launch flags, and simplifies the picker surface by moving the logic to where it belongs.
* Keep model probe evidence by agent, not host
Store probed model IDs in agent-keyed cache independent of host cache, so
evidence persists across host eviction. Prevents retired models from being
treated as valid when host cache entries are evicted.
* Store agent in enrichment entries instead of separate proof-evidence map
Model probe evidence is now tied to enrichment entries rather than maintained in a separate per-agent map, eliminating the need for eviction logic that could disconnect proof from entries.
---------
Co-authored-by: Orca <help@stably.ai>
A bare `setlocal` inherits delayed expansion from the caller, so every `!`
in a percent-expanded value on the curl line is eaten as a delayed
reference -- mangling paneKey and dropping worktreeId. `!` is legal in a
Windows path.
Also adds the test file the script never had: structural checks that run on
every platform, plus real cmd.exe runs covering unset/empty/normal/trailing
backslash/envelope/`!`/inherited `/v:on`.
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* Add Recent Chats & Terminals to Cmd+J palette
On empty query, the palette leads with recently accessed chats and terminals,
addressable via ⌘1–⌘9 shortcuts. Digit chords are intercepted while the
palette is open, preventing workspace switches behind the overlay. Status
dots reflect agent blocking and activity.
* Fix Cmd+J recent order capture and backfill behavior
- Capture unfiltered recent tab order when palette opens, avoiding frozen filtered subsets on search reopen
- Backfill recent section when rows drop out (mid-open narrowing) instead of rendering empty
- Move create-workspace action to the end as a fallback, not competing with real matches
- Add test coverage for order freezing, backfill, and create-action ranking
* Fix Cmd+J selection backfill when recent tabs hydrate late
Track auto-selected item to distinguish user-moved selections from auto-picks. When recent tabs arrive after the palette opens, reset selection to the top item unless the user explicitly moved it. Use useLayoutEffect to capture the pre-hydration worktree order before tabs render, preventing flashing. Fix worktree budget cap for sessions with no open tabs. Add test coverage for late hydration and user-moved selection preservation.
* Rank Cmd+J results by match relevance, lead with stronger section
Adds match-relevance scoring to rank search results by which field matched and
match position (prefix > word-start > mid-word). When a typed query has results
in both worktrees and open tabs, whichever section holds the stronger hit now
leads the list, matching user intent — a prefix hit in a tab beats a mid-name
hit in a worktree.
Decouples live status dots from the palette body subscriptions to prevent the
whole list from re-rendering on every agent transition or pane-title change.
Dots now own their subscriptions via PaletteLiveStatusProvider, while the body
reads status maps as a snapshot, refreshed only when the palette opens or tabs
change. Freezes the snapshot identity during animation to keep the selection
stable even while closing.
* Handle decomposed accents in Cmd+J match relevance scoring
Include Unicode combining marks (\p{M}) in word-boundary detection so
decomposed characters like café are treated as word boundaries instead
of mid-word matches.
* feat(ai-vault): validate session-delete targets for single-file providers
Add the pure judgement layer for deleting an Agent Session History entry.
`validateAiVaultSessionDeleteTarget` decides whether a session may be removed:
the agent must be one of the nine providers where a single file is the whole
session (gemini, copilot, cursor, hermes, devin, openclaw, droid, pi, omp),
the host must be local, and the renderer-supplied path must resolve inside
that agent's own session roots and match its discovery predicate.
To keep the delete roots from drifting from the scanner's own roots, the
WSL-expansion helper moves to session-scanner-root-dirs.ts and the OpenClaw
root derivation + session predicate become shared helpers that
discoverOpenClawFiles itself consumes.
The result is path-only and never touches the filesystem; a returned
`allowed: true` still requires an lstat/realpath re-check in the executor
(S-2) before removal, documented as a caller contract on the result type.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* feat(ai-vault): move a validated session transcript to the trash
Add the filesystem executor behind session deletion. It calls the S-1 path
validator, then performs the fs-side guards that validator documented it
could not: lstat().isFile() rejects a directory or symlink, and realpath is
re-fed through the validator so a regular file reached through a symlinked
parent that escapes the agent's roots is rejected too. Only then is the file
moved to the OS trash via shell.trashItem, with ENOENT treated as success so
a delete racing an external removal stays idempotent.
WSL UNC paths (no Recycle Bin) are delegated to tryDeleteWslUncPath before the
Windows-local fs guards, mirroring fs:deletePath. Any non-ENOENT error is
returned as a failure result rather than thrown, since IPC payloads are
untyped at runtime.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* feat(ai-vault): delete-session IPC handler, preload bridge, cache invalidation
Wire the S-2 delete executor to an IPC endpoint and expose it on the preload
bridge. The renderer calls aiVault:deleteSession with { agent, filePath,
executionHostId }; the handler fetches WSL homes, delegates to the executor
(which re-validates and trashes), and on a real delete invalidates the caches
that could otherwise keep serving the deleted session.
Cache invalidation is generation-guarded: a scan already in flight when the
delete lands carries an older generation and must not write its pre-delete
result back into the cache. Without this, an in-flight scan resolving just
after the delete would resurrect the deleted session for the 15s TTL — and
force-refreshing the panel only masks it for the desktop, not for the paired
mobile client or runtime RPC that share the same cache module. Both the shared
local-scope cache and the desktop multi-host cache carry the guard, with
regression tests for the in-flight race.
