The scrollback demotion feature for eviction-exempt hidden worktrees is no longer needed. Retention budget limits are now sufficient without this additional bound. Remove the terminal-hidden-scrollback-demotion module, the selectScrollbackDemotedTerminalWorktrees function, and related per-pane demotion logic.
When a parked pane reveals mid-Command Code turn, the new detector
cannot re-observe the already-passed idle composer. Cancelling the settle
leaves the row stranded at 'working', so dispose now flushes the pending
settle instead.
Extract readInFlightCommandCodeTurn to shared space and seed detectors
with in-flight turns so remounts complete mid-flight commands. Also
memoize SSH model probes to prevent double timeouts on reattach.
Seed the byte watcher with in-flight turn state from agent status: the
watcher is recreated per park cycle with no startup command to arm it,
and the banner scrolled away before parking. Also memoize
eviction-exempt checks and use SSH PTY ID builder in tests.
Implements the C1 retention contract: preserve worktree `hiddenSinceMs` through a
background-measure window (so TTL/ranking stay honest), but re-park waits for a
full `coldParkDelayMs` cool-down after the measure ends. Without the cool-down,
every ~3s measure lease on a past-deadline worktree thrashes remount/reattach.
Core changes:
- Terminal.tsx: add measure clock (measuringTerminalWorktreeIdsRef) and post-measure
cool-down tracking (terminalWorktreeParkCooldownUntilRef); gate parking candidates
until cool-down expires.
- Extract snapshot replay choreography to shared terminal-snapshot-replay-paint.ts
(used by SSH reattach + daemon restore paths).
- Add SSH model snapshot timeout (750ms) with fallback to relay replay.
- Move cold-park recheck deadline logic to terminal-cold-park-recheck-deadlines.ts;
add cool-down deadline to scheduling.
- useTerminalTabColdParking: implement matching measure-clock contract with per-tab
cool-down gate to keep tab deadlines synced with worktree retention clock.
- Add resolveTerminalMountScrollbackRows() to demote new xterms under demoted
worktrees (pane births during demotion must take the demoted tier at create).
- Add kill switches: terminalSshViewParking, terminalHiddenWorktreeRetentionBudget,
terminalHiddenScrollbackDemotion.
The first draft added a second remote repo mid-session, whose pane pty
spawn misroutes to the local daemon with the remote cwd (pre-existing
multi-repo issue, reproducible without any retention override — a seeded
local repo plus one remote repo shows the same misroute). The spec now
budgets across three worktrees of the ONE connected repo, created through
the product createWorktree path (an external git-worktree-add only lands
as a detected worktree needing adoption) and polled through the relay's
transient post-connect reconnect window. Verified green on the local
Docker lane in 20.8s.
Co-authored-by: Orca <help@stably.ai>
ORCA_E2E_TERMINAL_RETENTION_LIMIT flows preload → e2e-config →
getTerminalParkingPolicyOverrides (exposeStore-gated, positive-integer
only) so a spec can shrink the force-park budget to 1. The Docker-gated
spec opens two remote worktrees on one relay target (second pre-seeded
remote repo), disables terminalSshViewParking to make both un-parkable,
hides both behind the local context, and proves the older one force-parks
while the last-active exemption spares the newest; re-activating the
evicted worktree restores the marker tail via relay replay.
Co-authored-by: Orca <help@stably.ai>
Parked fact-mode watchers omitted onCommandFinished/onCommandCode*, so
OSC 133;D and Command Code scrape signals went dark while parked. New
parked-terminal-command-status.ts ports the store-level subset: git-UI
nudge on every command finish, same-turn status-row drop for SSH PTYs
(exact mounted-path parity — the foreground tracker refuses SSH ids),
and the Command Code working seed / 1500ms done settle. Byte mode scans
the same shared parsers for authority-off parity. Local-PTY status drops
stay with the mounted pane: they need pty-connection's process-confirm
ladder to tell a leaked nested-shell 133;D from a real agent exit.
Co-authored-by: Orca <help@stably.ai>
pendingSideEffects grew without bound under background timer throttling
(~64 drained/s vs hundreds queued/s overnight). Cap at 512 entries with
oldest-first eviction: titles drop (last-wins), a pending bell latches
onto the next survivor, agent-status payloads collapse onto the survivor
keeping the newest 16 (last-wins store state, KB-scale strings).
Co-authored-by: Orca <help@stably.ai>
isEvictionExemptTerminalTab read only tab.ptyId — the FIRST leaf's pty —
while the coverage veto that makes a worktree a retention candidate walks
every pane. A split tab whose second leaf held an unrestorable pty therefore
failed coverage (→ force-park target) yet looked exempt-free, so force-park
unmounted it and orphaned the live shell. The exemption now resolves panes
through the same resolveParkedTerminalPaneCandidates, keeping tab.ptyId in
the union for the no-layout/no-capture case.
Also from the same review round:
- force-park's capture passes includeLocalBuffers:false like every other
shutdownBufferCaptures caller; it was serializing up to 512KB/pane of
scrollback into the store inside a fix meant to bound renderer heap.
- Terminal.tsx unmount resets the scrollback-demotion registry — module
state with no reset path, read by a pane effect that runs before the host
effect that would clear it, so a stale verdict trimmed restore replays.
- memoize watcher coverage per tab within the parking pass; the retention
candidates re-asked it for every mounted worktree, not just the parked few.
