Skip to content

feat(orchestration): add a Settings switch to turn worker heartbeats off - #34070

Merged
brennanb2025 merged 8 commits into
mainfrom
brennanb2025/orch-heartbeat-toggle
Oct 26, 2026
Merged

brennanb2025 merged 8 commits into
mainfrom
brennanb2025/orch-heartbeat-toggle

Conversation

@brennanb2025

@brennanb2025 brennanb2025 commented Oct 26, 2026 •

Copy link
Copy Markdown
Contributor
Files Added Deleted Net
Test 12 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​152 $\color{#cf222e}{\Huge{\mathbf{−}}}$​8 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​144
Prod 25 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​183 $\color{#cf222e}{\Huge{\mathbf{−}}}$​26 $\color{#1a7f37}{\Huge{\mathbf{+}}}$​157

Problem

Every dispatched orchestration worker is told to send a heartbeat to its coordinator every 5 minutes. There is no way to turn that off, so users who don't want the extra coordinator mail have to work around it in their own agent instructions.

Change

Settings → Agent Orchestration gets a Worker heartbeats switch, on by default (today's behavior unchanged).

  • Off: newly dispatched workers' briefs no longer mention heartbeats at all, and the legacy orchestration run loop's stale-heartbeat check is skipped.
  • Workers already running keep the brief they were given; the setting applies to workers dispatched after the change (the description says so).
  • Like the other orchestration settings, it is read from the runtime's own store, is not writable through the CLI settings RPC, and is hidden on paired web clients. An older host ignores it and keeps 5 minutes.

Plan

The heartbeat interval was two constants that had to agree: 5 minutes in the worker brief and a 10-minute (2×) stale threshold. Both now come from one resolver (src/shared/orchestration-heartbeat.ts) that returns the interval in minutes or null when off. All six places that build the brief pass that value, and the stale check uses the same value. No saved field or wire change beyond the optional setting itself, where a missing value means on.

We also considered recording each dispatch's interval on its row so the stale check would judge against what that worker was told. We chose not to, because the stale check's only output is a log callback that the production coordinator leaves empty (gates.ts constructs it without onLog). A stored column would make something no one sees more precise. Intended difference from the common pattern of recording the interval per session: here the brief itself carries the interval to the worker.

A follow-up PR adds the interval setting, shown while heartbeats are on.

Testing

Full-path review completed: correctness and architecture review, persistence/consumer trace, cross-platform/mobile readiness scan, performance pass, and final reference comparison. No production defect found. Two verification gaps were filled, then the tests were reduced to the focused contracts requested by the lane owner.

  • Real Store migration/save/reopen: older profiles inherit on; saved false survives reload and the runtime resolves it to off.
  • Renderer: default-on/save-off, plus a saved-off profile toggled back on.
  • Real CLI argv through handlers, RPC validation, runtime and resolver: off omits heartbeat text from the delivered/returned brief and retains completion, both outcomes, ask, escalation and mailbox-check instructions.
  • 63 tests passed across those three named suites. The original changed-file suites also passed before the test-only additions. Normal commit hooks and the explicit main-merge check passed.
  • Branch-only Settings screenshot confirmed the new switch is on by default with the complete description visible. The isolated copy stopped gracefully with no survivors. Settings/orchestration use qa=tests; no agent QA run is required.

Final CI 40412354258 is green on 3845982: static analysis/typecheck, all unit shards, cross-version compatibility, packaging and platform persistence checks passed. The screenshot and reference evidence were captured on 779d33f. Subsequent own commits change only tests/support; main merges 2161f4c, e35e544 and 3742278 passed narrow checks. CI-exposed fixture contracts were corrected, the affected CLI suite passed again (2/2), and the final test/support delta is 147 net lines against 147 production additions.

Dispatched workers are told to send a heartbeat every 5 minutes. A new
"Worker heartbeats" switch under Settings > Agent Orchestration (on by
default) lets users turn that off: newly dispatched workers are no longer
told to send heartbeats, and the stale-heartbeat check is skipped.

The interval the brief states and the stale check both derive from one
resolver over the live setting, read from the runtime's own store like
the other orchestration settings.
@brennanb2025
brennanb2025 marked this pull request as ready for review October 26, 2026 21:06
@coderabbitai

coderabbitai Bot commented Oct 26, 2026 •

Copy link
Copy Markdown
Contributor

Review in Change Stack →

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

  • @coderabbitai resume to resume automatic reviews.
  • @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

  • ▶️ Resume reviews
  • 🔍 Trigger review

No actionable comments were generated in the recent review.

Recent review info
Run configuration
  • Source mode: Auto
  • Repository: Unavailable
  • Branch: Unavailable
  • File: No configuration file from this repository applied
  • Commit: Unavailable
  • Layers: Repository UI
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 8a4948dd-5813-4084-b0cd-5c8ff0760dc8
Commits

Reviewing files that changed from the base of the PR and between 65358c5 and 3845982.

Files selected for processing: 1
  • src/main/runtime/rpc/methods/orchestration/cli-runtime-boundary.test.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 1 remain after this review.

