Skip to content

[upstream #10035] fix(daemon): gate agent startup cwd after safe default fallback - #7

Closed
innocarpe wants to merge 89 commits into
mainfrom
fix/daemon-agent-startup-cwd-fallback
Closed

innocarpe wants to merge 89 commits into
mainfrom
fix/daemon-agent-startup-cwd-fallback

Conversation

@innocarpe

Copy link
Copy Markdown
Owner

Portfolio mirror of my contribution to upstream stablyai/orca.
Exhibition only — the real review/merge target is upstream.

Upstream

Summary

Description User-facing: Launching a recognized agent (e.g. OpenCode) through the daemon PTY path with no explicit workspace cwd no longer fails with “Automatic agent startup command … requires a non-root workspace working directory.” The spawn now uses the same s

Note

  • Do not merge this into innocarpe/orca main until the upstream PR is merged.
  • After upstream merges: sync fork from upstream, then close this mirror PR.
  • This open PR exists so visitors see in-flight work on this fork's Pull requests tab.

nwparker and others added 30 commits July 22, 2026 03:32
…idebar rows (stablyai#9850)

* fix(agent-status): keep Claude in-process teammates visible as idle sidebar rows

Claude Code 2.1.21x runs named Agent-tool agents as turn-based in-process
teammates: SubagentStop and TeammateIdle fire at every TURN end while the
teammate stays alive awaiting mail (verified live on 2.1.217). Treating
those events as finish signals deleted the child row seconds after each
burst, so the sidebar showed no subagents for most of a teammate's life.

Root-cause fix: the roster now tracks a working/idle state per child.

- One-shot children (hyphen-free ids) keep remove-on-stop: their
  SubagentStop is a true finish.
- Teammate-shaped rows park as idle on SubagentStop/TeammateIdle and
  revive to working via the next SubagentStart (same lifecycle id,
  first-observed startedAt preserved).
- Idle rows never gate the pane 'working' (stablyai#8825's done-gate rule).
- Only TeammateIdle-confirmed idle rows survive a complete lead-Stop
  fold; a stopped workflow lane wearing a teammate-shaped id is reaped
  there (or immediately, once a fold tagged it listedAsSubagentTask), so
  the pre-stablyai#8825 idle pile cannot rebuild.
- At the wire cap, the oldest idle row is evicted to admit a working
  spawn; working children are never displaced.
- Hydrate keeps pruning idle snapshots: idle-teammate liveness cannot be
  proven across a restart, and a live teammate re-earns its row.

* fix(agent-status): restore inventory-confirmed workflow lanes
* feat(skills): land remaining hybrid stubs

* fix(build): exclude skill stub sources from packages
…gration (stablyai#9798)

Opening the Quick Commands sheet right after connecting over relay races
the relay->direct cutover, which rejects the in-flight one-shot
settings.getTerminalQuickCommands with LogicalClientCutoverError while
connState stays 'connected'. The sheet then strands on "RPC interrupted
by connection migration" with an empty list until closed and reopened.

The read is side-effect-free, so replay it on cutover (capped at 5,
cancelled if the sheet closes or the client is replaced). Same failure
class and pattern as stablyai#9794 (capability probe) and stablyai#9796 (terminal
create).
…rch terms (stablyai#9967)

The Language setting's native word (语言 / 언어 / 言語 / Idioma) only reached
settings search via the localized title in that word's own UI locale — so a
Chinese speaker on the default English UI could not find it by typing 语言.

Always-index the native word for 'language' in every supported language (plus
the previously-omitted Spanish native name Español), so speakers can locate and
switch to their language from any starting locale. Native words are
locale-invariant constants, so they are plain keyword literals with reviewed
localization-coverage allowlist entries.

Co-authored-by: Orca <help@stably.ai>
stablyai#9861)

* fix(agent-status): map codex request_user_input questions to waiting

Codex 0.145 asks user questions via the auto-allowed request_user_input
tool (experimental default_mode_request_user_input): PreToolUse fires
while blocked on the answer with no Stop, so Orca showed the pane as
working/idle instead of Needs You. Map that PreToolUse to waiting
(mirrors grok's ask_user_question), exempt question waits from the codex
yolo auto-approval suppressor, and deliver native-chat answers to the
digit-commit selector by option number (typed labels are ignored and
Enter commits the highlighted first option). Older codex versions emit
no such event and are unchanged.

* fix(native-chat): preserve codex question answer semantics
…n mobile (stablyai#9801)

* fix(mobile): persist per-device tab selection so worktree return restores the last open tab

A phone's tab selection lived only in the host's in-memory
ClientSessionTabSelectionStore. Any host restart wiped it, and the
per-device projection then fell back to deterministic topology, so
returning to a worktree on mobile always landed on the first tab
instead of the tab last opened on the phone.

Persist the per-device selections in the Store (keyed deviceId ->
worktreeId), hydrate them when the runtime constructs, and guard
projection so an early empty snapshot after restart cannot wipe a
hydrated selection before tabs arrive. Selections are pruned with the
worktree/repo and on device revoke, and malformed persisted payloads
degrade to empty instead of throwing.

* fix(mobile): harden persisted tab selection cleanup

* fix(mobile): preserve tab selection across worktree rename
…ed (stablyai#9780)

* feat(mobile): mount ProtocolBlockScreen when protocol compat is blocked

ProtocolBlockScreen existed since PR stablyai#1440 but was never mounted: on a
'blocked' compat verdict the only output was a console.warn, so a future
MIN_COMPATIBLE_RUNTIME_CLIENT_VERSION bump would have silently shown a
broken host UI instead of the update screen.

Add HostProtocolGate — a choke point in app/h/_layout.tsx above every
/h/[hostId] route — that consumes useHostStatusGates and replaces the
blocked host's entire UI (sidebar + detail stack) with ProtocolBlockScreen.
The host list and other hosts stay usable; the screen's own 'Back to
hosts' escape hatch routes to '/'. Both block reasons render their
respective CTAs (mobile-too-old → App Store, desktop-too-old → GitHub
Releases). Compat logic stays in the src/shared mirror contract — no fork.