One contract matching the code: status IMPLEMENTED around force-park (not
the unmount proposal), real kill-switch names with coupling + revert
matrices, the true retention-floor formula with measured per-pane and
demotion numbers, an explicit when-OOM-is-still-possible paragraph naming
the H2 pendingSideEffects residual, the applyMainBufferSnapshot deadlock
constraint inside the slice-A section, stable-signal phrasing instead of a
capability latch, fail-open AND foreign-worktree exemption class, verified
cites, and a planned/landed/follow-up test matrix.
Co-authored-by: Orca <help@stably.ai>
Pad the session with ~180KB of output after the numbered markers so the
earliest marker falls outside the relay's 100KiB rolling replay buffer while
staying inside main's ~5k-row headless model; asserting marker_1 after
reveal now proves the headless-model paint rather than passing under the
relay fallback.
Co-authored-by: Orca <help@stably.ai>
Whole-worktree background mounts (browser-automation bootstrap lease, mobile
mounts, agent wakes) open a ~3s self-clearing measure window that previously
deleted hiddenSince — every remount restarted the 30s hysteresis and the
45min retention TTL, so a periodically re-mounted force-parked worktree
never re-parked. The measure window still pauses parking/eviction verdicts
(all selectors skip measuring candidates); only the clock survives, so the
prior verdict resumes as soon as the window closes. Visible and
portal-holding worktrees still reset the clock.
Co-authored-by: Orca <help@stably.ai>
The last-active exemption means a single hidden un-parkable worktree never
force-parks — and slice C previously only targeted exempt-tab worktrees, so
its panes held full scrollback forever. Demotion now also covers un-parkable
non-exempt worktrees past the retention TTL that are absent from the
force-parked set (last-active spared, or slice B switched off). Membership
stays time-monotone for fixed inputs; covered by new idempotence/monotone
selector tests.
Co-authored-by: Orca <help@stably.ai>
One eviction-exempt tab (fail-open/foreign pty) previously vetoed force-park
for its whole worktree, pinning co-located remote-runtime tabs forever. The
worktree now force-parks while exempt tabs keep their mounted panes via a
per-tab exclusion mirroring the Activity-portal pattern (legacy watcher sync,
legacy render, and the overlay cold-parking hook). Ordinary parking is
untouched — a worktree with an exempt tab still cannot ordinary-park.
Slice C now also demotes exempt tabs' panes as soon as their worktree
force-parks under the count budget (they are the only panes left mounted).
Co-authored-by: Orca <help@stably.ai>
Per the approved contract each slice reverts behind its own switch: slice C
now requires only the master terminalHiddenViewParking plus its own
terminalHiddenScrollbackDemotion flag. The TTL wakeup timer fires for
demotion candidates even with the budget switch off. No DEFAULT_SETTINGS
entries exist for sibling flags (defaults are the '!== false' optional
pattern), so no explicit defaults are added.
Co-authored-by: Orca <help@stably.ai>
A relay restart empties the replay buffer; the reveal previously painted
nothing even when main's headless model held the session. The reattach now
prefetches the model snapshot when no structural replay exists (SSH-shaped
ptys only) and paints it inside the coordinator; emptiness is judged on the
composed payload (scrollbackAnsi + data + pendingEscapeTailAnsi) so an
alt-screen snapshot with an empty screen frame still paints.
Co-authored-by: Orca <help@stably.ai>
Docker-gated (ORCA_E2E_SSH_DOCKER=1) spec: SSH tab parks behind a decoy and
reveal restores marker content at multi-viewport scrollback depth. DESIGN.md
records the as-built deltas (inline paint, force-park shape, last-active
floor) and the residuals so follow-ups aren't lost.
Co-authored-by: Orca <help@stably.ai>
applyMainBufferSnapshot runs its own structuralReplayCoordinator.run; calling
it from applyReattachPayload (already inside the coordinator when a relay
replay exists) deadlocks on the coordinator's tail chain. The model paint now
mirrors the daemon-snapshot branch inline (folded scrollback + rehydrate +
screen, dimension-matched, escape tail last) and arms the restored-snapshot
seq baseline so deferred/live chunks the snapshot covers dedupe instead of
double-painting. Also falls through (no early return) so reattachPayloadApplied
still latches. Adds the folder-workspace id parity unit case.
Co-authored-by: Orca <help@stably.ai>
The retention budget (slice B) must exempt worktrees holding fail-open or
foreign-worktree ptys — a remount would fresh-spawn and orphan the live
shell — which would leave that class unbounded again. Instead, past the same
45min retention TTL their hidden panes drop to the minimum scrollback tier
(measured: ~19MB -> ~1.3MB V8 heap per 50k-row pane; trimmed history is
gone by design, reveal restores the configured cap for future output).
- terminal-hidden-scrollback-demotion.ts: module-state verdict registry
(parked-watcher pattern) with content-equality notify damping; applied in
the existing scrollback-rows effect in use-terminal-pane-lifecycle.
- selectScrollbackDemotedTerminalWorktrees: pure, TTL-gated, time-monotone.
- Retention TTL wakeups now also cover exempt worktrees so demotion fires.
- Kill switch: settings.terminalHiddenScrollbackDemotion (default on).
Co-authored-by: Orca <help@stably.ai>
Un-parkable worktrees (remote-runtime ptys, uncoverable tabs, SSH with the
slice-A switch off) had unlimited retention: the parking cap/TTL only ever
saw eligibility-passing worktrees, so one bad tab pinned a whole worktree's
panes forever. Retention is now memory-bounded, not eligibility-bounded.