The delete result type moves to shared/ai-vault-types.ts so the renderer can
import the same contract the executor returns. To keep ai-vault.ts within the
max-lines budget after adding the delete wiring, two cohesive pieces are
extracted to their own files: the delete orchestration (ai-vault-delete.ts)
and listAiVaultSubagentSessions (ai-vault-subagent-list.ts). The latter is the
only handler with no dependency on this module's private cache state, so it is
the one piece that moves verbatim without threading state through a seam.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* feat(ai-vault): renderer judgement for whether Delete is offered
Add the renderer counterpart to the main-side delete validator: given a
session, decide whether the row menu shows Delete enabled, or disabled with a
reason a tooltip can render. It reuses the shared deletable-agent set and
unsupported-reason map so the two sides can never disagree about which agents
are deletable, and reuses the existing local-host / synthetic-path renderer
helpers.
This is intentionally not a security boundary — it validates neither the path
root nor the file predicate. Those are the main process's untrusted-input
defense; the renderer only picks the affordance, and the main side re-checks
on delete regardless.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* docs(ai-vault): correct deletability parity claim; test multi-reason agent
The renderer deletability check runs host -> synthetic -> agent, while the
main validator runs agent -> host -> synthetic. The two layers agree only on
deletable-or-not (renderer-false is a subset of main-false), not on the reason
code a doubly-failing session carries. Document that explicitly instead of
implying the orders match, and add the antigravity case (two reason codes) so
the agentReasonCodes array shape is actually exercised.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* feat(ai-vault): add Delete to the session row menu with a confirmation dialog
Wire the delete affordance into AI Vault. Both the dropdown and the context
menu gain a destructive Delete item; a session that can't be completely deleted
(remote host, synthetic OpenCode-SQLite path, or a directory/registry-backed
agent) shows the item disabled with a reason surfaced both as a tooltip and as
an aria-label so keyboard and screen-reader users learn why. Confirming opens a
dialog that names the session and states it will no longer be resumable from
the provider's own CLI, then calls the delete IPC and force-refreshes the list
for immediate feedback (the main side has already invalidated its caches).
The confirmation copy says the session "will be deleted" rather than "moved to
the trash": on Windows a WSL session is deleted with rm inside the distro (no
Recycle Bin), so promising recoverability would be a lie on that platform.
Deletability is computed once per row and shared by both menus so they can
never disagree. New pure logic — the reason-to-tooltip mapping (including the
multi-reason join) and the delete action hook's deleted/rejected/failed
branches — is covered by unit tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XDLggjSAjDnaWi3Y8U622i
* fix(ai-vault): state that Delete is unavailable without naming the cause
The disabled Delete item explained a provider's storage layout to the user
("Claude sessions can't be deleted here: stores sessions as a folder, not a
single file"). That is Orca's problem, not the reader's — the tooltip now says
which sessions are affected and stops there. The non-local-host string stays as
it was: it states scope, not a cause, and tells the user what would work.
The reason-code plumbing existed only to compose that tooltip, so
AI_VAULT_UNSUPPORTED_DELETE_REASONS, AiVaultUnsupportedDeleteReasonCode, and the
renderer result's agentReasonCodes field go with it. Why each agent is excluded
moves into the comment above AI_VAULT_DELETABLE_AGENTS, where a reader looking
up the deletable set will find it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KGuzChimmQ1dYX2raecrH7
* feat(ai-vault): delete claude, rovo, and grok sessions by their directory
These three were excluded only because the delete unit was one file. Their
sessions are directories — claude keeps Task subagent transcripts in a sibling
`<uuid>/subagents/`, rovo and grok keep everything under `<sessionId>/` — and
nothing in them is shared with another session, so a directory-aware delete is
still a complete delete. Supported goes from 9 agents to 12; the four that
remain (antigravity, kimi, codex, opencode) are blocked by a registry or a
SQLite row, which no delete unit fixes.
Validation now returns an ordered removal plan instead of a single path. Each
removal carries the kind it must be on disk and the roots its realpath must
stay inside, so the executor's guard is the same shape for a file and for a
directory. Companions come first and the transcript last: the transcript is
what puts the row on screen, so a part-way failure leaves the row to retry
from rather than dropping it and stranding the rest on disk.
Claude's `session-env/<uuid>/` goes with the transcript — it holds that
session's generated shell exports and nothing else. Its sibling
`file-history/<uuid>/` deliberately does not: it is the rewind buffer holding
earlier versions of the user's own files, and retiring a session is no reason
to take away the only copy that can restore them.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KGuzChimmQ1dYX2raecrH7
* fix(ai-vault): remove a claude session's own directory, not just its subagents
Deleting a claude session trashed `<uuid>/subagents/` and left `<uuid>/` behind
as an empty directory — one per deleted session, accumulating under every
project. The directory is named after the transcript, so it belongs to that
session as a whole; take it rather than the one subdirectory inside it. Still
derived from the scanner's own subagents path, so the two cannot drift.
Reaching the parent means a degenerate stem now matters: `..jsonl` passes the
extension check and its stem is `.`, which would resolve the session directory
to the project directory holding every session. Reject it instead.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KGuzChimmQ1dYX2raecrH7
* fix(ai-vault): keep a session row collapsed when a menu action is chosen
Radix portals the row's dropdown and context menus out of its DOM, but React
still bubbles their clicks back through the component tree, so every menu
selection also hit the row's own click handler and expanded it. The trigger
button already stopped propagation, which is why opening the menu looked fine
and only choosing an item misbehaved.