* fix(mobile): fence incompatible host routes efficiently

* fix(mobile): route Android updates to releases
* feat(source-control): show current branch in header

* fix(source-control): keep header focused on branch

* fix(source-control): compact detached head identity

* fix(source-control): make branch identity keyboard accessible

* fix(source-control): keep create review in checks
…erer OOM (stablyai#9872) (stablyai#9936)

The live `agentStatusByPaneKey` map had no size cap. `setAgentStatus` rewrites it
with a full spread copy on every status ping and keeps rows until a pane/tab
teardown event removes them; a missed teardown (agent killed without a Stop hook,
pane/tab closed while its status lingered) orphans a row forever. On long
multi-agent sessions orphaned heavy rows accumulate without bound, and because
each ping spread-copies the whole map, once it is ~1.9 GB one ping transiently
doubles it past the 3586 MB old-space limit -> renderer crash (exit -36861).

Cap the live map at MAX_LIVE_AGENT_STATUSES=500, shedding only rows whose pane is
provably gone (a mounted tab's rooted layout no longer lists the leaf) or long
idle, and never an open pane's row (any state, incl. needs-input waiting/blocked).
Rootless/empty-snapshot, not-yet-hydrated, and no-renderer-tab (orchestration
worker) rows are unprovable: kept while a fresh agent could own them, shed only
past the stale window or by a hard-cap fallback that guarantees the bound.
Eviction bumps the status/sort epochs so the retention sync snapshots disappeared
done rows. Cost is one Object.keys length check on the reducer's already-O(n)
spread under the cap; the layout walk + sweeps run only on the rare over-cap ping.

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
…ablyai#9782)

* feat(mobile-pairing): surface unpaired-device auth failures instead of silent 4001 loops

Desktop: when a phone repeatedly fails direct-transport E2EE auth with a
token missing from the device registry (pre-v1.4.106 pairing-path bug left
desktops that regenerated their registry rejecting paired phones forever),
throttle to one notification per session and show an actionable toast
pointing at Settings -> Mobile to re-pair.

Mobile: map a bare 4001 close onto the existing auth retry budget (the
encrypted e2ee_error is undecryptable when the desktop keypair changed, so
the close code is the only surviving signal) instead of looping the generic
reconnect forever, and make the auth-failed verdict say 'Pairing invalid -
re-pair with your desktop' instead of a bare 'Auth failed'.

* fix(mobile-pairing): handle stale keys and startup notification races

* fix(mobile-pairing): isolate auth notification failures

* fix(mobile-pairing): keep recovery alert actionable
* perf(mobile): gate host polling on foreground

The mobile host screen ran two 3s polls (routed + embedded), each firing worktree.ps
AND repo.list, with no foreground/background gate — so a connected phone kept pinging
every 3s (worktree.ps is a full multi-repo process scan) plus a radio wakeup, including
brief background windows while the socket stays parked.

Consolidate both into one startHostWorktreeRefresh lifecycle and AppState-gate the
interval so BOTH polls stop while backgrounded and refresh immediately on foreground
return. worktree.ps keeps its 3s cadence while foregrounded (it carries live agent
status/preview/unread that no push event replaces). repo.list stays on the interval as
an AppState-gated, self-throttling (REPO_METADATA_REFRESH_MS=60s) convergence safety-net
— desktop Settings repo edits notify only the renderer, not the runtime clientEvents
stream, so it can't be made purely event-driven without going stale — and additionally
gets a reposChanged/worktreesChanged fast-path and reconnect-replay refetch.

Verified in a deps-installed mobile checkout: full mobile suite 2232 pass, typecheck,
oxlint (within the frozen max-lines budget), and oxfmt --check all clean.

Co-authored-by: Orca <help@stably.ai>

* chore(mobile): drop stale fetchRepoMetadata dep from the reconnect effect

Address CodeRabbit nitpick: the reconnect effect no longer calls fetchRepoMetadata
(that refetch moved into startHostWorktreeRefresh), so it shouldn't remain in the
effect's dependency array.

Co-authored-by: Orca <help@stably.ai>

---------

Co-authored-by: Orca <help@stably.ai>
stablyai#9979)

* fix(mobile): keep quick-commands button steady while capabilities load

The tab-row quick-commands button only rendered once the capability probe
resolved true, so it popped in after the row was already visible (and
vanished during reconnect re-probes). Render it whenever support is not
confirmed absent and disable it until the probe settles — pre-quick-commands
hosts strip agentPrompt, so the action (not the button) must wait for
confirmation. Confirmed-unsupported hosts still hide it entirely.

* fix(mobile): explain unsupported quick commands on tap instead of hiding

Per feedback on the disabled/hidden states: the button now always renders
and stays tappable. Tapping against a desktop that confirmed no support
shows "Desktop update required for quick commands" (mirroring the browser
streaming copy); tapping while the capability probe is still resolving says
to try again in a moment. The sheet still opens only once support is
confirmed, since pre-quick-commands hosts strip agentPrompt.

* docs(pr): add QA screenshots for quick-commands button states

* test(mobile): lock quick-commands button stability

Add a focused source-contract test for the always-mounted tab action and confirmed-support sheet gate. Keep the non-obvious safety comment concise, and remove PR screenshots now hosted as GitHub user attachments.

* test(mobile): structurally guard quick-command action mount
…9843)

react-markdown's <Markdown> has no internal memoization: it rebuilds the whole
unified remark->rehype->highlight->katex processor and re-parses the document on
every render. MarkdownPreview re-renders on internal state that does not affect
the rendered output — most visibly, every keystroke in Find (query/match-index
state) — so a large doc re-ran the full parse + syntax-highlight + KaTeX pass per
keypress, making Find laggy.

Hoist the two fully-static plugin arrays to module scope (a fresh array identity
per render would defeat the memo) and render the body through a React.memo'd
MarkdownBody keyed on content + components. The pipeline now re-runs only when the
rendered content or the components map actually changes; Find/review-pulse/copied-
note re-renders skip it. The components map was already memoized, so its identity
is stable across those re-renders.

Behavior unchanged: 106 existing MarkdownPreview tests pass.