Walkthrough

The changes add a worker-heartbeat setting that is enabled by default and can be disabled in local runtime settings. The runtime resolves the setting to a five-minute interval or null. Dispatch preambles include heartbeat instructions only when an interval is configured. The coordinator uses the interval to determine when to warn about stale dispatches and skips those checks when heartbeats are disabled.

Priority: Low

Merge Risk: Minimal · up to 38459

No actionable merge-blocking risk was identified in this test-only change.

Pre-merge checks: Passed 3 · Failed 2

Failed Failed checks: 2 warnings

Check name Status Explanation Resolution
Docstring Coverage Warning Docstring coverage is 28.57% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 21 functions across 29 files. Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check Warning The description explains the problem, change, implementation plan, and testing in detail, but it does not follow the repository template. It omits the required ELI5, What Changed, Why, Linked Issue, V… Update the description to include all required template sections. Add an actual issue reference under “Linked Issue” using Fixes #... or the maintainer exception. Add the required before/after visual proof, or write N/A with a reason if…
Passed Passed checks: 3 passed
Check name Status Explanation
Title check Passed The title clearly and concisely describes the main change: adding a Settings switch to disable worker heartbeats.
Linked Issues check Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check Passed Check skipped because no linked issues were found for this pull request.
Full details: Description check

Explanation

The description explains the problem, change, implementation plan, and testing in detail, but it does not follow the repository template. It omits the required ELI5, What Changed, Why, Linked Issue, Visual Proof, AI Disclosure, Review, boundary, Notes, and Checklist sections. It also does not provide an actual linked issue or the required visual-proof declaration or attachment.

Resolution

Update the description to include all required template sections. Add an actual issue reference under “Linked Issue” using Fixes #... or the maintainer exception. Add the required before/after visual proof, or write N/A with a reason if applicable. Complete the testing checkboxes, AI Disclosure, Review, Agent skill upstream boundary, Notes, and Checklist sections.

Finishing Touches: 💡 1
📝 Generate docstrings: 💡
  • Commit to this branch
  • Create a new PR
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@brennanb2025

brennanb2025 commented Oct 26, 2026 •

Copy link
Copy Markdown
Contributor Author

Workers always received five-minute heartbeat instructions, so users could not avoid the extra coordinator mail. This PR adds Worker heartbeats in Settings → Agent Orchestration, on by default. Turning it off omits heartbeat instructions from future worker briefs and skips the legacy stale check. Existing briefs and incoming completion/heartbeat reports continue to work.

The full review found no production defect. Two missing test contracts were added: older settings load/save/reopen with the opt-out preserved, and real CLI arguments through RPC/runtime to the delivered brief. The tests were then narrowed at the lane owner's direction to those contracts plus the pane test. Final CI also exposed a legacy partial-runtime fixture cast after the heartbeat stub changed its expression; a line-specific safety explanation was added without changing fixture behavior. The first executed typecheck then found an incomplete authority mock and a CLI import crossing the Node project boundary in that integration test. Both were corrected using existing patterns; the real argv/RPC/runtime path and assertions are retained. No findings remain deferred.

Verification: 63 tests passed in the three retained suites, and the two-test CLI suite passed again after its fixture/import corrections; normal commit hooks and all three explicit main-merge checks passed. Correctness, architecture, performance, cross-platform/mobile readiness, final reference comparison and the narrow final merge check passed. Final CI 40412354258 is green on 3845982104a: static analysis/typecheck, all five unit shards, selection evidence, cross-version compatibility and packaging passed, alongside the platform persistence and WSL checks. The bounded test-fix review also passed. Settings and orchestration use qa=tests; no real-agent QA round was required. The isolated Settings screenshot copy stopped gracefully with no survivors.

The live runtime preference is an intended difference from the common recorded-session pattern: the legacy stale callback reaches no production consumer and never fails, releases or wakes a worker. There is no dispatch migration or wire change. The interval control remains a separate follow-up.

For Brennan's review: the screenshot shows the switch enabled by default and its complete description. The key spots are the scope description and the heartbeat-only preamble branch.

Worker heartbeats in Settings

Screenshot/reference production evidence is from 779d33f32c897bd173d26f517c5451cf820a86a4. Since then, 06e33929f82 added test contracts, 3b38f8e0782 pruned them to the retained scope, and clean main merges 2161f4c67b2, e35e5448c8e, and 37422780d94175362e55805b2ba99023df37721b incorporated main and its CI repairs. 65358c53bc3 adds only that fixture safety annotation. 3845982104a corrects only the CLI test fixture/import. All four own commits change only tests/support, including the runtime test harness that accepts the settings fixture; the settings and CLI tests cover that tests-row flow on the final head, while the unchanged Settings screen retains its original screenshot; all main merges passed narrow review and preserve the reviewed heartbeat changes. Prior evidence carries. Final source/merge review is on 37422780d94175362e55805b2ba99023df37721b; the test-only deltas and final CI are on 3845982104a8d34f5df7e6b9bd56dec4bf1feb4b.