It shows worst on Delete: the row expands behind the confirm dialog, so
cancelling leaves the list rearranged under a dialog the user just backed out
of. Toggle details only for clicks that land in the row's own subtree — that
covers the context menu and any future portalled surface, not just this one.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KGuzChimmQ1dYX2raecrH7
* fix(ai-vault): harden the delete-confirmation flow against IPC rejection and mid-delete dismissal
Two robustness gaps flagged in review:
- handleConfirmDelete only branched on result.outcome. The main handler
resolves with a 'failed'/'rejected' outcome rather than throwing, but the
IPC invoke itself can still reject on a transport/serialization error, and
the caller fires it with `void`. That reject would surface as an unhandled
rejection with no toast. Catch it and show the same generic failure toast.
- handleDialogOpenChange cleared sessionPendingDelete on every open=false.
The Cancel button is disabled mid-delete, but Radix still fires its
Escape/outside-click/X close, which could dismiss an in-flight delete out
from under itself. Ignore close requests while deletingSession is true.
Both covered by regression tests (verified failing without the fix).
* fix(ai-vault): route WSL UNC directory removals through the WSL rm branch
Directory-shaped deletes (claude's subagents/session-env dirs, rovo/grok's
session dir) gated the WSL branch on kind === 'file', so on Windows a session
under a WSL distro home fell through to shell.trashItem — which can't trash a
WSL-volume item (no Recycle Bin) and throws, or worse is silently stranded when
the 9P filesystem's unreliable lstat false-reports ENOENT and the executor
treats that as success. Single-file deletes predate the directory kinds, so the
file-only gate was correct until directory removals were added.
tryDeleteWslUncPath already supports recursive removal; pass recursive for
directory removals so they take the same WSL rm path as files instead of
shell.trashItem. Covered by two regression tests (file: non-recursive,
directory: recursive), verified failing without the fix.
Also drops the internal ledger-ID references (D-*, S-*) from comments in these
two files; they pointed at a private design doc a reader can't see.
* docs(ai-vault): drop internal design-ledger IDs from shipped comments
Comments across the session-delete feature cited decision/slice IDs (D-1..D-7,
S-1..S-5) from a private design document. Those references are meaningless to
anyone reading the code without that doc, so remove the IDs while keeping the
reasoning each comment carried. No behavior change.
* test(ai-vault): e2e-cover the real on-disk session delete
The unit tests mock lstat/realpath/trashItem, so nothing proved the whole IPC
path actually removes files. This spec seeds sessions into the E2E harness's
isolated HOME and deletes them through window.api.aiVault.deleteSession:
- a single-file session (gemini): the transcript is gone from disk and drops
out of the list.
- a directory-shaped session (claude): the transcript, the <uuid>/ session
directory (subagents included, no empty shell left), and the session-env
companion are all gone, while the file-history rewind buffer is preserved.
Verified failing when the executor's removal is stubbed out. Runs on Linux CI.
* fix(ai-vault): address review findings on the session-delete flow
Three points raised in review:
- Disable Delete for a still-running session. resolveAiVaultSessionDeletability
now gates on liveState (working/blocked/waiting) last — an otherwise-deletable
session that is mid-run shows "wait for it to finish" instead of an enabled
Delete, so trashing a live agent's transcript can't drop writes it is still
appending. Unsupported/remote sessions keep their permanent reason.
- Realpath the roots, not just the target, in the executor's escape check. The
roots were only resolve()'d (text), so a session under a symlinked root
(~/.claude -> /Volumes/…) was falsely rejected; realpath each root (falling
back to its text form when it can't be resolved) before the membership check.
- Invalidate the parse cache with the raw filePath, not resolve(filePath). The
cache is keyed by the exact path the scanner discovered, so resolve() could
normalise it away from the stored key and miss. Drops the now-unused import.
Also moves AiVaultDeleteSessionArgs/Result out of ai-vault-types.ts (which the
upstream merge pushed over the max-lines limit) into the ai-vault-session-deletion
domain module they belong to, and updates importers.
Regression tests added for the live gate, the symlinked-root accept, and the
reason string; verified failing without each fix.
* fix(ai-vault): type the deleteSession preload bridge as its real result
The bridge declared Promise<unknown> while AiVaultApi.deleteSession promises
AiVaultDeleteSessionResult, so the preload object leaned on the api-types
declaration to stay honest instead of being checked against it.
Co-authored-by: Orca <help@stably.ai>
* refactor(ai-vault): tighten the session-delete code to house style
Comments across the delete flow explained HOW alongside WHY and ran to a dozen
lines; they now carry only the non-obvious reasoning. The excluded-agent
rationale, the caller contract on the validator, and the file-history carve-out
are kept — those are knowledge, not narration.
Also removes three duplications the feature introduced:
- AiVaultSessionDeleteExecutionResult was an alias for AiVaultDeleteSessionResult
whose comment pointed at a module the type no longer lives in.
- The synthetic-path predicate existed twice under near-identical names; the
renderer now re-exports the shared one it already had a sibling import of.
- The delete-failure toast was written out verbatim in both the rejected and
the thrown branch.
Co-authored-by: Orca <help@stably.ai>
* refactor(ai-vault): use a design-system dialog width and a stable row selector
The confirm dialog pinned an arbitrary sm:max-w-[440px]; every other dialog in
the right sidebar uses a scale token, and md (448px) covers the role.
The row-expand test selected the row by [draggable="true"], which stopped
naming the row when draggable moved to the title element upstream. It still
passed by bubbling, so the comment was the only thing wrong — now it selects
the title deliberately and says why the query is first-match (Radix's asChild
trigger repeats the subtree, so screen.get* sees duplicates).
Also types the e2e delete helper as AiVaultDeleteSessionResult instead of a
hand-written { outcome: string }, now that the preload bridge returns it.