Co-authored-by: Orca <help@stably.ai>
…i#9842)

The iOS MJPEG and Android scrcpy device streams are gated only on the pane
being the active tab (isActive, PR stablyai#7382). When the emulator tab is frontmost
but the whole Orca window is hidden/minimized/occluded/display-asleep, the
full-fps pipeline keeps running: main-process socket read + JPEG/H.264 decode
+ IPC + renderer decode. Renderer background-throttling (stablyai#9395) cannot stop it
because the pipeline is IPC-push driven from main.

Gate showStream additionally on window visibility via a new occlusion-safe
hook that honors the terminal stale-visibility latch (so a display-sleep
occlusion wedge can't freeze the emulator on a black frame) and delays the
visible->hidden park by 500ms so a quick Cmd+Tab round-trip doesn't renegotiate
the device stream.

Co-authored-by: Orca <help@stably.ai>
…#9995)

Zone.js patches the global Promise with a non-native thenable. When a bare
`new Promise(...)` crosses the Electron executeJavaScript boundary, it's
serialized as-is, losing { page, target } and exposing __zone_symbol__*
fields instead. Wrap in an async IIFE to return a native promise that
Electron always unwraps correctly.
…blyai#9985)

Production crash diagnostics measured ~128 `git worktree list` execs/min
(9,400 in one 80-minute session, ~16% of wall-clock in git subprocesses):
the resolved-worktree scan fans out over every registered repo on a 30s
cache TTL, and most registered repos on the affected installs were
agent-CLI scratch repos (~/.codex-tmp capsules, vendor imports, skill
checkouts) that need no freshness.

Classify agent-scratch repo roots with a curated shared matcher and stamp
their scan-cache entries with a 5-minute TTL instead of 30s. Orca-driven
mutations still bypass the TTL via the per-repo generation bump, so only
passive pickup of external changes slows for scratch repos. Expected
steady-state reduction on the measured install: ~82% fewer git spawns.
…7936) (stablyai#9826)

* fix(daemon): retire macOS daemons whose login session died (stablyai#7936)

A daemon that survives a full macOS logout is unsalvageable: its PAM
context can no longer host login(1) spawns (every new PTY becomes a
'Login incorrect' prompt zombie) and its Mach bootstrap namespace has
lost the system DNS resolver, so terminals it hosts have no egress.
Today it also keeps the stablyai#9301 preflight's cached 'accepted' verdict, so
it keeps wrapping spawns in login(1) forever; only a manual daemon
restart recovers.

GUI-spawned daemons now watch for login-session death from the inside:
a fresh cache-bypassing PAM probe (triggered by PTY-exit bursts, fresh
client hellos, and a slow periodic timer) must conclusively reject
three consecutive times AND the in-process system resolver must be
degraded; then the daemon exits crash-style so session meta stays
unclean and the replacement daemon cold-restores scrollback. A
conclusive rejection also flips the spawn-wrapper cache off
immediately.

Headless serve/SSH daemons never get the watch (they must survive their
spawning session ending), and a session that never conclusively
accepted login(1) never arms it — a PAM anomaly alone can't kill a
healthy daemon (fast user switching keeps accepting, so switched-away
sessions are preserved).

* test(daemon): e2e seam to drive login-session death oracles from a verdict file

A dead macOS login session cannot be fabricated without root (PAM owns
audit-session teardown), so live lifecycle QA drives the death watch's
probe and resolver oracles from ORCA_E2E_LOGIN_SESSION_PROBE_FILE:
'alive' → accepted/healthy, 'dead' → rejected/unhealthy, anything else
inconclusive — with compressed watch timing. Mirrors the existing
ORCA_E2E_DAEMON_INIT_DELAY_MS seam; inert unless the env var is set.

* fix(daemon): close the hang-shaped gap in login-session death detection

The conclusive-PAM-verdict trigger had one blind failure shape: login(1)
hanging at the prompt past the probe bound (killed → inconclusive
forever → the watch never fires). Three changes close it:

- The death-watch probe gets its own 4s bound (the 500ms preflight bound
  exists for spawn-path latency, which doesn't apply off-path), so a
  slow-but-answering PAM stack isn't misread as a hang.
- An inconclusive pipe probe escalates to a PTY-hosted probe via
  script(1) — a dead session's PAM stack may only misbehave under a real
  tty (the pipe-vs-PTY fidelity limit the preflight documents).
- A streak of timeout-killed probes (which a live session never
  produces) is a second retirement trigger, at a higher threshold (5)
  and still gated on the degraded resolver, logged with a distinct
  cause so field logs discriminate the two paths.

Every dead-session behavior — fast reject, prompt-then-EOF, or hang —
now fires retirement; all inconclusive states still fail toward
preserving the daemon.

* fix(daemon): keep login-session retirement conclusive

* fix(daemon): stop login watch before clean shutdown

* fix(daemon): make login-session PTY probe reliable

* fix(daemon): close login-session watch races

* fix(daemon): ignore health probes for login watch activity
…ai#9980)

* test(e2e): verify Claude is prefilled with issue URL on start

Regression test for stablyai#6613: when starting a workspace from a newly
created GitHub issue, ensure the issue URL is passed to Claude via
`--prefill` and `--dangerously-skip-permissions` flags. This prevents
context loss after issue creation.

* test(e2e): fix GitHub-issue-start prefill test flakiness

- Reorder mock API handlers to ensure `/labels` and `/assignees` paths match before the specific issue endpoint
- Replace regex heading matcher with exact string for more reliable assertions
- Refactor terminal content polling to capture text once and reuse in subsequent assertions
* fix(terminal): defer remote output ACKs until parse

* test(terminal): document synchronous credit claims

---------

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
AmethystLiang and others added 22 commits July 22, 2026 21:30
…blyai#10007) (stablyai#10029)

* Revert "Enable accessibility tree (`ax`) command on iOS emulator sessions (stablyai#10007)"

This reverts commit 43ae014.

* fix(emulator): expose iOS accessibility tree

* fix(emulator): support device-only iOS AX

* fix(emulator): normalize iOS ax to 0..1 and heal missing axUrl

serve-sim's helper /ax reports element frames in absolute pixels, but
tap/gesture take normalized 0..1 coords. Normalize the raw AX node tree
into a compact nested shape whose frames are 0..1 over the device screen
(first root's frame), mirroring serve-sim's own normalizeAxTree, so agents
can feed ax output straight back into input commands.