- terminal-hidden-worktree-retention.ts: retention budget (12 hidden / 45min
TTL, sized from the measured 2.5-19MB per-pane V8 cost, DESIGN.md §2) over
hidden worktrees ordinary parking can never evict; reuses the hot-retain
ranking so last-active exemption, deterministic ties, and deadline-driven
rechecks hold. Fail-open/foreign-pty tabs are eviction-exempt (a remount
would fresh-spawn and orphan the live shell).
- Terminal.tsx: force-parked ids join the parked set AFTER the coverage veto
(darkness for uncoverable tabs is the accepted cost); buffers captured via
the sleep-flow registry before the unmount render; retention TTL added to
the recheck deadlines for budget candidates only.
- Verdict stays out of its own effect deps; policy test asserts idempotence
and time-monotone membership (flip-loop dwell regression).
- Kill switch: settings.terminalHiddenWorktreeRetentionBudget (default on).
Co-authored-by: Orca <help@stably.ai>
SSH ptys were blanket-excluded from hidden-view parking, so a hidden SSH
worktree retained every pane forever (C1: renderer heap climbs to the V8
ceiling). SSH bytes transit local main — fact-mode watchers already cover
them, and main keeps a headless model served over pty:getMainBufferSnapshot
that the SSH reattach path never consulted.
- isParkRestorableTerminalPty: snapshot-backed OR (SSH + policy); threaded
through both park verdicts, both selectors, watcher coverage, and the
watcher start guard. Remote-runtime/fail-open/foreign/null unchanged.
- Parked-SSH reveal paints from main's headless model (dimension-matched,
~5k rows) and degrades to the relay 100KiB replay unless the snapshot is a
non-empty source==='headless' payload — never a blank/stale paint.
- Kill switch: settings.terminalSshViewParking (default on).
DESIGN.md records the approved plan and the H1 magnitude non-claim.
Co-authored-by: Orca <help@stably.ai>
* fix(native-chat): surface draft launch context in chat composers
Creating a workspace from a GitHub issue delivers the issue link only into
the agent TUI's input buffer (argv prefill or startup paste), so the chat
view showed no trace of it on desktop or mobile.
Desktop: draft launches now seed an in-memory launch draft keyed by tab id
(direct work-item launches, background GitHub work-item creates, quick-create
composer, and new-tab draft deliveries). The chat composer adopts the seed
once as its editable draft, declines permanently if the composer already has
text, and drops an untouched copy when any user turn lands (the one-line TUI
input means the prefill was submitted or deliberately cleared) or on its own
send, whose existing input pre-clear retires the TUI copy.
Mobile: the host publishes the draft as an optional launchDraft field on the
mobile terminal tab snapshot (additive, no protocol bump) and the mobile
composer adopts it with the same once-only/decline/resolve semantics. Mobile
chat sends now also pre-clear the TUI input line (Ctrl+U, desktop parity) so
a pending prefill cannot concatenate with the sent message.
Completion seeding resolves the launch tab from the synced store tabs when
the backend spawned the terminal and activation reports no primaryTabId.
Split the Windows shell-quoting tests into their own file to stay within the
max-lines budget.
* revert(mobile): drop incidental pnpm-lock churn from the launch-draft branch
The libc binding fields and the @typescript-eslint peer re-resolution came from
a local install, not from this change; mobile/package.json is untouched.
* fix(native-chat): resolve launch drafts without trusting cross-host clocks
The rule required a user turn stamped at or after the seed. Grok omits row
timestamps, so a Grok launch draft never resolved; and the seed time is a
renderer clock while the stamp comes from the executing host's JSONL, so a
remote workspace whose clock trailed never resolved either. Both left the
composer adopting an already-submitted prefill, which re-sends it as a
duplicate turn.
Resolve on any user turn that is not PROVABLY older than the seed (a launch
draft's session starts with zero user turns), with the existing cross-host
skew slack, plus a timestamp-free backstop for wider skew: a new tail user
turn since the draft was first observed. "Load earlier" prepends, so it
cannot move the tail and cannot over-resolve.
Split out of native-chat-pending.ts to stay under the max-lines ratchet.
* fix(worktrees): seed the launch draft on the agent's own tab, never on tabs[0]
Two defects in the completion seed:
- The tab was resolved by array position. buildStartupOpt returns undefined on
the backend-spawn path, so applyDefaultTerminalTabs stamps launchAgent on no
tab and the launchAgent guard was dead there. A repo with default terminal
tabs ("dev server", "logs", ...) got the draft on a tab that runs no agent,
and then published it to mobile as THAT tab's launchDraft. Correlate on the
backend startup tab, then on a launchAgent-stamped tab, then on primaryTabId
(which is the agent tab whenever the renderer owns startup); never tabs[0].
- Runtime-owned worktrees mirror their session tabs async, so tabsByWorktree
was empty at seed time and the seed was silently dropped for that whole host
class. Defer to the first mirrored tab via the existing delayed-delivery
queue, which now holds every pending delivery for a worktree instead of one
(setup/issue commands and the seed both wait on the same first tab).