Co-authored-by: Orca <help@stably.ai>
* refactor(ai-vault): consolidate agent sources and use system dialog
Discovery and deletion now share the same agent source definitions, eliminating the risk of them drifting apart. A single `AI_VAULT_AGENT_SOURCES` table declares each agent's root directories, file extensions, and acceptance predicates. Replaced the custom delete confirmation dialog with the system dialog, simplifying the delete action hook and removing boilerplate state management.
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: Orca <help@stably.ai>
Google binds a signed-in session to the browser identity that created it, and
rejected Orca's Electron/Chrome-shaped UA on accounts.google.com — so sessions
died at ~1h and re-copying cookies only reset the same losing clock.
Present a Firefox identity scoped to the Google/YouTube sign-in hosts at every
layer Google probes: the request headers (with sec-ch-ua* stripped, since real
Firefox sends none), the per-WebContents UA that auth JS reads, and the CDP
emulation override a viewport preset installs. All three must agree — a
disagreement between them is a sharper bot signal than any single wrong UA.
Native-UA profiles opt out at every layer, including popup child windows and
after a browser-native cookie import.
Validated live against a real account: freshness cookies issued and rotating on
a ~10-min cadence, session still signed in and self-renewing past 10 hours,
against a pre-fix baseline of death at ~1h.
Closes#11518.
* feat(native-chat): render omp transcripts
omp already ships as a first-class launchable agent with session_id resume, but
its transcripts had no decoder, so native chat could not render it — the agent
runs and the conversation stays a raw terminal. This adds the decoder and wires
it through the same path Claude, Codex and Grok use.
omp writes one envelope per line, `{ type, id, parentId, timestamp, … }`, where
conversation turns are `type: 'message'` and the rest is session bookkeeping.
Reasoning arrives as a `thinking` content block inside the assistant turn, so
the mapping follows Claude rather than Codex: thinking becomes a text block on
an assistant message, where Codex and Grok emit a separate reasoning role only
because their transcripts carry dedicated reasoning records.
- toolCall -> tool-call, arguments passed through as the object omp writes
- toolResult -> tool role, isError preserved
- developer -> system, matching the Codex non-user/non-assistant fallback
- blob-handle images drop, as the Claude mapper drops an image record with
neither path nor url
- bookkeeping and unrecognized types skip rather than throw
Session files are `<ISO timestamp>_<session id>.jsonl` under a per-cwd directory,
so the resolver matches the id as a base-name suffix the way Codex rollout files
are matched, and honors OMP_CODING_AGENT_DIR through normalizeAgentSessionsDir
so it stays consistent with the AI Vault scanner.
omp records no interruption or abort event, so unlike Claude and Codex there is
no NATIVE_CHAT_INTERRUPTED_STATUS_TEXT path.
Verified against 94,603 lines of real omp transcripts across four sessions:
50,546 records decoded, zero malformed, zero thrown.
* fix(native-chat): complete omp record coverage and gate remote transcripts
Review fixes on the omp transcript decoder.
omp writes several record types with no `content` field, so they decoded
to zero blocks and disappeared from the chat view entirely:
- `bashExecution` / `pythonExecution`: TUI `!command` runs, now a tool turn
- `fileMention`: `@path` attachments, listed by path (never `files[].content`,
which is an auto-read dump)
- `custom_message` and legacy `custom` / `hookMessage` rows, gated on
`display` the way omp's own renderer gates them
Also:
- `stopReason: 'aborted'` turns now surface as the interrupted row, matching
the Claude and Codex decoders. An abort carrying partial content keeps it.
- A cancelled command cell now reads as errored. Every omp cancel path emits
`exitCode: undefined`, which JSON drops, so an `exitCode !== 0` check read a
cancelled run as a clean success.
- omp joins Grok in requiring a locally readable transcript. Its hook reports
no transcript path, so under Model-A SSH the chat view opened against a disk
this process cannot read and never loaded. Applies on mobile too, which
shares the same allowlist.
- The session-file walk prunes omp's per-session subagent artifact
directories, matching the AI Vault scanner. It was returning a subagent
transcript instead of the parent session, and cost a full recursive readdir
on every resolve.
* style(native-chat): apply oxfmt to the omp review fixes
Mobile CI gates `oxfmt --check`; the two root files were unformatted too,
just ungated there. Line wrapping only, no behavior change.
---------
Co-authored-by: plotarmordev <299844489+plotarmordev@users.noreply.github.com>
Co-authored-by: Brennan Benson <79079362+brennanb2025@users.noreply.github.com>
* fix(serve): recognize CLI-form serve args on the Electron process
When the binary is launched as `… serve --port …` without the CLI rewrite
that injects `--serve`, normalize argv so isServeMode, headless GPU flags,
and serve option parsing all engage.
Preserves existing `--serve*` flag behavior for the CLI-spawned path.
Fixes#12677
* fix(serve): treat only CLI subcommand position as serve
Parse bare `serve` as the first positional token after flags/values so an
option value named `serve` cannot enable headless mode.
Addresses CodeRabbit on #12818.
* fix(serve): keep CLI redirects ahead of the serve argv rewrite
Rewriting argv before maybeRedirectAppImageCliLaunch replaced the `serve`
positional with `--serve`, so the redirect's command-name lookup saw a port
number and bailed — dropping AppImage serve launches out of the CLI path.
Also translate `--port=6768` (the CLI accepts it, getServeOptions only reads
the next token) and the mixed `--serve --port` form, so a security-shaped flag
like `--no-pairing` can no longer read as accepted while pairing stays on.
Map lookups replace `in` on object literals, which turned a stray `serve
toString` positional into a function spliced onto argv.