Also heal sessions that were registered without an axUrl: stablyai#9924 only
derived /ax at parse time, so already-active sessions had no endpoint.
The bridge now derives it from the session's mjpeg stream URL, guarded to
the /stream.mjpeg suffix so a non-mjpeg URL never fabricates a bogus /ax.

* docs(emulator): mark ax working on iOS with correct raw-AX-tree shape

Both skill guides and the CLI summary described iOS ax as unsupported (or,
via the reverted stablyai#10007, as a normalized "screen + elements" shape that
never matched the endpoint). ax works on both backends: Android via
uiautomator, iOS via the serve-sim helper. Document the real iOS output —
a raw AX node tree (labels, roles, nested children) with frames normalized
to 0..1 — and regenerate the bundled skill guides.

* chore(skills): regenerate skill bundle manifests

CI verify failed because generated skill artifacts were stale after version/skill revision bumps.

* fix(emulator): read ax from explicit device without active session

Fall back to udid-keyed session lookup when a worktree has no active emulator,
allowing `--device` targeting to work the same way for ax as it does for tap/type.
Also clarify in docs that AX frames are normalized 0..1 with top-left origin,
and show how to tap an element at its frame center (x+width/2, y+height/2).

* fix(emulator): cap iOS AX tree at 500 nodes

Unbounded accessibility trees can flood agent output. Enforce a 500-node limit (matching serve-sim's snapshot cap) and mark truncated parents so consumers know the tree was cut.

---------

Co-authored-by: 5Hyeons <ohs2251@naver.com>
…grid (stablyai#9997)

* feat(dashboard-popout): size the terminal dialog's PTY to the dialog grid

The popout agent-terminal dialog rendered the pane's serialized frame at its
original cols/rows and CSS-scaled it down to fit. The dialog now claims the
PTY grid for its own box through the remote-desktop viewer registry: the PTY
reflows to the dialog's dimensions (crisp, unscaled), the main-window pane
parks at the held grid like any remote viewer hold, and closing the dialog
releases the claim so the pane reclaims its geometry. A phone-driven PTY
keeps the floor; the dialog then falls back to the scale-to-fit rendering.

Any grid change under a live preview stream (fit landing, host reclaim,
phone takeover) now pushes a resync so the dialog repaints from a snapshot
at the new grid — this also fixes garbled dialogs when the pane resized
while a dialog was open.

* fix(dashboard-popout): harden terminal grid claims

* perf(dashboard-popout): bound terminal preview resize work
* fix(terminal): resume output after recovery resync

* fix(terminal): harden recovery resync boundaries

* fix(terminal): prevent double rendering on frame-drop resync

After an untagged resync snapshot, the client resets to the snapshot's output
high-water. Buffered output generated while serializing is already included in
the snapshot, so replaying the entire buffer afterward renders those bytes
twice. Trim to output after the snapshot seq.

Tagged snapshots feed side consumers and need all buffered output unchanged.
Add exponential backoff for truncated recovery retries to avoid stampeding
flooded servers.
…stablyai#10011)

Adds guarded host-authoritative orphan PTY adoption and keeps Active Server durable preference mutations exclusive to its explicit settings control.
Credits remote terminal output only after parse or intentional discard, with bounded adaptive windows, fair draining, recovery cleanup, and RTT/parser benchmarks.
…blyai#10110)

electron-builder 26 dropped the app-builder-bin Go binary, so the
signed-installer staging step failed with 'node_modules/app-builder-bin/
win/x64/app-builder.exe is not recognized'. Blockmap generation now lives
in app-builder-lib's pure-JS buildBlockMap; call it through a small script
in both the release-cut and signing-rehearsal workflows.

Co-authored-by: Orca <help@stably.ai>
stablyai#10115)

- orca-runtime: 'stop cannot be proven' now requires the fresh PTY inventory
  to still show the pty live (empty inventory proves exit post-stablyai#10106).
- git-common-watch: swap worktrees dir->file synchronously so no poll tick
  observes the transient ENOENT window and fabricates a delete.


(cherry picked from commit e11736d7e170982eeebe09256eb549446a4f4f7b)

Co-authored-by: Orca <help@stably.ai>
stablyai#10120)

- Tighten combobox row padding (py-2 -> py-1.5) in the Project and Run-on
  lists so items aren't so vertically loose.
- Use a Monitor glyph (not a server) for the local machine in Run-on.
- Pin 'Add host' as a compact single-row footer with one clean divider,
  matching the Project picker's 'Add a new project' footer (removes the
  taller two-line in-list row and its extra separator).
- Give the smart name field a solid bg-background so it matches the other
  inputs on light mode instead of reading translucent over the dialog.

Co-authored-by: Orca <help@stably.ai>
…blyai#10023)

Replace the api.onorca.dev fallback pattern with same-host retries on
www.onorca.dev, the endpoint that owns crash delivery. This ensures
consistent behavior and clearer error reporting when failures occur.
All earlier WeChat groups are full; point the README QR and copy at group 5.
…yai#9627)

* fix(quick-open): guide rg install after fallback budget errors

* fix(quick-open): show local host wording for local install guidance

The install-rg guidance component hardcoded 'on the remote', so the new
local fallback path told local users to install ripgrep 'on the remote'
— wrong for the exact case stablyai#9627 targets. Parse the location out of the
message and render the matching wording; add the local locale string and
a render test that guards against the 'on the remote' regression. Also
harden the reason capture against a stray ')' in the error text.

Co-authored-by: Orca <help@stably.ai>

---------

Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: Orca <help@stably.ai>
* fix(terminal): restore a dark-background contrast floor

Fully disabling xterm minimumContrastRatio on dark backgrounds (stablyai#9599)
left near-background body text unreadable — Antigravity paints #262b30
on #1e242a (~1.1:1). Keep light backgrounds at WCAG-AA 4.5 and use a
milder dark floor (3) so dark-on-dark body text is lifted without the
full light-bg correction strength.