* fix(store): evict nativeChatLaunchDraftByTabId on every teardown path
The new map was absent from all four paths its sibling
nativeChatLaunchPromptByTabId participates in: tab close, the orphan terminal
sweep, the bulk worktree purge, and the removeWorktree teardown. A stranded
entry is worse than a plain leak here because sync-runtime-graph keeps
publishing it to mobile as that tab's launchDraft.
* fix(native-chat): only seed single-line unsubmitted launch drafts
The unsubmitted-delivery branch seeded on every draft delivery, which also
caught the agent-session-fork path whose prompt is multi-line scraped context.
The chat send pre-clears the TUI with Ctrl+U (kill-to-start-of-LINE), so a
multi-line prefill cannot be fully cleared and its earlier lines would glue
onto the next message. The GitHub work-item draft this feature targets is a
bare issue URL, so narrowing costs it nothing.
Also assert the composer retires the seed after a send — deleting that call
previously failed no test.
* fix(mobile): stop the chat pre-clear from wiping a just-pasted image
The text write set clearInputFirst unconditionally. On the image path that
Ctrl+U lands AFTER pasteMobileNativeChatImagePaths already pasted the image,
so the agent receives the text alone while acceptSend still renders the
thumbnail on the sent bubble — silent image loss.
Desktop's image path clears exactly once, before the paste, and never again;
mobile now matches: pre-clear only when nothing was deliberately pasted first.
The image paste already leads with its own Ctrl+U, so a launch-draft prefill
parked on the input line still cannot glue onto the message.
Pinned at both levels: the controller test drives the real send hook and
asserts clearInputFirst per branch, and the send module asserts the wire text
carries no leading \x15. The image-attachments test injects its own baseSend,
so it structurally could not observe this.
* fix(mobile): hold the launch-draft prefill until the transcript settles
session.tabs delivers launchDraft before the transcript read resolves, so the
seed effect could run against an empty in-flight message list and miss the
user-turn decline. Launching from an issue, submitting the prefill in the TUI,
and never opening desktop chat (nothing else clears the host seed) then
prefilled the mobile composer with the already-sent issue link — a send tapped
before it retracted duplicated it to the agent.
Thread the session's loading state through and skip the seed while the read is
in flight. idle/waiting-session still seed: no session means no user turns.
* fix(runtime): publish a launch draft to mobile only for the tab's own agent
The publish had no agent check while the desktop consumer declines on
mismatch. The seed is keyed by tab id, which survives a pane's agent switch, so
mobile could adopt a draft desktop refuses — seed for claude, never open
desktop chat, switch the pane to Codex, and mobile prefills the Codex chat with
the Claude-era issue link. Align publish with the consumer.
* fix(native-chat): take the launch-draft baseline only after the transcript loads
The timestamp-free backstop snapshotted the transcript's user turns on first
observation of the draft, which can happen while the read is still in flight and
`messages` is []. A pane bound to a session that already had user turns then
backfilled above that zero baseline with a different tail id, so clause 2
resolved and silently dropped the seed — the launch context never appeared, and
the feature no-oped for exactly the panes it was meant to serve. Clause 1 was
already correct there (that history is provably older than the seed).
Gate baseline capture and resolution on the transcript read settling, the same
shape mobile's drafts hook uses. Clause 1 is unchanged; while loading the merged
list is empty anyway, and a pane with live appends is never reported 'loading'.
Also restore clause 1's short-circuit: it scans with .some() again and only
allocates the user-turn list when falling through to the backstop.
NativeChatView sat at exactly the 400-line cap, so the composer's two
launch-draft props are now spread from the hook result they already mirror.
* fix(native-chat): reject multi-line launch drafts inside the seed helper
The single-line guard lived in deliverLaunchPromptToAgentTab, so the two
other seeding entry points (worktree create, direct work-item launch)
bypassed it — and every Linear launch is multi-line by construction
("Linked Linear issue: STA-…" + url). The chat send pre-clears the TUI
with Ctrl+U, which kills to start of LINE, so those earlier lines stay
parked to glue onto the next message.
* fix(worktrees): keep the deferred agent seed off ambiguous mirrored tabs
The runtime-owned deferred path fell back to tabs[0], which the module's
own docstring forbids: with repo default tabs ("dev server", "logs") the
seed lands on a tab running no agent, where mobile withholds it and
desktop's agent check ignores it — the feature is silently dead for that
create and the entry leaks until tab close.
The queue entry is consumed before delivery, so there is no retry to fall
back on; accept the first mirrored tab only when it is the worktree's
only one and so unambiguously the agent's.
* fix(mobile): treat a launch-draft-only session-tab frame as a change
mobileSessionTabEqual's terminal branch never compared launchDraft, and
the route keeps `prev` when tabs compare equal — so a publish whose only
delta is the draft appearing or retracting was discarded and never
reached the composer. Live QA passed only because agentStatus happened to
change in the same frame.
MobileSessionTab's terminal variant did not declare the field either
(the controller read it through the structurally wider
MobileNativeChatTab), which is why TypeScript never flagged it.
* fix(mobile): judge a launch prefill only from its own settled transcript
Two ways the drafts hook was reading a transcript that was not the active
chat's:
- transcriptLoading came from `status`, a plain useState written by a
passive effect declared before the drafts hook. On the commit where the
tab identity changes it still holds the previous tab's value, so the
guard was off on exactly the render that seeds: first entry saw
status 'idle' with an empty list and seeded an already-submitted link,
and a tab switch declined the new tab's prefill from the old tab's
turns. The session hook now tracks the identity its messages describe
and reports transcriptLoading until they agree; the retire effect gates
on it too.