* fix(serve): close the CLI-form serve gaps found in review
second-instance: shouldActivateDesktopForSecondInstance matched only `--serve`,
so a duplicate `<binary> serve --port …` — the ExecStart shape documented in
docs/reference/headless-linux-server.md — promoted the live headless server to a
desktop window, un-fixing #11935 on exactly the launch shape this PR legitimizes.
findServeSubcommandIndex consumed a flag's value unconditionally while the
rewrite consumed it only when the next token was not flag-shaped. The two could
disagree and swallow the `serve` token, leaving `--serve` uninjected: #12677
again in a new shape (`--port --port serve`, `--port -- serve`). Both scans now
share one definition of value consumption.
`<binary> serve --help` / `serve help` bound a network-exposed runtime server
with pairing on and printed nothing; the AppImage redirect already routes those
three tokens to the CLI, so refuse them here too.
`--no-pairing=false` translated to `--serve-no-pairing` with the value dropped,
disabling pairing for an operator who asked for the opposite. The CLI reads its
serve booleans as `flags.get(name) === true`, so a boolean is now translated only
in its bare form and the `=` form rides through as the CLI treats it.
Tests: spec-derived parity between src/cli/specs/serve.ts and the rewrite,
covering both ends of the contract (serveOrcaApp and getServeOptions); a
source-text lock on the index.ts redirect/rewrite ordering, which reverted
silently green before; an exhaustive self-consistency property test; and the
real GUI launch argv shapes that must never enter serve mode.
---------
Co-authored-by: Jinwoo-H <Jinwoo-H@users.noreply.github.com>
cleanupExpiredRemoteClipboardFiles read the entire OS temp root with
readdir({ withFileTypes: true }) and mapped every entry into Promise.all,
so the prefix filter only ran after a promise already existed per entry.
The sweep is fire-and-forget from registerClipboardIpcHandlers at startup,
so a large %TEMP% froze the main process before the window came up.
Stream the root with opendir, skip foreign entries before allocating, and
cap in-flight removals at 8.
Wake idle Run coordinators with durable orchestration mail pointers while keeping message payloads in the store until check consumes them. Preserve waiter, Cursor, restart, real Codex title, and PTY replacement behavior.\n\nPart of #12953.
* fix(terminal): gate Ctrl+Enter CSI-u on a negotiated kitty pane
Ctrl+Enter emitted \x1b[13;5u unconditionally, so a pane that never
negotiated the kitty keyboard protocol (local Windows ConPTY, plain
shell) printed the escape verbatim into the prompt. Mirror the
Shift+Enter guard and fall back to the legacy CR every emulator sends
for this chord. Keeps the intercept, so IME commit ordering and the
single-send dedupe still apply.
Fixes#12329
Co-authored-by: Orca <help@stably.ai>
* test(e2e): negotiate kitty via PTY output in the Ctrl+Enter spec
The Ctrl+Enter gate reads the PTY-output kitty tracker, which
enableKittyKeyboardReporting never feeds (it writes straight into
xterm's parser), so the spec pressed the chord on a pane the policy
still saw as un-negotiated and got the CR fallback. Negotiate from the
application side like the neighbouring Shift+Enter spec, and reset the
flags afterwards for the serial suite.
Co-authored-by: Orca <help@stably.ai>
* fix(terminal): preserve trusted Ctrl+Enter routing
* fix(terminal): scope IME redispatch ownership
* fix(terminal): reject conflicting Ctrl+Enter evidence
---------
Co-authored-by: Orca <help@stably.ai>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* fix(mobile): never auto-advertise virtual bridge addresses for pairing
Container/VM bridges stay manually pickable, but automatic defaults skip them so
QR codes do not race an unreachable direct path. Relay pairs without a local
address; LAN-only and runtime pairing fail closed on bridge-only hosts.
* fix(mobile): never auto-advertise virtual bridge addresses for pairing
- Set endpoint to null when no direct address is advertised, so the QR
doesn't show an unreachable address to the scanning phone
- Distinguish "No address selected" (bridge exists but not advertised)
from "No interfaces found" (genuinely nothing to pick)
- Add tests for NetworkInterfacePicker placeholder behavior
* fix(github-project): index fork upstream slugs for project row matching
Project cards often reference the public upstream repo while the open
clone's origin is a personal fork. Map the parent slug to the same Repo
so selected-repo filters no longer hide every board row.
Preserves origin-based getRepoSlug identity for non-project callers.
Fixes#12647
* fix(github-project): match project rows against fork upstream slugs
Resolve the referenced call to a nonexistent `resolveRepoUpstreamSlug` and
match the persisted `repo.upstream` parent instead of issuing an extra
`github.repoUpstream` RPC per repo on every index build — that lookup shells
out to `gh repo view` for non-forks, so it would have gated the Projects tab
on N network calls. `repo.upstream` is already resolved at repo-add time and
backfilled at startup, so the fix costs no IPC.
Origin matches take precedence over upstream ones so an open clone of the
upstream repo itself is never made ambiguous by someone's fork of it.
Also covers the two surfaces the origin-only match broke alongside the desktop
table: mobile's project row matcher and the store-slice row-mutation routing.
* fix(github-project): scope fork upstream matching by host and selection
Round-1 review fixes on top of the upstream-slug index:
- Apply origin-over-upstream precedence among *selected* repos instead of
globally. An open-but-unselected clone of the upstream repo was shadowing the
selected fork, so #12647 still reproduced for anyone holding both — and repo
selection collapses to one repo per project key, which is exactly that case.
- Scope a fork's upstream identity key to the fork's own origin host.