Fixes stablyai#10104

* fix(terminal): extend dark-bg contrast floor to preview + mobile terminals

The dark-background minimumContrastRatio floor (stablyai#10104) is applied per
`new Terminal()` construction site. Beyond the live pane, agent output also
renders in the dashboard popout preview and the mobile WebView, which were
still at the floor-1 default, so Antigravity output stayed unreadable there.

- AgentTerminalPreview: gate via resolveTerminalMinimumContrastRatio
- mobile WebView: port the gate as resolveTerminalContrastFloor (Chrome-74 JS)
- tests: builtin-catalog guard + mobile vm-harness coverage

Co-authored-by: Orca <help@stably.ai>

---------

Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: Orca <help@stably.ai>
…etion (stablyai#9263)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
…ad of failing them closed (stablyai#10151)

Inline setup/onboarding terminals (skill installs, feature tips) and the floating terminal use host-agnostic synthetic worktree ids with no worktree/repo row. Since stablyai#9994, connectPanePty gated their transport on resolveWorktreeOperationRouteResult, which fails closed ('missing') for unknown ids — so the Orca CLI skill Update button, every other inline setup terminal, and the floating terminal showed "Workspace identity is ambiguous across hosts" instead of running.

Route them through the shared resolveTerminalWorktreeRoute (which already exempts the floating terminal and folder workspaces), and add the missing ephemeral-setup exemption so setup terminals follow the single active runtime (remote skill installs land there) or run locally when none is focused. Genuinely unknown/stale repo-backed worktrees still fail closed, preserving stablyai#9994.

Adds terminal-worktree-route unit tests and connectPanePty regression tests (proven to fail without the fix).

Co-authored-by: Orca <help@stably.ai>
createPtySubprocess asserted assertSafeAgentStartupCwd on the raw
opts.cwd before applying opts.cwd || getDefaultCwd(). An omitted cwd
was treated as root-like and threw even though the post-fallback home
default is safe — while LocalPtyProvider already gates the effective
cwd after fallback.

Compute requestedCwd first, then gate. Update the daemon unit test to
assert omitted-cwd agent launch uses the safe default (stablyai#9578).
@nwparker
nwparker force-pushed the fix/daemon-agent-startup-cwd-fallback branch from b02027d to f2134eb Compare July 23, 2026 11:49
OrcaWin pushed a commit that referenced this pull request Jul 29, 2026
…(C1) (stablyai#10625)

* fix(terminal): park SSH worktrees like local ones (C1 retention, slice A)

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(terminal): bound hidden-worktree retention with a force-park budget (C1, slice B)

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>

* fix(terminal): demote hidden scrollback for eviction-exempt worktrees (C1, slice C)

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>

* fix(terminal): paint the SSH model snapshot inline, not via nested coordinator (C1 slice A fix)

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>

* test(terminal): SSH park+reveal e2e round-trip + as-built design notes (C1)

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>

* fix(terminal): paint SSH reveal from main's model even when the relay replay is empty (C1 review #1)

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>

* fix(terminal): decouple scrollback demotion (slice C) from the retention-budget switch (C1 review #2)

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>

* fix(terminal): scope eviction exemption to the tab, not the worktree (C1 review #3)

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>

* fix(terminal): demote un-parkable worktrees the force-park lever spared (C1 review #4)

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>

* fix(terminal): keep the hidden clock running through transient background-measure windows (C1 review #5)

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>

* test(terminal): make the SSH park+reveal depth assertion prove the model paint (C1 review #6a)

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>

* docs(terminal): rewrite DESIGN.md as the single as-built C1 contract (review #7)

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>

* fix(terminal): resolve the eviction exemption per pane, not per tab (C1 review #8)

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.

* docs(terminal): drop DESIGN.md — the as-built C1 contract moves to the PR body

Co-authored-by: Orca <help@stably.ai>

* fix(terminal): cap the deferred PTY side-effect queue (C1 residual H2)

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>

* fix(terminal): carry command-lifecycle facts through parked watchers (C1 follow-up)

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>

* test(terminal): retention-budget force-park e2e with a retentionLimit override (C1 6b)

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>

* test(terminal): retention-budget e2e via same-repo remote worktrees (passes docker lane)

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>

* fix(terminal): prevent remount thrashing during post-measure cool-down (

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.

* fix(terminal): detect Command Code completion in parked mid-turn panes

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.

* fix(terminal): flush pending command-code settles on reveal remount

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.

* fix(terminal): remove scrollback demotion (C1 slice C)

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.

* test(terminal): assert bounded probe during stalled reveal

Add assertion to verify that a stalled reveal operation makes exactly one
`getMainBufferSnapshot` call, ensuring retry logic doesn't introduce
redundant probes that would extend the timeout window before relay fallback.

* fix(terminal): implement C1 retention budget for hidden parked worktrees

Addresses OOM regressions in hidden parked terminals by force-evicting
worktrees past a retention budget: at most 12 mounted while hidden, none
past 45 minutes (absolute, not exempted by last-active). Eviction is
least-recently-hidden-first. Exempt tabs (unrestorable local PTYs) keep
their panes to avoid orphaning shells; worktrees are force-parked even
if they contain exempts, and their buffers released elsewhere. SSH/remote
worktrees serialize buffers pre-eviction for reveal; local worktrees keep
daemon snapshots. Command Code's done-settle window is transferred across
park/reveal boundaries so the row cannot strand at 'working'. Model probe
on SSH reattach is scoped to park-reveal only, not ordinary reconnects.
Includes new E2E suite proving the budget actually releases memory.

* memoize eviction-exempt terminal tabs to avoid redundant store reads

Each tab's exemption check re-reads the store and walks the layout tree.
Introduce selectEvictionExemptTerminalTabIds() to resolve all exempt tabs
for a worktree in a single pass, then memoize the result in Terminal.tsx
and useTerminalTabColdParking. This prevents O(n) store reads when checking
exemptions across multiple tabs and ensures the set remains stable across
unrelated re-renders.

* refactor: reformat hidden-worktree retention comments

Reflow to 80-character lines and remove internal ticket references
(C1, C1 slice C).