- Leaving chat view nulled launchDraft while draftKey stayed the same,
which the hook could not tell from a host retraction — it declined the
prefill permanently, so peeking at the terminal dropped the context.
The controller now passes the raw field plus an explicit chatActive
flag, and both effects hold their state when the tab is not on chat.
The controller wiring was previously unasserted: replacing both props
with constants left all 795 mobile session tests green.
* fix(native-chat): keep the launch-draft baseline across a transcript reload
baselineKey went null whenever the transcript was loading, and the null
branch DISCARDED an already-valid baseline taken from a settled read. It
was then re-taken from the fuller list, swallowing the very user turn
that resolves the draft — so a stale prefill gets re-adopted as a
duplicate turn. Key the baseline on draft identity alone and gate only
the capture.
session.status is also not a truthful read-in-flight signal: a live
'working' hook outranks 'loading', so the guard could be off over an
in-flight empty list. Expose the read phase itself and gate on that.
* test: cover the launch-draft reducers and the sync-key skip gate
Every consumer test injects the three launch-draft reducers as bare
vi.fn()s, so reducing markNativeChatLaunchDraftAdopted to a no-op left
2609 tests green — while in the app the composer would resurrect the
prefill after every manual clear.
canSkipRuntimeMobileSessionSyncKeyBuild had no launch-draft case either:
when it skips, the sync key is never even built, so the existing
getRuntimeMobileSessionSyncKey case cannot catch its removal.
* fix(native-chat): hold the launch-draft baseline in state, not a render-mutated ref
react-compiler rejects reading or writing a ref during render. Adjust the held
baseline with the sanctioned render-time setState instead, keeping the local
copy so the render that first sees a settled transcript resolves against it.
* fix(mobile): carry the transcript identity in the session read state
react-doctor flags the separate loadedIdentity state as an extra render for a
derivable value. Hold status alongside the identity it describes in one state
written by the subscription effect, so transcriptLoading derives from it.
* test(native-chat): assert the readPhase contract without the hook-status race
The test asserted status === 'working', which depends on liveStatusOverride
winning over ambient transcript state — green locally, red under CI load. The
contract is that readPhase stays 'loading' once live content unmasks status,
so assert exactly that; it still fails if readPhase derives from status.
* fix(mobile): derive pre-read chat status instead of writing it from the effect
react-doctor's no-derived-state-effect flags idle/waiting-session/loading being
set in the subscription effect: all three are pure functions of the props. Derive
them during render and keep state only for the genuinely async outcome, tagged
with the identity it describes.
The tag now gates `messages` too, so a just-switched tab never sees the previous
tab's transcript at all rather than seeing it behind a loading flag.
* fix(mobile): drop a settled chat read once its subscription is torn down
The settled outcome was only ever replaced by a newly arriving frame, so any
effect re-run that landed back on an already-settled identity resurfaced it over
a list the same effect had just cleared: 'ready' with no messages and
transcriptLoading false. Toggling out of chat view and back hit this every time
(the agent goes null, then returns), flashing the "start a chat" empty state over
a real conversation and opening the launch-draft seed's decline check on an empty
transcript. A reconnect did the same via the client dep.
Identity and client are the effect's only inputs, so tagging the read with both
and dropping it during render when either moves covers every re-run.
* Improve translations for resource manager and related UI elements
Standardize terminology ("daemon" vs "service"), complete missing translations, and refine wording across Spanish, Japanese, Korean, and Chinese locales for consistency and clarity.
* Improve translations for resource manager and related UI elements
* fix(test): update zh name-mode label expectation after translation fix
The resource-manager translation pass correctly changed the Chinese
"Name" label from 姓名 to 名称; update the localized options unit test
to match so CI passes.
An account switch decides pane eligibility from a single cached
inspectProcess read. When that read reports the pane's shell for a live
Codex session, the pane silently loses its restart card - no error, no
retry. Re-check shell readings on Orca-launched Codex panes with the
existing fresh-scan confirmForegroundProcess before trusting them; only
an affirmative codex answer flips the decision, so a genuine exit to the
shell stays uncarded and unsupported providers keep today's behavior.
* feat: [AI-GEN] add Trae CLI as a supported TUI agent
Closes#10579.
Wire trae-cli into the desktop and mobile agent catalogs following the
same integration pattern as other CLI agents (e.g. Ante, Devin):
- src/shared/types.ts, tui-agent-config.ts: register 'trae' with
detectCmdAliases (traecli/trae-agent) and argv prompt injection,
matching trae-cli's `trae-cli [prompt]` contract. The CLI's own
third documented alias `ta` is intentionally excluded — too generic
a 2-letter name to use as a PATH-existence detection signal without
false-positiving on unrelated tools.
- src/shared/trae-headless-command.ts: recognize `--print`/`-p` and
`--output-format json|stream-json` as one-shot headless invocations
(same shape as claude-headless-command.ts) so they aren't mistaken
for a live interactive session.
- agent-kind.ts, telemetry-events.ts, agent-status-types.ts,
agent-type-label.ts, tui-agent-display-names.ts,
tui-agent-permissions.ts (YOLO via trae-cli's own --yolo flag),
tui-agent-selection.ts: standard per-agent registrations.