Persistence strips upstream.host, so GHES forks never matched their own rows
and a GHES fork's parent could bind a same-named github.com row.
* fix(github-project): skip the fork alias when its own origin is unresolved
Round-2 review fix. `githubHostFromIdentityKey` cannot tell "origin resolved to
github.com" from "origin did not resolve" — both yield no host. A GHES fork
whose slug resolution had failed (auth lapse, unreachable runtime) therefore
landed in the github.com namespace, so an unrelated public Project row matched
it and Start work opened the wrong clone on the wrong server.
Require a resolved origin before indexing the upstream alias: it is the only
host evidence there is, and a repo with an unresolved origin was already absent
from the origin index, so nothing is lost that origin matching had.
* fix(repos): persist the fork upstream host instead of dropping it
`sanitizeRepoUpstream` kept only `{owner, repo}`, so a fork's parent lost the
server it lives on every time the record round-tripped through disk.
That forced the Project row matcher to re-infer the host from `origin`. The
inference is right for an API-resolved fork parent — `getRepoUpstream` stamps
`origin.host` there precisely because "a fork parent lives on the same server as
the fork". It is wrong for the other branch: a local `upstream` remote carries
its own host, so a github.com clone with a GHES `upstream` remote was indexed
into the github.com namespace, where an unrelated same-owner/name public repo
could claim it and Start work would open the wrong clone.
Keeping the host removes the guess. Absent stays absent, so records written
before this hydrate unchanged and the origin-derived fallback still covers them.
Also fixes the avatar for rehydrated GHES forks, which resolved against
github.com for the same reason.
* docs(github-project): correct upstream host fallback comment
Persistence now keeps non-empty upstream.host; originIdentityKey remains
the host fallback for older records without one (CodeRabbit nit).
* fix(github-project): own slug-index retry timer cleanup
Move the failure-retry setTimeout into its own effect so cleanup always
clears it. Scheduling from the async buildIndex then-handler failed the
react-doctor effect-needs-cleanup gate in static analysis.
* test(github-project): guard the slug-index retry timer, fix the mobile twin comment
Two follow-ups on 52298d82 and 2f89c20d:
- Cover the retry timer both ways: a failed resolution still re-resolves after
the TTL and recovers the match, and the pending timer is gone after unmount.
The second fails if the timer moves back into the async then-handler, so the
property is guarded by more than the lint rule.
- The mobile matcher's comment made the same stale "persistence strips
upstream.host" claim that 2f89c20d fixed on the renderer side.
* test(github-project): unmount slug-index hooks so React cannot flush after teardown
CI shard `tests node 24 6/16` failed with 10 unhandled
`ReferenceError: window is not defined` traced to this file. The tests mounted
hooks without unmounting, so React scheduler work flushed after the DOM
environment was disposed. All assertions passed; the shard failed on the
unhandled errors alone.
`cleanup()` after each test unmounts the trees. Does not reproduce locally in
isolation — it needs CI's worker pooling and file ordering.
---------
Co-authored-by: Jinwoo-H <Jinwoo-H@users.noreply.github.com>
* fix(permissions): surface macOS silent Local Network denial with diagnostic and workaround (STA-3505)
On macOS 27 beta, NECP silently denies Orca's whole process tree Local
Network access: no prompt fires, the app never appears in System
Settings, and terminal child processes fail with EHOSTUNREACH. The
Settings trigger swallowed the probe's socket error and reported
'unknown' + a 'Permission request sent' toast, indistinguishable from
success.
Classify the mDNS probe outcome (EHOSTUNREACH/EHOSTDOWN -> denied,
clean send -> granted, else unknown), remember the verdict for the
status chip, and render an inline diagnostic with the documented
NECP re-evaluation workaround when denial is detected.
* fix(permissions): avoid false Local Network grants
* fix(permissions): use standard Local Network request flow
* feat(permissions): add local network connection test
* fix(permissions): nest local network connection test
* fix(permissions): collapse connection test by default
* fix(permissions): emphasize connection test action
* fix(permissions): restore outlined connection action
* fix(azure-devops): retry with -preview api-version and keep project-level Git base for on-prem Server (STA-3494)
Azure DevOps Server rejects api-version=7.1 with 400
VssInvalidPreviewVersionException unless the -preview suffix is supplied,
so auth and every Git endpoint failed. Retry once with -preview on that
rejection and remember the requirement per origin. Also stop letting a
same-origin ORCA_AZURE_DEVOPS_API_BASE_URL (collection-level, needed only
for the connectionData auth probe) override the project-level base derived
from the remote for Git endpoints; cross-origin (proxy) overrides keep
working.
* fix(azure-devops): constrain preview retry and base override
Reverts the classification change and keeps behavior at base.
cursor-agent's native OSC title is the bare literal "Cursor Agent" and never
carries a status word, so it names the agent without proving one is present.
The title tracker drops it live, so main records it only when the stale-working
timer strips the spinner off the synthesized "⠋ Cursor Agent" — and that fires
both when Cursor parks idle and when cursor-agent exited and the shell reclaimed
the pane. The two states are observationally identical: same title, same null
foreground read.
Classifying it as an agent therefore removes a refusal rather than adding
evidence. Guarded sends auto-submit Enter, so the false positive types into the
user's shell. A null foreground is also not "unreadable" on the default local
provider, which returns null when the pty is gone.
hasPty, probePtyLiveness, hasChildProcesses and inspectProcess were each checked
as corroborating signals; none separates alive-with-agent from alive-with-shell
when the foreground read is unavailable.