* fix(lint): split overlay slot and eviction-exempt tabs under max-lines

Static analysis failed because TerminalPaneOverlayLayer (401) and
terminal-parked-tab-watchers (304) exceeded oxlint max-lines. Extract the
slot component and eviction-exempt helpers into dedicated modules.

* test(terminal): stabilize retention budget e2e control arm

Stage un-parkable remote pty ids only after both worktrees are hidden, and
keep re-staging during the control-arm poll so a late updateTabPtyId cannot
flip the decoy back to park-restorable and ordinary-park it before budget
engages.

* test(terminal): pin retention e2e decoy to a mounted pane snapshot

Use the active pane-identity snapshot for the decoy tab instead of all
worktree tabs, and re-assert un-parkable ids after the control-arm hold so
a deferred/empty tab id cannot fail the budget-off mounted-count check.

* fix: memoize terminal eviction exemptions on layout leaf PTYs

Splits add leaf panes to the layout store without changing the tabs
array. A memo keyed only on tabs misses this change, leaving new panes
unexempted for unmount. Include layout leaf PTYs in the exemption memo
key so it recalculates when splits occur or PTYs are re-minted.

---------

Co-authored-by: Orca <help@stably.ai>
@innocarpe

Copy link
Copy Markdown
Owner Author

Upstream stablyai#10035 merged on stablyai/orca — closing portfolio mirror.

@innocarpe innocarpe closed this Jul 29, 2026
nwparker added a commit that referenced this pull request Oct 19, 2026
…iew off the deprecated owner resolver (stablyai#30562)

* refactor(renderer): fold browser creation, file actions and doc-preview callers off the deprecated owner resolver

The creation-action policy and browser creation (#3, #48) fold together, so the provider the
policy offers is the one creation uses. Browser panes, page lifecycle and conversion, tab
duplication, the explorer tree pane, doc-preview actions and editor path moves (#21, #43, #45-#47,
#49, #50) read runtimeEnvironmentIdWithLegacyFallbackUnlessTwin. File-explorer writes on a folder
(#51) take the owner from the rows alone, never from the focused server. Deprecated importers
16 -> 6.

* refactor(renderer): fold file preview off the deprecated owner resolver; Open in Orca Browser refuses a twin

A project two hosts publish reads as no server for the doc-preview plan, but the file:// path now
refuses it with its own message instead of opening this machine's copy (#7). Deprecated importers
6 -> 5.

* fix(renderer): a twin's file preview and open-externally refuse instead of opening this machine's copy

Codex P2 on the F7 folds: a folder twin read as no server, so the preview plan built a local
file:// tab and open-externally handed the path to this machine's OS. Both now check the owner
for ambiguity first. Also fix two test fixtures that broke the renderer typecheck.

* refactor(renderer): move the file:// open target out of file-preview.ts

file-preview.ts crossed its line limit after main grew it; the file:// target, its type and its two
refusal messages move to workspace-file-browser-open-target.ts, which they never depended on
file-preview for.

* test: mock the replacement resolver in the quick-action availability suite (Codex P2)
innocarpe pushed a commit that referenced this pull request Oct 21, 2026
…the same path (stablyai#31055)

* fix(runtime): project registration dedupes only against rows on this machine

Codex review 6 #7: create() matched paths across hosts, so creating a local
project at a path an SSH project used returned the SSH row and made no
directory. add() with no host had the same gap. One predicate,
runtimeRepoIsRegistrationCandidate, now backs add, create and clone.

* test(runtime): cover nested-SSH rows and clone reuse in host-qualified dedupe

Review follow-up (Opus P3s): runtime:<env> rows carrying a nested SSH
connectionId, add() for a runtime host, the clone controller, and a stricter
create() assertion.

* test(runtime): clone dedupe fixture commits without signing
innocarpe pushed a commit that referenced this pull request Oct 21, 2026
…p ones (stablyai#31096)

* feat(ci): ratchet raw repo connectionId host reads; one drop-in resolver

Codex review 6 class fix (#4/#6/#7): host decisions read off a repo row's
legacy connectionId miss rows that name SSH only as executionHostId.

- getRepoRoutingConnectionId (shared/execution-host.ts): drop-in for a raw
  repo.connectionId routing check that also finds executionHostId-only SSH.
- check-repo-host-read-ratchet: per-file count of raw repo connectionId
  reads in src/main, src/shared, src/cli; counts may only shrink.
- Migrated 66 cheap reads: runtime GitHub/GitLab commands, PR/MR head ref
  fetches, preserved-branch cleanup, ref queries, base-ref IPC, worktree
  preparation, trash sweep, scan cache TTL, usage metadata.

* fix(ipc): PR base IPC routes executionHostId-only SSH rows; ratchet is a stated lower bound

Review follow-up (Opus on stablyai#31096):
- worktrees:resolvePrBase decided git/gh routing on the raw connectionId while
  its head-ref fetches already used the resolver; one resolved connectionId
  now drives all of it (test fails on main).
- getRepoRoutingConnectionId's contract is stated exactly: never local when the
  raw field is set; on a contradictory ssh row executionHostId wins.
- The ratchet regex no longer matches report.connectionId; header and docs
  say it is a lower bound.
- Regression tests for a runtime GitHub and GitLab command site.

* test(ci): pin the repo host-read ratchet step to the static phase

* fix(ci): repo host-read ratchet skips verbatim field copies into projected rows

A copy of connectionId into a projected row decides nothing; migrating it to
the resolver would change what the projection publishes. The sidebar feed
catalog's copy (added after the baseline) and two older projections no longer
count; baseline pruned to 91.
nwparker pushed a commit that referenced this pull request Oct 23, 2026
…ds a terminal asks main's layout (stablyai#32002)

* refactor(runtime): every reader of a terminal's pane goes through paneOfPty (L80)

With the layout-owner switch on, main's layout answers which tab and pane hold a terminal, and
nothing reads or writes the PTY record's tabId/paneKey copies or their graph stamp. Switch-off
reads the copies exactly as before.

* refactor(runtime): read a terminal's pane through the runtime's paneOfPty method

* fix(runtime): main's pane index answers paneOfPty before any window graph

A headless host never installs a window graph, so the query asked a publisher that did not exist
and every terminal read as paneless. The index is now made on first query; it publishes only once a
graph is installed, as before.