@brennanb2025
brennanb2025 merged commit ca8eb29 into main Oct 26, 2026
49 checks passed
nwparker added a commit that referenced this pull request Oct 27, 2026
…ts (#34123)

* fix(source-control): fit argv agent prompts to the local command line by truncating the diff

Antigravity and the other promptDelivery: 'argv' agents pass the whole
commit or PR prompt on the command line. The prompt builders capped the
diff at 200,000 characters, but runLocalSourceControlPlan refuses argv
over 30,000 UTF-16 units on Windows (and any argument over 120 KiB on
Linux), so a large branch diff failed before the agent started.

For local targets, when the planned argv would exceed that limit, pick
the largest diff budget whose prompt fits (binary search, so Windows
units and Linux bytes are both measured on the real plan) and rebuild
the prompt with it. This covers the default action template path as
well as plain prompts. WSL targets are left to the existing guard
because the WSL launcher re-wraps the command. The pre-spawn guard
stays as the backstop when the non-diff sections alone are too large.

Fixes #34109

* Unify Source Control action permissions and card layout (#30449)

* Source control actions start chats and terminals on their saved model

A "start an agent" action's model lived only in free-text CLI arguments, which a native chat
never reads, so AI Fix opened a chat on the default model. Each action now saves a structured
model and effort that both routes apply: chat create options, or the agent's own flags in a
terminal (winning over the same flag in CLI arguments). Desktop, dialog, fix-checks and mobile
launches carry it; Settings rows get Model/Effort selects and the CLI arguments field says it is
terminal-only.

Tracked in STA-11102.

* Name the model in the repository override note

* Type the dialog's model hand-off; keep Agent in English in Japanese copy

* fix: preserve action chat model choices through launch

* Preserve action selection ownership across chat launches

* Keep terminal host contract fixtures aligned with preflight

* fix: wait for action Settings agent inventory

* fix: consolidate direct agent routing imports

* fix: use standard action option labels

* fix: align action launch contracts with shared types

* fix: sync action option localization metadata

* test: provide action Settings inventory in mounted fixtures

* Keep action options row height stable while loading

* Let each Source Control action choose its chat permission mode

An action's saved chat choice may now carry a permission mode, with or without a model;
"Same as new chats" saves nothing. A model is replaced only when the action names one,
so a permission-only choice keeps the chat's remembered model and options.

* fix(native-chat): leave older hosts their own permission mode and keep action options readable

- An action's saved permission mode is sent only to hosts that advertise
  agentSession.create.options-mode.v1. Older hosts get no mode, in create
  options or a later setOption, so the chat starts on that host's new-chat
  default, as a phone launch to such a host already does. Hosts from
  1.4.225 to 1.4.227 refused a mode their tables lacked (OpenCode, Grok,
  OMP), leaving the action unable to start.
- Settings action rows give Model and Permissions full rows in the options
  column so saved values are readable; the loading placeholder reserves the
  same rows.
- Remove two unused locale keys; fix two stale comments.
- Tests: the older-host create carries no permission mode; a permission
  mode picked alone keeps the remembered model on an acquired create.

* fix(settings): reserve one options row while loading an action with nothing saved

Terminal and permissions-only rows no longer shrink when loading ends; a chat
agent with a model list may grow by its Model row once known.

* fix(native-chat): keep the create options mode a checked literal type

* fix(settings): share verified action option fields

* fix(settings): keep action selectors on standard typography

* Keep the launch files under their length limits after the main merge

The structured route imports its result types where they are defined, and the two older-host
action-launch tests share one setup helper.

* One action settings section for chat and terminal launches

A Source Control action's Model, model options and Permissions now apply whether it opens a
chat or a terminal; terminals receive them as the agent's own flags, and a repository's own
CLI arguments keep the flags it sets. Extra CLI arguments stay terminal-only.

* fix(source-control-ai): one terminal projection for saved action choices

Ask in a terminal now states its permission instead of leaving it to the
user's own config: Claude and Grok pass --permission-mode default, Codex
passes -a on-request -s read-only -c approvals_reviewer="user" (the
interactive CLI has no untrusted policy, so a read-only sandbox is what
asks before edits). Agents whose CLI has no flag for a mode show the
existing "Applies to chats only" hint.

The strict launch-preferences projection now requires the launch agent at
every caller, composes model ids through the resolver's own step, carries
a mode only where the agent's terminal has a flag for it, and the local
string wire keeps booleans as 'true'/'false', so Cursor Thinking/Fast reach
the terminal and an OpenCode permission-only choice is not refused.

The argument parser knows flag arity, so removing a boolean permission
flag keeps other options' values, following positionals and '--'.
Pre-admission and executor share one terminal admission check on the
options the terminal will receive. Customize with Save launches the recipe
as it will resolve once saved; legacy partial repository overrides are
edited as the complete recipe they resolve to. Empty Extra CLI arguments
start collapsed.