- agent-catalog.tsx, agent-favicon-assets.ts,
mobile/src/tasks/mobile-tui-agents.ts,
mobile/src/components/mobile-agent-icon-assets.ts: catalog entries
and bundled favicon (fetched from docs.trae.cn, required by mobile's
offline-icon invariant test).
- i18n: add the "Trae" label to all five locale catalogs (en/es/ja/ko/zh).
- Tests: agent-process-recognition, agent-status, tui-agent-startup.
Verified with `pnpm typecheck` (desktop + mobile), the relevant vitest
suites (869 tests across 12 files, all green), oxlint (clean), and a
real end-to-end launch of the actual trae-cli binary through Orca's
pty.spawn IPC path (confirmed via the OS process table).
* fix: [AI-GEN] point Trae catalog entry at the real CLI quick-start doc
docs.trae.cn/cli (what the installed CLI's own --help text prints as
its "User manual" link) soft-404s — the docs site restructured and the
working page is docs.trae.cn/cli_get-started-with-trae-cli (confirmed
by HTTP fetch: real page title "TRAE CLI 快速开始" vs the old path's
"404 - 页面不存在"). Addresses CodeRabbit's homepageUrl review comment.
* fix: [AI-GEN] detect Trae on traecli, not the ambiguous trae-cli name
Per @AmethystLiang's review: the open-source bytedance/trae-agent
project (MIT, ~12k stars) registers its own console script as
`trae-cli` (pyproject.toml: `trae-cli = "trae_agent.cli:main"`), an
entirely unrelated CLI with a different contract (`trae-cli run
"task"`, `-p` short for `--provider`). Detecting on bare `trae-cli`
would false-positive on that project's installs and break launch for
anyone who has it instead of the actual TRAE CN CLI.
- tui-agent-config.ts: detectCmd/launchCmd/expectedProcess -> `traecli`
(TRAE CN's own installer symlinks this alias too, but the other
project does not ship it). Dropped the `trae-agent` alias entirely —
it's the colliding project's literal repo name, the highest
false-positive string available.
- agent-catalog.tsx: cmd -> `traecli` to match; faviconDomain ->
`www.trae.cn` (bare `trae.cn` 404s on Google's favicon service;
`www.trae.cn` is the product-root domain that actually resolves).
- mobile-tui-agents.ts: faviconDomain -> `www.trae.cn` to match.
- Tests updated: agent-process-recognition now asserts `trae-cli` and
`trae-agent` are NOT recognized as Trae (regression guard against
reintroducing the collision); tui-agent-startup updated for the new
launch command.
promptInjectionMode stays `argv` and the headless-command file stays
as-is — both verified against the real TRAE CN CLI's actual --help
output (pasted in the PR review thread), not assumptions.
* refactor: [AI-GEN] share one print-mode headless matcher across agents
trae-headless-command.ts was a rename-only fork of claude-headless-command.ts,
and ante-headless-command.ts carried a third copy of optionName. Collapse both
print-mode files into print-mode-headless-command.ts, dispatch from a
Partial<Record<TuiAgent, ...>> table instead of an if-chain, and compress the
Trae comments to the repo's one-line style.
* fix: [AI-GEN] terminate Trae flag parsing before the positional prompt
`traecli` is a Cobra CLI with subcommands, so an argv prompt starting with
`help`, `config`, `-…` was dispatched as a subcommand or flag instead of being
run as the task. Add `argvPromptSeparator: '--'` (same reason Grok has it), and
stop the shared print-mode headless matcher at `--` so a prompt that reads like
`--print` no longer drops the pane out of agent recognition.
* docs: [AI-GEN] name both Trae CLIs explicitly in the detect-name comment
Co-authored-by: Orca <help@stably.ai>
* docs: [AI-GEN] drop the vendor tag from the Trae union comment
Co-authored-by: Orca <help@stably.ai>
* fix: [AI-GEN] guard the nullable startup plan in the Trae separator test
Co-authored-by: Orca <help@stably.ai>
---------
Co-authored-by: 陈泽榜 <chenzebang@jianzhikeji.com>
Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: Orca <help@stably.ai>
parsePositiveSafeIntegerNumericText mirrors the CLI's own Number()
coercion on purpose: text like `600000.000000000000001` is the budget
the CLI will actually wait on, so rejecting it here would leave the
relay and SSH kill timers shorter than the CLI's and cut the request
short. Document that and pin it with regression cases.
Removes the lowercase hourly/weekly/custom entries from the AutomationSchedulePicker catalog node in all five locales. They were superseded by keys derived from the rendered labels in #11068, and the extractor only adds keys, so they stayed behind unreferenced.
Follow-up to #11068.
The cadence picker now resolves Hourly, Daily, Weekdays, Weekly, and Custom cron through the renderer i18n catalog, with corrected zh/ko/es translations.
Supersedes #10044 (thanks @innocarpe) and #10045 (thanks @fsdwen).
Fixes#10043
* ci(pr): run E2E when a PR touches tests/e2e paths
Regression specs under tests/e2e never ran on PR CI — only schedule and
release called e2e.yml — so a red regression test could merge green.
Path-filter and workflow_call the E2E suite when E2E-relevant files change.
Use merge-base diffs so base-branch drift does not false-trigger E2E, fail
the detector when git diff cannot compute the PR range, and pin
least-privilege contents:read on both the detector and reusable E2E workflow.