* test(runtime): pin that a query with no window publishes nothing

* test(runtime): a terminal summary reads its pane once (L80)

* feat(runtime): pin bind-first at registration, and one own-handle rule with or without a window (switch-on only)

* test(runtime): census and tests that every create path binds its pane before it registers (L80)

* test(runtime): a bind-first fixture that stands in for the spawn commit (L80)

* test(runtime): fixtures bind a terminal's pane before registering it, as the spawn commit does (L80)

Also, with the switch on: skip the bind-first check for a paired server's workspace, release a
closed tab's terminal ownership once main has dropped the pane, and publish a pane's agent status
from the terminal main's layout binds it to.

* feat(runtime): an exit after main's close still finds the pane it retires (switch-on only, L80 Q2)

Main's layout write records every pane it drops whose terminal no pane holds any more, in a new
optional session field closedTerminalPanesByPaneKey (same TTL and cap as closed-tab records).
Exit cleanup reads main's layout first, then that record by terminal id and incarnation, and
removes the entry it used. Older builds ignore the field.

* refactor(layout): one dropped-leaf step every live layout edit runs (switch-on only, L80 Q2)

The step receives each dropped pane (host, worktree, tab, leaf, terminal, incarnation), computed
across the whole layout so a moved pane is not dropped, and runs only for a commit that declares a
live edit. Today it records the closed-pane entries; a drop is recorded once per terminal.

* fix(runtime): a phone or headless tab close still stops and releases its live terminals after main's layout drops the panes (switch-on only, L80)

The headless close reads each row's live terminal before main's close drops the panes it is found
by, and a renderer-retired tab releases the owned terminal main left in no pane, as the explicit
close already did. Close tests now commit through main's layout and bind panes first.

* fix(runtime): a local workspace's terminal tab left in a paired server's partition is main's pane (switch-on only, L257)

Older builds could save a local worktree's tab rows into a runtime partition (stablyai#18202); its terminals
are local, so main's index takes those tabs for workspaces the router homes locally. A tab or
terminal the local partition also holds is indexed once, from there; SSH and server-owned
workspaces are skipped as before.

* test(layout): the legacy-close comparison leaves out the closed-pane record only layout-owner builds write (L80)

* test(pty): a pane whose failed start withdrew its tab resolves to terminal_not_found in layout-owner builds (L80 #7)

Main's layout no longer holds the pane, so there is no pane to answer for; window-owned builds keep
the disconnected stub from the exited terminal's record.

* test(runtime): orca-runtime suites bind panes first, as the real store and spawn commit do (L80)

Fake PTY spawns commit the binding into the pane the runtime wrote, stores list their partitions,
and SSH sessions sit in their ssh:<conn> partition; terminals registered directly bind first.

* fix(runtime): main's layout names a pane's terminal only when the workspace and stamped incarnation match (switch-on only, L80)

A record made before its adoption registers (a host restart) or a local id reused after one
matched a saved binding by id alone, so a phone or paired desktop could be handed a second handle
for the pane before the terminal's own handle arrived and stay on it.

* test(runtime): a restored pane is only unnamed in the adoption gap, and no local record rescues a paired server's tab (L80)

* test(runtime): recovery, close and routing suites set up terminals as production does (L80)

A recoverable pane is one the host inventory dropped with no exit, live terminals register
rather than appear only in the window graph, and SSH stores route as the real Store does.

* test(runtime): suspects 3, 4 and 5 bind, register and apply as production does; PR-1's main-layout closes in the part-10 and part-11 close tests (L80)

* test(runtime): a fixture store applies main's layout commands as a profile does; PR-1's own closes in the part-05, part-08 and UUID-leaf close tests (L80)

* test(runtime): the profile-store stand-in moves to its own fixture, keeping bind-first under the line limit (L80)

* test(runtime): orphan adoption reads the pane from main's layout with the switch on, not the record copy (L80)

* test(runtime): a phone's pane close and a pinned tab close are main's layout closes with the switch on (stablyai#31571), in the no-widening and surface-close tables (L80)

* test(runtime): a paired phone close commits in main's layout with no relay ack when the switch is on (stablyai#31571) (L80)

* fix(terminal): a parked pane resumes its agent once main has closed the other pane that held the session (layout-owner build, L80 P1)

* test(runtime): the parked-pane suite's restart holder is a saved pane with no registered terminal, under the line limit (L80 P1)

* fix(terminal): a spawn commit that binds a reused local terminal id unbinds it from the saved pane of the older process (layout-owner build, L80 P2)

* test(runtime): the legacy OSC row test uses the UUID leaf main's layout holds and registers its live terminal with the switch on (L80 E5)

* test(runtime): the pty-exit echo close of a dead headless tab is expected to fail with the switch on until W4's close PR (L255)

* fix(terminal): a terminal record that arrives, changes incarnation or moves workspace after main composed its panes re-composes them, so a restored pane lists once (layout-owner build, L80)

* fix(terminal): a pane bound to a terminal a paired server runs keeps that binding and host, and is never indexed as a local spilled tab (layout-owner build, L80/L257)

* fix(terminal): a pane main owns no longer proves its terminal runs; the host's inventory and the terminal's record decide, so a dropped shell reads disconnected (layout-owner build, L80)

* fix(terminal): an SSH pane whose host says its shell ended keeps its leaf on a negative or failed code, as the window does, so recovery starts a shell in it; a stale handle learns the connected replacement (layout-owner build, L80)

* fix(agent-launch): a terminal main's layout still binds counts as running through a host listing gap, so a launch never starts a second shell in its pane (layout-owner build, L80)

* fix(terminal): a disconnected terminal main still binds stays a stop target, never reads stopped unproven, and keeps an exit wait waiting until a certified exit (layout-owner build, L80)

* test(terminal): the wait unit fixture answers the exit-settled dependency

* test(terminal): realistic fixtures for three phone-row tests: unique leaf ids per pane, and a window that draws main's own panes (layout-owner build, L80)