* fix(source-control-ai): an edited repository action saves complete

Every Settings edit of a repository action (agent, model, permission,
prompt or arguments) now starts from the complete recipe the action
resolves to, whether its stored override is owned or a legacy partial.
The resolved agent selection is kept (never an agent only detected to run
a default), an explicit null args clear stays null, and empty args stay
empty instead of being dropped. Clearing the last preference therefore
means the action agent's own defaults rather than re-inheriting Global.

The action text draft is lossless: unset or cleared prompt and args stay
so until the user types, and the inputs show them as empty, so saving a
prompt no longer writes empty args that suppress the agent's default
flags. Untouched sibling actions are not rewritten by the edit.

* fix(settings): style the Extra CLI arguments toggle through Button's own variant and size

CI's changed-code quality check refuses restyle classes on Button and
CollapsibleContent. Use the ghost variant at size xs with a margin, and
put the content spacing on a plain inner wrapper.

* fix: keep saved action and restart launch recipes consistent

* fix: describe inherited arguments and chat-only permissions accurately

* test: check repository action settings fixture type

* test: map saved action options into terminal compiler inputs

* fix: retain saved action controls during agent detection

* fix(i18n): retire obsolete Portuguese action labels

* test: align folder launch fallback mock with outcome export

* fix: carry resolved connector launch intent and mode together

* test: pin shared action launch branches to recorded agent output

* test: preserve saved source control choices through detection and migration

* test: align launch fixtures with integrated result contract

* test: retain recipe and restart contracts after main integration

* Unify Source Control action permission settings

* Keep native Ask copy with its owning change

* Preserve action drafts and old-host permission intent

* Consume rejected Source Control action saves

* Follow the moved Settings persistence fixture

* Keep permission integration contracts in the cross-version test project

* Include the action permission inheritance label in English catalogs

* Align launch contracts with shared permission recipes

* Keep Source Control action controls within their Settings columns

* Use primitive variants for compact action settings controls

* Show action arguments only for terminal defaults and widen recipe rows

* Apply terminal-only action arguments visibility to repository overrides

* Align action permission control with model controls

* Use shared settings label for action permissions

* Hide action model controls until argument migration ships

* feat(wsl-host): start a WSL host's distro when one of its automations comes due (E9 B) (#33839)

* feat(wsl-host): start a WSL host's distro when one of its automations comes due

E9 connect policy, part 2. A WSL host's automations run inside its distro, so a
distro that is off (a reboot, an idle release) missed them. While a host is connected
the desktop now reads its enabled automations' next run (on connect and every 15 min)
and keeps the soonest one per host in userData. When that time comes with the host off,
it boots the distro once, and orcad's startup catch-up fires the run within its missed-
run grace. A run due while the computer was off fires once Orca starts. Nothing wakes
past the grace window, and a connected host is left to its own scheduler.

Co-Authored-By: Claude <noreply@anthropic.com>

* test(wsl-host): a due automation boots a parked host once, never with WSL hosts off

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(wsl-host): keep the following run, and forget changed automations (E9 review)

Review of #33839:
- A run handled while connected dropped the host's next due time. The wake now keeps
  each host's enabled schedules and computes the following run locally whatever
  happens to this one (wake failed, past grace, handled by the host), and re-reads the
  real one from a connected host.
- A deleted, disabled or rescheduled automation kept waking the distro at its old
  time. A successful automation.create/update/delete from this desktop now re-reads the
  host's automations, as do the 15-minute re-read and every connect. Nothing boots a
  distro to check: on wake, a distro already running is just reconnected.
- wakeForAutomation/isConnected live here now (moved out of #33838); isConnected reads
  the host's reachability, never booting.

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor(wsl-host): say isConnected answers for this process; keep the service in bounds

isConnected reports whether the host answers through this process; false is not
"stopped" (a fresh process has no bridge yet, and the wake just connects). The
service's quiet record read moves beside the other record helpers so the file stays
within its line limit after the E9 additions.

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor(wsl-host): give the connector contract its own file; one wake path for focus and automations

After rebasing onto #34063, #34093 and #33110 the WSL host service was over its line
limit. The connector contract (what the service drives) moves to
wsl-host-connector-contract.ts, re-exported from the service, and the focus entry and
the automation wake share one private wake step: lift the park, then reconnect unless
the host answers, read with #33110's focusReachability so neither pays a WSL round trip
when this process has no bridge yet. An automation wake no longer reconnects a host
that already answers.

Co-Authored-By: Claude <noreply@anthropic.com>

* test(wsl-host): an automation wake with no bridge skips the reachability probe

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>

* test(connector): freeze the shipped connector schema behind an upgrade test (#34115)

Freezes the connector's shipped table definitions: later schema changes must be additive statements. Adds a baseline fixture and an upgrade test (baseline, then current schema, compared with a fresh database on SQLite and Postgres), plus the statement-shape check relay already uses. No runtime change.