Closes#10518
Co-authored-by: Wooseong Kim <innocarpe@gmail.com>
Co-authored-by: Orca <help@stably.ai>
* ci(pr): make the E2E path gate actually block, and match the real config path
Two fixes to the new path-filtered E2E job.
The gate did not gate. pr.yml's `verify` job is the required check, and it
enumerates its dependencies explicitly — `e2e` was in neither `needs` nor the
result list, so a failing shard left `verify` green. That reproduces the exact
hole this job exists to close: a red spec merges green, just with a red box
further down the page. Add `e2e` to both.
Because the job is path-filtered, `skipped` is the normal result on a PR that
touches no E2E files and has to keep passing. That allowance is checked after
the strict loop rather than inside it, so it can never leak to the six jobs
that are always required.
The `playwright.` pattern matched nothing. The config is
tests/playwright.config.ts — beside tests/e2e/, not inside it — so no tracked
file starts with `playwright.` and editing the runner config would silently
skip E2E. Anchor it at `tests/playwright.`.
Adds a contract test alongside the existing release-e2e one. Verified it fails
when either fix is reverted, and simulated the gate across
success/skipped/failure/cancelled plus the skip-must-not-mask-a-real-failure
case.
* test(ci): close two gaps in the E2E gate contract
CodeRabbit was right on both counts — verified by reverting each and watching
the contract stay green.
The path filter was unasserted, so `e2e` could lose its `if:` and run on every
PR — the cost the filter exists to avoid — without failing anything.
The strict-loop check hardcoded four of the six required jobs, so dropping
GIT_COMPATIBILITY or SHELL_CONTRACTS left them unenforced while the contract
passed. Derive the list from verify.needs instead, so a newly added required
job that misses the loop fails here rather than silently going unchecked.
* ci(pr): land the E2E path gate advisory instead of blocking
The E2E suite is currently failing every scheduled run on main — 22 of the last
22 — so making verify depend on it would block any PR touching tests/e2e/**,
including the PRs that fix the suite. This PR's own run reproduced that: 3 of 12
shards failed on specs unrelated to it (agent-session resume, Jira linking,
plugin containment, terminal artifacts).
So the job runs and reports on E2E-path PRs but is left out of verify.needs for
now. The detector, the tests/playwright. path fix, and the contract tests are
unaffected — those stand on their own and were the substance of the review.
Flipping to blocking is a three-line change once the suite is green; the exact
wiring, including why the skipped allowance must sit outside the strict loop, is
recorded on verify's Require-successful-checks step. The contract test pins the
advisory choice so it reads as deliberate rather than as the unwired-gate bug it
originally caught, and still fails if the path filter, the strict-loop coverage,
or the config path regress.
---------
Co-authored-by: Wooseong Kim <innocarpe@gmail.com>
Co-authored-by: Orca <help@stably.ai>
#8321 mixed the selected card's wash into the opaque --worktree-sidebar
surface, lifting dark mode to 16% (#4b4b4b). Card text lost too much
contrast against it.
Return both modes to a translucent wash (light 8%, dark 10%) so the
brighter selection border added by #8321 carries the selected state
instead of the fill.
Co-authored-by: Orca <help@stably.ai>
* fix(browser): keep the address bar editable in a narrow toolbar
Every other browser toolbar control is shrink-0, so the address bar was
the only flexible item and absorbed the entire squeeze: below roughly
420px of pane width it collapsed to the leading globe icon with a
zero-width input. Clicking it only opened the suggestion dropdown, which
inherits `--radix-popover-trigger-width` and so rendered at icon width —
there was no way to type or edit a URL in that tab.
Focusing a squeezed bar now lifts the form out of the toolbar flow and
overlays the row edge to edge, giving a full-width editable field that
navigates on Enter (and a full-width suggestion list for free). A
measured slot stays in flow so the overlay cannot feed back into its own
width, and the slot keeps a min width so the globe remains a real hit
target instead of being overlapped by neighbouring buttons.
Fixes#11090
Claude-Session: https://claude.ai/code/session_01Mx53f7erbtw5NraS8HdXKE
* fix(browser): use the documented floating shadow for the expanded bar
STYLEGUIDE.md defines exactly three elevation levels and forbids a
fourth; shadow-md was not one of them. The overlaid address bar is a
floating surface, so it takes the documented floating shadow already
used by the other floating surfaces in this pane.
Claude-Session: https://claude.ai/code/session_01Mx53f7erbtw5NraS8HdXKE
* test(browser): make the narrow-toolbar regression deterministic
The spec passed only from a clean profile. Two preconditions it set once are
actively undone by the app:
- BrowserPane re-focuses a blank tab's address bar across several animation
frames plus the blank-url did-finish-load handler, so a single blur() was
reverted and the bar never reached its squeezed resting state.
- Startup paths re-open the right sidebar. At a fixed 700px window that leaves
the pane ~70px, so the overlay had nowhere to go and the field measured 0px.
Settling these separately let whichever settled first drift back while the next
one ran. Re-assert them in one loop until they hold simultaneously, and size the
window from the chrome actually measured instead of assuming a fixed 700px.
Verified 8/8 green, and still fails at the overlay assertion when the fix is
disabled, so the regression coverage stays real.
---------
Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>
* Revert "fix(skills): stop reporting a failed update when the CLI succeeded (#11105)"
This reverts commit a8660839ee.
* fix(skills): stop offering an update the CLI provably cannot perform
The Update skills dialog reported "The update didn't finish" / "Some skills
could not be updated" with an armed Retry, directly above the runner's own
"All global skills are up to date" log line, on a clean exit 0.