* fix(terminal): a stale handle learns a pane's replacement only when it was resolved for that pane, its own terminal ended, and the replacement is connected there now (layout-owner build, L80)

* test(session-tabs): switch-on variant for a window's chat placement (main's layout order rules); headless phone tab order is L283

* test(terminal-close): switch-on close reds name their ledger rows (L255, L256, L259a, L259b); a same-id respawn's commit stamps its incarnation; the crossing terminal binds first

* test(terminal): the crossing registration is refused by main's workspace fence (L256 switch-on variant); the placement test's tabs are main's saved layout

* refactor(terminal): the stale-handle guard reads an issue-time pane stamp on the handle record, not a second map; moved-pane, restart and single-reader tests (layout-owner build, L80)

* test(session-tabs): switch-on variants: the desktop's tab name is main's rename, and a mobile-created tab's live shell is the one its commit registers (layout-owner build, L80)

* fix(terminal): a re-issued handle keeps its issue-time pane while no pane holds its terminal (review)

* fix(terminal): stop targets and exit waits use main's holds check (workspace and incarnation), and any exit event settles a later wait (review)

* test(session-tabs): name the switch-on placement and split-group reds L288 and L289

* fix(terminal): an exit drops a closed pane's record in the same write that retires the pane, and an unbind folds into the binding write (layout-owner build, L80)

The exit used to remove main's closed-pane entry in a session write of its own, before the
retirement write. It now reads the entry, and the retirement write for that pane drops it,
re-checked by terminal id and incarnation. An exit that keeps the surface (relay contact lost,
intentional stop) keeps the entry until a confirmed exit or its TTL.

A reused terminal id's unbind is now a pure computation that applyPtyBinding applies through
its existing writes. Two split-pane tests still expected a -1 exit to close the pane; they now
follow the ruling that any negative code keeps it. Static-gate fixes: test fixtures and tests
use typed member access instead of Reflect, and no longer use type assertions.

* test(runtime): the shared fixture store announces its session writes and the L283 store lists its partitions, as a profile Store does; L289's test passes on that store (layout-owner build, L80)

The fixture store bindPaneFirst fills now tells write listeners of every session write, direct or
by a layout command, so main's post-commit merge runs as it does on a real Store. The notice is
not enumerable, so a test's own `{ ...store }` copy, which keeps its writes elsewhere, does not
inherit a subscription it would never fire. With it, the targeted split-group create (L289) passes
at this head, so its it.fails mark goes. L283's test runs on a store that names its partitions and
stays it.fails("L283").

* fix(terminal): recovery checks the caller's handle after the cheap guards again, so the switch-off build re-issues no handle on a refused recovery (layout-owner build, L80)

The stale-handle switch point had moved the handle lookup ahead of the pane, terminal and
workspace checks. That lookup can re-issue a handle, so a switch-off recovery refused on those
checks gained a side effect base never had. The lookup runs after them again, and the switch-on
replacement check lives in one method.

* test(runtime): the layout-owner start fixture's spawn commits the pane binding before main registers the terminal, as the host does (L80 switch-on variant)

The window-tab start test came in with the main merge after bind-first; its fake spawn returned
a terminal no pane was bound to, so registration refused it. The fixture's spawn now binds the
pane first, as every other bind-first fixture does.

* fix(terminal): a new process under a reused terminal id starts with no exit code, so an exit wait on it cannot settle on the old process's code (layout-owner build, L80)

The record of a reused id kept the earlier process's lastExitCode, and with the switch on an exit
wait settles on any recorded code once the terminal reads disconnected. A new incarnation now
clears the code and cause. Also drops the L283 test's partition-list option: the fixture store
already lists its partitions.

* test(runtime): switch-on variants for four tests the full local run never loaded (L80)

- A restored pane's host lists the terminal it started; a pane main owns vouches for none.
- The other pane holding a slept session is in main's layout, so an unverifiable holder blocks
  the resume as it does through a window.
- Main sleeps a workspace whose panes it lists itself, even with a renderer attached (stablyai#31571).
- A graph sync no longer reconnects a disconnected terminal of main's pane; the inventory decides.

* test(runtime): the phone Markdown route tests use stablyai#31667's realistic store, and the untitled cases wait on L268 with the switch on

The window's tab uses main's id (the file path), and the store lists its partitions and saves each
open editor's unified tab and group, as the window does (the fixture from stablyai#31667). With the
switch on, the three untitled cases fail on L268, main's row merge dropping the untitled flag,
which stablyai#31667 fixes; they are it.fails until A1b's history includes it.

* refactor(runtime): two merged files back under the line limit, by moving A1b's additions to their helpers (L80)

After the main merge, the holder-pane check and a closed tab's live-terminal lookup pushed two
files past the line limit. The holder check is now paneHoldsSession (main-terminal-surfaces.ts)
behind one method, and the closed tab's lookup is one parent-class method. Behaviour unchanged.

* test(runtime): the headless chat tests open the chat on the publish path, as every product path does; L283 stays it.fails for the main-owned reorder

openChat recorded the chat only in the runtime's snapshot, so main's layout never held it and the
phone's reorder was refused as an unknown tab. On the publish path main's layout holds it; the
reorder is accepted but does not survive the row rebuild after a close (L283, reopened).

* test(runtime): bind-first census follows the one spawn-commit step (stablyai#31596); the unbound-registration test keeps its spawn from the suites' stand-in commit

After the main merge both start lanes register in spawn-commit-assembly.ts, after
savePtySpawnPaneBinding (or, adopted, on the live owner's pane); the census lists that file. A
phone create now writes its pane before the spawn, so the suites' stand-in spawn commit bound it
and the test of a path that registers unbound passed for the wrong reason; that test's controller
now opts out of the stand-in commit.

* test(runtime): convert four switch-on tests the disguised full runs hid

Real switch-on runs showed four files whose fakes registered terminals unbound
or confirmed a stop without its exit. Bind first, stamp the source as main
does, and deliver the exit a confirmed stop delivers. Switch-off unchanged.

* test: type the two converted fakes instead of casting them

The PTY controller fake now takes the controller's optional pane fields, and the
switch-on store is a typed RuntimeStore, so neither needs an assertion.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

10 participants