* test: remove 30 low-value cases and seven suites (#34120)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* test(terminal): a pinned parked tab's process exit asks first, then closes as a pty-exit (#34121)

Co-authored-by: Claude <noreply@anthropic.com>

* fix(ssh): move enabled SSH automations to the host's Orca server (#34044)

* fix(ssh): move enabled SSH automations to the host's Orca server

Connecting an SSH host moves its saved state to the host's managed Orca
server. An enabled SSH automation refused that move on every Connect
("automations (1), automation runs (4)"), leaving the host unconnectable;
Move anyway connected it but stranded the automation on the desktop, where
it no longer ran.

An enabled automation whose runs have all finished now moves with its run
history. The manifest still carries it disabled, which every server version
accepts; once the move commits the desktop disables its retained copy and
turns the server's copy back on through automation.update, re-sending its
start time so the next run is the next future occurrence. A server that
can't is noted in Settings; an unreachable one is retried on the next
Connect. While a move fences the host, the desktop scheduler, Run now and
the launch re-check write nothing for its automations, so a tick can no
longer break the fenced export; releasing the fence resumes them. A run
still in flight keeps blocking and the refusal now says to wait for it.

Fixes #29083

* test(ssh): mock the server automation enabler in the delta-move handler test

* test(e2e): keep the Move anyway specs refusing now that enabled automations move

The inline Move anyway spec seeded an enabled automation to force a
refusal, which this branch now moves instead. It seeds one pinned to an
earlier registration of the host, which still stays behind. The relay
bridge spec's Move anyway lookup also matched the refusal toast's
button since #34041; it now takes the pane's own.

* fix(connector): one turn id from the host's stamp everywhere (#33909)

The host's turn-start timestamp is the only turn id the connector uses: snapshots, turn ends and replies all carry it, and the desktop's separate turn cache is gone, so they can no longer disagree. The unused turn-bound cancel was removed (Stop is parked). The hook server reads the main agent state through the same shared accessor.

* feat(orchestration): add a Settings switch to turn worker heartbeats off (#34070)

* feat(orchestration): add a Settings switch to turn worker heartbeats off

Dispatched workers are told to send a heartbeat every 5 minutes. A new
"Worker heartbeats" switch under Settings > Agent Orchestration (on by
default) lets users turn that off: newly dispatched workers are no longer
told to send heartbeats, and the stale-heartbeat check is skipped.

The interval the brief states and the stale check both derive from one
resolver over the live setting, read from the runtime's own store like
the other orchestration settings.

* Test heartbeat preferences across settings and worker dispatch

* Keep focused heartbeat preference contracts

* Explain partial runtime fixture assertion

* Correct CLI integration fixture contracts

* fix(wsl-host): a cold wake that loses orcad recovers by itself instead of leaving a dead pane (#33147)

* fix(wsl-host): a cold wake that loses orcad recovers by itself; stopped-distro refusals stay final

* fix(wsl-host): identity and pairing refusals stay final; no second wake after a stop; settling is retryable

* fix(files): open host files outside workspaces for any client with terminal access (#33958)

* fix(files): open host files outside workspaces for any client with terminal access

Paired clients that can open terminals on a host can already read and write
anything its user can, so the per-terminal provenance gate on files outside
workspaces protected nothing. The host now grants any existing file outside
its workspaces; older hosts get a clear update message instead of a refusal.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(files): refuse SSH, device and WSL-home paths instead of misreporting them

Co-Authored-By: Claude <noreply@anthropic.com>

* perf(files): size grant digests to the file and skip status checks for local fallbacks

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(files): read procfs files whose stat reports size 0

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>

* fix(mobile): resume history sessions with desktop's resume and host rules (#33943)

* fix(terminal): keep ConPTY's viewport top through a width change (#34061)

* fix(terminal): keep ConPTY's viewport top through a width change, checked against a captured transcript

A captured ConPTY 1.23 resize (wsl.exe) shows it keeps the line at the top
of its viewport through every resize and repaints only that viewport. xterm
keeps the cursor's rows instead: a wider grid that rejoins wrapped rows pulls
rows out of scrollback (ConPTY's repaint then overwrites them), and a
narrower one that rewraps pushes rows into it (the repaint then repeats
them). The windowsPty row rule covered neither. ConPTY-backed emulators now
put the top line back on its row after a resize.

* fix(terminal): ConPTY resize correction respects unfinished escapes and history replay

- Skip the widen correction while the parser is inside an escape sequence,
  so a resize landing mid-SGR/OSC no longer aborts the app's sequence.
- Skip it when an app set a scroll region.
- Replay daemon history (cold restore from log, durable rebuild) with the
  live emulator's ConPTY model, so replayed and live screens agree.
- Keep the saved cursor with its text when rows come back on narrowing,
  and never drop a background-coloured blank row.

* fix(terminal): replay a saved log with the resize model of the daemon that wrote it

- A daemon now stamps resizeModel into each session's meta; the history reader uses the ConPTY
  replay model only for a stamped session and keeps xterm's for one an older daemon recorded.
- Fix the DECSC adjustment: xterm keeps savedY absolute, so popping a blank row already keeps
  DECRC on its text; the old +1 sent it to the top row.
- Tests: narrow-direction replay, older-daemon replay, DECSTBM skip, DECSC/DECRC after narrowing,
  a background-coloured blank row kept.