Tests pin every no-evidence branch fail-closed and document the mechanism, so
both attempted fixes fail loudly if reintroduced.
Real gap tracked in #12946.
* fix(ai-vault): group OMP task subagent transcripts under their parent session
OMP persists task-child transcripts inside the parent session's same-named
artifact directory (<stamp>_<uuid>/), and discovery scanned them as ordinary
top-level sessions - a coordinator's history drowned under its own workers.
Extend the existing Claude subagent model to OMP, classifying purely by the
artifact-dir layout (never by a transcript's parentSession field, which also
describes non-task lineage):
- prune artifact dirs from the top-level scan (name-pattern predicate)
- count direct-child transcripts onto the parent row (local readdir; remote
walks partition their listing instead, mirroring Claude's SSH posture)
- list children on demand via the existing listSubagentSessions IPC, titled
by their coordinator-given task label and linked to the layout-derived
parent id
- refresh the count on zero-turn cache reuse, matching Claude
- extract session-scanner-roots.ts so the renderer-supplied-path allowlist
for both agents lives in one module
Fixes#9330
* review: harden OMP subagent classification and cover its uncovered branches
Prune predicate now skips depth 0 (the workspace dir), so a workspace whose
name happens to look like a session stem keeps its sessions. Drop degenerate
OMP roots in ompSessionsRootDirs: OMP_CODING_AGENT_DIR='/' normalizes to '',
which resolve()s to the process cwd and would have allowlisted it for the
renderer-supplied subagent path.
Rename session-scanner-omp-subagents.ts to -omp-subagent-transcripts.ts so it
mirrors Claude's transcripts/lister split by role rather than inverting it.
Correct two comments that asserted things the codebase contradicts: OMP task
children do carry their own sessionId and would resume by path (OMP's own
picker globs `*/*.jsonl`, so it never offers them either), and workspace dir
names are not uniformly dash-prefixed.
Cover branches the change added with no test: the remote/SSH partition wiring,
the IPC `omp` gate and per-agent allowlist, the parse-cache zero-turn recount,
and the executionHostId disk-ownership guard. Extract the remote scanner's
in-memory provider into a fixtures module to stay under the max-lines cap.
* review: note why child rows carry an unrendered grandchild count
* review: describe the real OMP grandchild layout in the pattern comment
---------
Co-authored-by: Dan Cieslak <dcieslak19973@users.noreply.github.com>
Co-authored-by: Jinwoo-H <Jinwoo-H@users.noreply.github.com>
* refactor(mobile): demote address picker to optional disclosure on Relay
Relay provides remote access without requiring a specific local address,
so hide the picker behind a disclosure to keep the direct fast path
accessible without visual clutter. Reposition Sign in between the Relay
and LAN options to clarify it's Relay-specific. Keep custom addresses
always visible and force the disclosure open when settings search
targets the address picker.
* refactor(mobile): improve relay pairing guide and interface ranking
- Rank Docker/VirtualBox bridges below real LAN addresses so they're never auto-advertised as the default
- Clarify UI copy: 'Local network address (optional)' → 'Direct connection on this network'
- Better explain direct connection vs Relay roles and when each is used
- Fix Relay unavailability to be a build property, not dependent on current selection
* refactor(mobile): reframe local network address as optional in relay pai
Demote the address picker from primary action styling to an optional
disclosure with quieter visual treatment. Update messaging from "Direct
connection on this network" to "Also use a faster local path" to
clarify Relay is the default path and local addressing only applies
when nearby. Add explanatory hint text to set expectations that Relay
remains available when away.
* Retire SSH worktree metadata an authoritative scan proved gone
The metadata fallback's protection against resurrecting externally deleted
worktrees lived only in renderer module state, so it died on every reload
while the SSH WorktreeMeta it guarded against persists forever
(gcStaleWorktreeMeta exempts any repo with a connectionId, because a local
existsSync cannot probe a remote path). Repro: `git worktree remove` on the
SSH host, let the authoritative scan purge the row, restart — the startup
fetch runs before SSH connects and the fallback re-lists the deleted
worktree as a ghost row.
Chose option (a), deleting the stale persisted meta in main, over persisting
the removal memory: the metadata is the thing that outlives the worktree, and
Orca's own removals already delete it (removeWorktreeMetadataAndTransientState),
so external removals now converge on the same end state instead of accumulating
a second, parallel tombstone list that would itself need eviction. The
in-session memory stays for the window before the async delete lands.
New `worktrees:forgetRemovedForExecutionHost` only accepts SSH hosts, requires
an exact repo owner, skips metas owned by another host, and refuses folder
repos — a folder workspace's meta IS the workspace record (gcStaleWorktreeMeta
skips those keys for the same reason) and no remote scan can retire one. The
renderer only calls it from the authoritative-removal path, so a mere
disconnect never deletes anything.
Also:
- hoist resetAuthoritativelyRemovedWorktreeMemoryForTests into a top-level
beforeEach; removeWorktree writes that memory too, so suppression could leak
across describes and silently hide a row.
- cover the requireAuthoritative gate that skips the fallback, which had no test.
- replace the raw NUL byte committed inside the coalesce-key template literal
with a \0 escape; it made the file scan as binary to grep/ripgrep.
* test(worktrees): verify non-authoritative fallback skips removal
The non-authoritative fallback must not trigger worktree cleanup when it observes an absence — only an authoritative scan should. Tighten the expectation to ensure cleanup happens exactly once, when new data arrives after the connection state changes.