`skills update` decides what to do by comparing its lock's `skillFolderHash`
against the source tree and never reads disk (published CLI, dist/cli.mjs:
`latestHash !== entry.skillFolderHash`). So once the lock records a revision
the filesystem does not actually have, it reports up-to-date, exits 0 and
writes nothing. No retry converges it.
Gate eligibility on that instead of blaming the run afterwards: a name stays
updatable only while the lock's hash and the revision the DISK hashes to
agree. Those that disagree fall through to `needs-attention`, which
skill-freshness-display-status.ts already documents as the state for a copy
"out of date somewhere the update command cannot reach". The dialog no longer
offers them, so it can neither claim failure nor claim success.
Deliberately compared against disk, not the bundled manifest: the updater
pulls from the source repo, which legitimately runs ahead of what a build
ships, and gating on the bundle would withhold real updates. Both sides must
also be positively identified — an unplaceable lock hash is not evidence.
Also reverts a8660839ee, which forgave `outdated` post-run. That turned the
false failure into a false success ("Updated 2 skills" over copies nothing
wrote) and let an empty verdict swallow spawnError, so an offline run
published green.
Reachable by anyone who updated during the stub-conversion window — every
bundled skill has a stub -> full -> stub oscillation in its last three
registry revisions.
* fix(skills): do not gate a skill when a placement is unidentifiable
`diskTreeShas` drops digests matching no known revision, so `every` ran over
only the resolved half — one stale copy beside one unidentifiable copy gated
the name, contradicting the unknown-stays-eligible rule the function documents.
Require every observed digest to resolve before treating all placements as
mismatched.
Caught by CodeRabbit on #11110.
* fix(skills): judge convergence only over placements the update command writes
A same-name plugin-cache or repo copy could defeat the gate two ways: an
unidentifiable repack read as an unresolved placement, and a cache copy
parked at the lock's own revision read as an anchor — both re-arming the
unwinnable update on a drifted canonical. Filter to
SUPPORTED_GLOBAL_SKILL_TOPOLOGIES, matching eligibility and outcome.
* fix(agent-status): track Codex rollout subagents
* fix(agent-status): resolve cross-day Codex child rollouts and unblock CI gate
Codex files each rollout under its own local start date, so a session that
runs past midnight spawns children into a sibling day directory. Scanning
only the parent's directory left 13% of real subagent spawns (48/371 across
local rollouts) permanently unresolved, which pinned a phantom "working" row
and re-ran readdirSync every poll tick forever. Resolve the child's own day
directory from occurred_at_ms, and time-box a child whose rollout stays
unreadable so a deleted or never-written file can't leak a working row.
Also make the hook HTTP handler return void: the changed-code quality gate
keys findings by span overlap, so this PR's added line inside the pre-existing
async createServer callback resurfaced no-misused-promises as a new finding.
Tests cover cross-day resolution, grace-period retirement, and that the poll
re-arms across successive roster changes (the prior tests passed even when
the poll died after its first change).
* fix(agent-status): keep the Codex subagent poll alive across nested hooks
A nested non-codex CLI inherits its parent's ORCA_PANE_KEY, so its hook
POST reached scheduleCodexSubagentPoll and tore the timer down before the
source guard, silently ending polling while a rollout child was still live.
* perf(terminal): eliminate dense control and frame gate regressions
* test(terminal): keep gate labels in valid expect shape
* test(terminal): expose the surviving sub-threshold control-density case
The only adverse strip fixture sat at 50% control density, which is exactly
where the fallback fires and wins. A shape at 31 controls per 64-unit block
evades the trigger and still loses to the per-character legacy (0.67x), so the
benchmark structurally could not show it.
Add that fixture, pin both density literals in the staleness guard so a retune
fails loudly instead of silently measuring a boundary that moved, and export
the probe constant the equivalence test was hardcoding.
The Update skills dialog showed "The update didn't finish" / "Some skills
could not be updated" with an armed Retry, directly above the runner's own
log line saying "All global skills are up to date" — on a clean exit 0.
`skills update` compares its lock's recorded hash against the source and
never reads disk (dist/cli.mjs: `latestHash !== entry.skillFolderHash`).
Once the lock has advanced past the installed bytes it prints up-to-date,
exits 0 and writes nothing. The copy left behind is a recognised older
revision, so the post-run re-scan sees `outdated`, skillUpdateFailedNames
counted that as a failed run, and skill-update-run settled to state 'error'.
Retry re-ran the same command, which no-op'd again — a closed loop.
Reclassify `outdated` as "the command did not converge this", not "the run
failed". The freshness badge still marks the copy not-current, so nothing is
hidden; the run just stops being blamed for it.
A botched write is still caught: a half-written bundle hashes to
`unrecognized`, a wholly-degraded or removed copy leaves no convergent
placement, and process-level failure still surfaces via the spawn error.
Reachable by anyone who updated during the stub conversion window — every
bundled skill has a stub -> full -> stub oscillation in its last three
registry revisions.
Add translations for OrchestrationPage coordinator and child PR names, plus agent workflow status messages (initial states and progress beats) across all supported languages (English, Spanish, Japanese, Korean, Simplified Chinese).
Jinjing
Brennan Benson
Neil
OrcaWinandOrcaWin
Jinwoo HongandOrcaWin
YoyoandNeil