* fix(terminal): take the history resize model from the daemon hosting the session

Main writes history and rebuilds durable checkpoints, but a daemon outlives an app update, so
main's own code is not what the live emulator ran. The daemon now reports its resize model in
hello (older daemons omit it, meaning xterm's); main stamps it into meta on open, restamps it
when a checkpoint starts a fresh log, and rebuilds durable history with it. The client's token
read moved to its own module to keep client.ts under the line limit.

Also clamps a saved cursor (DECSC) on a popped row to the new bottom row, as xterm's reflow does.

* fix(terminal): a disconnect leaves the history resize stamp alone

The client keeps the last daemon's reported model through a disconnect (the next hello replaces
it); undefined, before any daemon has answered, now means unknown and skips the restamp instead
of reading as an older daemon. Tests: a disconnect mid-commit keeps the stamp, and a Sleep-cache
wake onto a newer daemon restamps with its first checkpoint (#34060).

* fix(terminal): read the rebuild's resize model through the history manager

The durable rebuild called client.getDaemonResizeModel() in the checkpoint path, which tests that
fake the client with only request() could not answer, so the rebuild fell back to the live
snapshot. The history manager already holds the hosting model's provider; the rebuild reads it
there, keeping the checkpoint path's client use to request().

* refactor(connector): the desktop takes Discord and Linear conversations as well as Slack (#33927)

The desktop's connector accepts conversations from Slack, Discord and Linear and lists all three sources in its hello; per-source details (the Sent from line) come from one table. Nothing changes until those adapters are configured; behind the experimentalConnectors setting.

* fix(settings): show cookie import as available before skill setup (#34097)

* Linux cookie import says when secret-tool is missing instead of blaming the keyring (#34099)

* fix(browser): report why Linux cookie keys cannot be read

* test: cover cookie warning forwarding across import hosts

* test: place wire fixture cast rationale at assertion start

* fix: declare Linux cookie import warning translations

* test: extract WSL host connector fixture to restore lint (#34132)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* Size tabs to titles and preserve floating panel drag space (#34025)

* Size tabs to titles and preserve floating panel drag space

* Stabilize agent tab layout during activity changes

* Balance tab widths across titles and tab types

* Reserve terminal tab icon space before agent detection

* Run floating header drag E2E in the isolated headful project

* Prevent native text selection on the floating panel drag handle

* Cap expanded tab width at 24rem

* Let tabs expand to fit titles in available space

* test: remove 81 low-value cases and 12 suites (#34130)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* fix(agents): launch Claude Agent Teams as plain orca on Orca servers (#34116)

* fix(agents): launch Claude Agent Teams as plain orca on Orca servers

A paired Orca server has no SSH connection id, so client launches treated it
as a local Linux desktop and typed `orca-ide claude-teams`. Headless servers
only put `orca` on PATH. Client launches now read the resolved execution host
(SSH or server, except WSL workspaces), and launches built on a headless Linux
server use the launcher that server publishes.

Fixes #33920. Reported by @williamhrs. Client-side approach from #33987 by
@coolwindX; #33975 by @Shubham-Padkonde showed the host-built launch gap.

* fix(orcad): keep orca-ide working for older clients on Linux servers

Older clients type the desktop orca-ide claude-teams into server terminals.
Write an identical orca-ide launcher beside orca. Idea and test from #33975
by @Shubham-Padkonde.

* fix(ssh): Stop and remove keeps an SSH host's managed server removed across reload and restart (#34107)

* fix(ssh): Stop and remove keeps the managed server removed across reload and restart (#34003)

* fix(ssh): recheck the removal hold after a queued Stop settles

* fix(ssh): a new managed-server claim lifts the Stop and remove hold

* fix(web): a TUI copy the browser refuses offers a Copy button (#34135)

Browser clipboards need a recent user gesture, and an OSC 52 copy from a
TUI arrives after a host round trip that can outlive it. The web client
now shows a replaceable toast whose Copy click writes the refused text.
Desktop keeps its existing failure toast.

Refs #33937

* fix(editor): let desktop edit host files outside workspaces (#34119)

* fix(files): open host files outside workspaces for any client with terminal access

Paired clients that can open terminals on a host can already read and write
anything its user can, so the per-terminal provenance gate on files outside
workspaces protected nothing. The host now grants any existing file outside
its workspaces; older hosts get a clear update message instead of a refusal.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(files): refuse SSH, device and WSL-home paths instead of misreporting them

Co-Authored-By: Claude <noreply@anthropic.com>

* perf(files): size grant digests to the file and skip status checks for local fallbacks

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(files): read procfs files whose stat reports size 0

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): let desktop edit host files outside workspaces

Paired clients with terminal access can already write any file on the host, and
phones can now save these files, so desktop grant tabs open editable too. Saves go
through the host grant, renewing it when it lapsed and keeping the disk-change check;
a grant tab never falls back to writing this computer's same path.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): renew lapsed host grants for reads, banners and restored tabs