* fix(daemon): detect severed macOS TCC attribution and surface daemon-restart remedy (STA-3491)
macOS pins the detached PTY daemon's TCC responsible process to the app
binary that forked it. Once that binary is deleted (packaged updates
replace the bundle), Accessibility/Automation grants on Orca silently
stop covering every daemon-hosted terminal: osascript/System Events
fails with -25211 no matter what the user grants.
- record spawnerExecPath in the daemon pid file at fork
- adoption checks it: severed + 0 live sessions -> replace the daemon
(reason severed_tcc_attribution); live sessions are preserved
- Settings (Developer Permissions + Manage Sessions) show a visible
banner pointing at Manage Sessions -> Restart while severed
* fix(daemon): harden TCC attribution recovery
* fix(menu): restore paste in macOS native dialogs
* test(menu): pin platform for paste routing coverage
CI runs unit shards on ubuntu-latest only, so the unpinned paste-routing
test asserted "no native paste" vacuously: a double-route regression
passed green on Linux. Cover darwin/linux/win32 explicitly so the
exactly-once contract holds on every platform, and assert the item exists
in the negative-only cases so a rename cannot pass them silently.
Also document why the native first-responder fallback exists.
---------
Co-authored-by: Jinwoo-H <Jinwoo-H@users.noreply.github.com>
* feat(dashboard): add experimental agent map view
* fix(dashboard): harden agent map behavior
* fix(dashboard): harden agent map recovery
* fix(dashboard): close map selection on view change
* fix(agent-map): center sparse layouts
* fix(agent-map): align completion and workspace actions
* Polish agent map interactions and repo labels
* feat(agent-map): add worktree lineage and project actions
* fix(agent-map): use marker for unread agents
* fix(agent-map): compact orchestrated families
* fix(dashboard): harden agent map actions and layout
* fix(agent-map): bound layout work and preserve interactions
* fix(agent-map): move unread marker to ring top-right
* fix(agent-map): seat unread marker on the ring's top-left edge
* feat(agent-map): restore the agent launcher and declutter map labels
Three gaps in the experimental Agent Map:
- The "start a new agent" picker was split onto a preserved branch during the
08-02 rebase (47829cb226) and never re-landed. Restores that commit and its
pop-out IPC, keyed on the raw worktree id rather than the map identity.
- Workspace labels draw at a fixed screen size with no collision handling, so a
zoomed-out map stacked dozens of names on each other. Adds a declutter pass
that seats project names first, then workspace names by attention, then
project counts in whatever room is left.
- The pop-out had no workspace right-click at all: its renderer has no store, so
the shared sidebar menu cannot mount there. Adds a snapshot-driven menu with
the launcher and Sleep, relayed to the main renderer.
* refactor(agent-map): fold the map's filter rail into the shared toolbar filter
The rail duplicated the toolbar's project filter and cost the canvas 14rem of
width on the surface that needs it most. Agent states move into the toolbar's
Filter dropdown (map view only — the board's columns already separate them) and
count toward its badge; project filtering falls back to the toolbar's own. Show
all is the dropdown's Clear all, and Fit already lives in the viewport controls.
* fix(agent-map): isolate map work from main renderer
* perf(agent-map): stream status updates to popout
* fix(i18n): add agent map catalog entries
* feat(agent-map): glow working entities
* fix(agent-map): prioritize attention ring status
* fix(agent-map): distinguish subagent connectors
* fix(skills): match official skill files despite local sidecars
Scope known-snapshot matching to manifest-listed files so agent-written
sidecars (e.g. agents/openai.yaml) no longer mark a package unrecognized
and block updates when official bytes still match.
Preserves fail-closed detection when a listed file's content drifts.
Fixes#12694
* fix(skills): scope lock trust and convergence to official files too
Sidecar tolerance stopped at the snapshot match, leaving three disk-vs-official
comparisons still judging the whole folder.
The lock-comparable hash covered every observed file, so a clean update beside
agents/openai.yaml reported as failed and read 'may be modified'. It is now
carried both whole and scoped to the current bundle's paths, and either may
satisfy the lock: the sidecar case only ever matches scoped, while an upstream
revision that ADDS a file only ever matches whole, so publishing one alone
would trade this bug for #11220.
Convergence re-derived the disk revision from that same whole-folder digest,
which no revision matches once a sidecar lands, retiring the stuck-lock gate
and arming an update the command provably cannot perform; it now honours the
revision observation already resolved.
Subset matching also let an older revision launder drift on a file the current
bundle lists, since that revision does not list it and so read it as a
neighbour. Identity now keys tolerance on what the current bundle owns.
---------
Co-authored-by: Jinwoo-H <Jinwoo-H@users.noreply.github.com>
httpProxyUrl is the only network setting stored via safeStorage. Two
failure modes silently killed the configured proxy on macOS:
- A keychain reset/denial makes decryptString throw at load; the raw
ciphertext then masqueraded as the configured proxy URL, so
applyElectronProxySettings silently fell back to DIRECT and the
garbage re-persisted forever (no self-heal).
- safeStorage.isEncryptionAvailable() throwing (keychain/API errors,
pre-ready use) was uncaught in encrypt/decrypt, failing the entire
state save - the data file never gained the proxy keys at all.
Load now validates the decrypted value and clears undecryptable
ciphertext (plaintext URLs still pass, preserving the pre-encryption
upgrade path), the availability check is exception-safe, and startup
logs when persisted proxy settings are invalid instead of silently
using direct networking.
Jinwoo Hong
Neil
OrcaWin
Avichal Dwivedi
Jinjing
Brennan Benson
Pongsakorn Paetrakul
Wooseong Kim
5hseok
plotarmordev
Shani Singh
BingZ
dcieslak19973
pstanton237