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): ask older hosts for an update instead of reporting a lapsed grant as deleted

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): keep grant tabs read-only on hosts that cannot renew grants

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): check a discarded save right before the grant write is sent

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): pin grant saves to one pairing and grant restored tabs before saving

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): open grant tabs read-only when the host cannot be asked, and restore read-only grants as read-only

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): pin a restored tab's grant request to the save's pairing

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): treat only a proven not-found as a missing granted file

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): keep restored drafts by refusing read-only grant saves instead of locking the tab

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): read a restored grant on the pairing that issued it and keep read-only grants off the tab

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): let every reader of a restored host-file tab ask its host for a grant

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): add no await before ordinary reads and saves pin their pairing

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): explain unrecreatable deleted host files and hide rename for grant tabs

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): move the grant capability probe out of the store-importing module to break an import cycle

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): renew grants without cross-workspace retargeting and settle saves whose reply was lost

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): read back a lost write reply through a fresh grant after a disconnect

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): trust only an unanswered write with byte-exact readback as saved

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(editor): read back only sent writes that lost their connection, including desktop timeout envelopes

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>

* fix(runtime-environments): keep session partitions of saved servers the listing hides (#34134)

The registry watcher now reconciles orphaned runtime sessions on every
write, so the reconcile must only drop hosts the file proves are gone. It
read the filtered listing, which omits newer-format records and servers
hidden by an unreadable sidecar or WSL overlay, and checked file presence
before reading, so a file vanishing in between read as an empty registry.
Build the known set from every raw record id, fail closed on a missing
file or an id-less record, and keep the catalog broadcast running if the
reconcile throws. Follow-up to #34101.

* fix(session-tabs): keep the server's per-device tab pick on the tab the desktop shows (#34082)

* fix(session-tabs): keep the server's per-device tab pick on the tab the desktop shows

The server hands its remembered pick back whenever the desktop has no tab groups of its own for a workspace. A desktop that kept its groups never reported what it showed, so that pick went stale and such an open landed on the wrong tab.

* fix(session-tabs): report only a ready visible pane, and retry after an error answer

Name the pane the desktop shows and require it ready, so a report never starts a pending sibling's terminal. A failure envelope no longer marks the pick as reported.

* fix(session-tabs): report a desktop's pane once per change and connection

Two windows sharing one device id could steer the pick back and forth. Report again only when this desktop's own pane changes, scope the record to the pairing and connection, and resolve the visible tab with the existing helper.

* fix(session-tabs): report only the on-screen workspace, compare panes, and remember agreement

Reports stay with the workspace on screen, whose pane attaches anyway, so a report never starts a terminal elsewhere. The pick is compared pane for pane, and an agreeing frame records the pane so a later change of this desktop's pane reports again.

* fix(session-tabs): settle before reporting, report on opening a workspace, and only to its owner

A click records its activation after an async hop, so a report now waits until the pane settles and compares against the latest frame. Opening a workspace reports from its latest frame. A workspace id two servers share is reported only to its owner. The e2e now drives the real flow: a stale pick, then a desktop restart.

* fix(session-tabs): keep one outstanding pick check per workspace, fenced to its connection

A busy workspace's frames no longer postpone the check. A frame from a replaced pairing or connection is dropped, and a frame without the per-device projection clears the cached one.

* fix(session-tabs): bound the pick report caches and clear them on mirror teardown

One record per workspace, with the connection in the value. Environment teardown drops reports, frames and timers. The tests use checked types, and the e2e notes why it waits and why it needs more time.

* fix(session-tabs): check again when the user moved on before a report finished

The host does not order activations, so a report that completes after the user picked another pane may have landed last. Clear its record and recheck from the latest frame.

* fix(session-tabs): type the visible pane's leaf as optional

* fix(session-tabs): send a browser or chat pick again when a late report landed after it

A terminal report that completes after the user picked a browser or chat tab may have had the last word. Re-send that pick the way a click does.

* fix(session-tabs): recheck a late report against the host's answer, and fence it to teardown

The activate answer is the host's per-device frame after the report, so a late landing is checked against it instead of a cached frame. A report answered after mirror teardown or a new connection does nothing.

* test(session-tabs): drop the in-connection suppression check from the restart e2e

Whether the desktop repairs a stale pick within the connection depends on which frames settle first; the guarded outcome is the pick after a restart.

* fix(session-tabs): reconcile the user's newer pick after any report outcome, timeouts included

A report whose answer timed out may still have landed after the user's newer pick. Settle every outcome the same way: retry on failure, then reconcile with what the desktop shows.

* fix(source-control): stream large Antigravity generation prompts over stdin (#34148)

* fix(rate-limits): show agy's own error when Antigravity usage can't be read (#34106)

When agy refuses a usage read with its own error envelope (for example the location eligibility rejection), Orca now shows agy's text, redacted and bounded to one line, instead of the generic "did not report a quota" message. Existing failure precedence and the last good quota bars are unchanged.

Fixes #33426

* fix(antigravity): save an agy account whose credential has no id_token (#33580)

agy 1.4.7 stores no id_token, so Save current account stayed disabled and the account could not be saved. Save now asks Google's userinfo endpoint which account owns the current access token (only on the Save click), then saves it, or updates the saved account with the same Google subject. The credential is re-read after the lookup, and nothing is saved if it changed.

Fixes #33323

* Refine lint performance and safe process execution (#34147)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* fix(agent-status): keep an Antigravity run working while its helper agents finish (#34128)

When an Antigravity parent ends its turn to wait for invoke_subagent helpers, agy reports the parent's Stop as fullyIdle=false, but its own terminal status says done at each turn end. After two minutes of hook silence Orca let that terminal done replace the hook row, so automations completed while helpers were still working. The parent's busy Stop now marks the row as held open for helpers, and a program's non-boundary done can no longer settle a row held open for child work. Late helper hooks after a completed turn are rejected before they touch parent state.

Verified with real agy 1.4.7: helpers slower than two minutes ended runs early on main (3/3) and complete with all helper output and the final report on this change.

Fixes #33973

* fix(mobile): match history sessions and scope paths with desktop's rules (#33955)

* test: load real modules before timed waits, and ratchet in-test vi.importActual (#34151)

* test: load real modules before timed waits, and ratchet in-test vi.importActual

Two tests paid a cold real-module load inside their timed wait:
- the empty-remote wake test's 1s vi.waitFor raced the post-create refresh's
  lazy session-tabs import; it now loads those modules in beforeAll;
- the runtime label-only create tests called vi.importActual inside each test
  (19-23s of the 30s budget on the first case); it now runs in beforeAll.

test-no-in-test-import-actual flags vi.importActual outside vi.mock/doMock/
hoisted factories, beforeAll and module scope; the 50 existing findings are
grandfathered in the flaky-test ratchet baseline.

* test: flag vi.importActual in a doMock factory registered inside a test

* test: flag vi.importActual in any vi.doMock factory; its load runs on the next import

* test: remove 47 low-value cases and 13 suites (#34136)

* test: remove 37 low-value cases and 3 suites

* test: delete ten more renderer wiring and copy suites

* test: drop deleted suite from its contract entry point

---------

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* Compact worktree project and host pickers with stacked details (#34098)

* Compact worktree project and host pickers with stacked details

* Align worktree picker markers and action gutters

* Extract WSL connector fixture to satisfy file length limit

* Restart CI after resolving upstream fixture conflict

* refactor: group remote runtime client modules (#34133)

* Revert benchmark-only startup loading study (#33912) (#34162)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* test: remove 294 low-value cases and 12 suites (#34156)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* fix(connector): the Discord gateway sends queued posts, so its replies don't wait for the API to wake (#34131)

The always-on Discord gateway worker now runs the same background sweeps as the API (queued posts, held starts, cleanup), so Discord replies go out at once instead of waiting for the scaled-to-zero API to wake. Posts are claimed with leases, so running the sweeps in both services never sends one twice.

* refactor(mobile): share agent child-row text and favicon domains with desktop (#33962)

* refactor(mobile): share child-row detail text and agent favicon domains with desktop

* refactor: share the agent tool preview line with child rows

* refactor: translate the child-row Ended label only when shown

* fix(mobile): show chat images stored on the host as thumbnails (#33731)

* fix(mobile): show chat images stored on the host as thumbnails

Fixes #33554

* fix(mobile): read stored chat images two at a time

A burst of megabyte replies over a slow link overflows the host's outbound queue and drops the phone's socket.

* refactor(mobile): tidy stored chat image read

* fix(mobile): skip stored image reads for rows that already left

Also ignore a result read through a different host client.

* feat(connector): each Slack app verified with its own signing secret at its own URL (#34125)

Groundwork for per-company Slack apps. Each Slack app gets its own row with encrypted secrets, and Slack posts to /slack/apps/<app id>/..., where the address only selects which secret to check; a request acts only in a workspace where that app is installed. Rate limits are per address and per app, so one workspace can't hold up another's posts. Schema changes are additive. The existing staging app keeps working on the old route until a documented cutover. Inert for users.

* test: remove 368 low-value cases and 4 suites (#34157)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

* fix(source-control): account for Windows quoting when fitting argv prompts

---------

Co-authored-by: Brennan Benson <79079362+brennanb2025@users.noreply.github.com>
Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Jinwoo Hong <73622457+Jinwoo-H@users.noreply.github.com>
Co-authored-by: OrcaWin <alpha-eng@stably.ai>
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
Co-authored-by: Kelvin Amoaba <97001695+AmoabaKelvin@users.noreply.github.com>
Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: Manan <100475418+thefrikidude@users.noreply.github.com>
Co-authored-by: Neil <neil@stably.ai>
@brennanb2025
brennanb2025 deleted the brennanb2025/orch-heartbeat-toggle branch October 27, 2026 02:38
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.

1 participant