Commit Graph
21681 Commits
Author SHA1 Message Date
Jinjing a54cea03c5 Improve notice details 2026-11-15 13:50:51 -08:00
OrcaWinandOrcaWin 6161632e6c test: remove 34 low-value suites (#42822)
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
2026-11-15 13:45:47 -08:00
Kelvin AmoabaandRadu Toader 79377cb989 feat(mobile): return to a chat tab where you left its transcript (#42584)
* feat(mobile): return to a chat tab where you left its transcript

One chat view serves every session tab, and a tab switch only changes its
scope, so the list jumped to the tail and opened tool runs collapsed on every
return. The page list now saves each chat's reading place and drawn rows
while the reader is away from the tail and puts them back when the chat shows
again. Opened tool runs, and the row keys they are saved under, are kept per
chat too. A chat left at its tail, or whose anchor rows are gone, still opens
at the tail and follows.

Co-authored-by: Radu Toader <4960306+radut@users.noreply.github.com>

* fix(mobile): keep a chat's saved place through a reload, and relative to the scroller

A tab switch to a chat whose rows are still loading no longer forgets its
saved place, so the list puts it back once the rows draw; sending or jumping
to latest still forgets it. Reading anchors now measure from the scroller's
top, so a place saved on an earlier visit holds if the scroller moved.

---------

Co-authored-by: Radu Toader <4960306+radut@users.noreply.github.com>
2026-11-15 21:43:59 +00:00
Neil 5b53483928 fix(native-chat): repair main import and projection lint failures (#42817)
* fix(native-chat): resolve moved async question module in journal status

* fix(native-chat): simplify optional status tool normalization
2026-11-15 13:36:10 -08:00
Brennan Benson e65baeb7ee fix(storage): reclaim stale managed runtimes on launch and gzip usage source sidecars (#42807)
* fix(storage): reclaim stale managed runtimes on launch and gzip usage source sidecars

* fix(storage): reclaim on orcad hosts too; skip throwaway gzip writes
2026-11-15 13:35:58 -08:00
Jinwoo Hong 9bebd77c77 fix: repair the main build after the Codex async question module move (#42818)
Point journal-sql-status-items at the moved codex-async-question-input module (#42620 moved it before #42459 landed with the old path) and inline a one-use callId so structured-agent-session-projection.ts is back under the 300-line limit. No behavior change.
2026-11-15 16:31:35 -05:00
Neil 69cfecdf70 refactor: organize shared lifecycle and runtime owners (#42792) 2026-11-15 13:18:34 -08:00
Jinjing abb8b7cf7c Fix sign-out and workflow translations (#41419)
* fix(i18n): refresh sign-out and workflow translations

* fix(i18n): clarify preserved branch cleanup status

* fix(i18n): align activity agent terminology
2026-11-15 13:08:56 -08:00
Kelvin AmoabaandTobani Esan-George 2c789f9871 feat(native-chat): run Cursor in Chat over its ACP server (#42193)
Cursor chats run `cursor-agent acp` with the host's Cursor login. Models
come from cursor/list_available_models with each model's own effort levels
and Fast default, and effort, Fast and model are written as Cursor's own
options. Cursor's question and plan requests name no session, so they are
routed to the chat and shown as cards instead of being refused, which made
Cursor fall back to a permission request Full access would have answered.
Only Claude and Codex treat an unpicked Fast as off.

Carries over #41494.

Co-authored-by: Tobani Esan-George <tobanigeorge@gmail.com>
2026-11-15 21:07:20 +00:00
Kelvin Amoaba c7d4cb5d52 feat(pi): show native chat subagents from the managed status extension (#42448)
Native chat strips the hook endpoint, so the managed extension's subagent
roster never reached Pi chats. In structured sessions the extension now
publishes the roster as a status line, and the Pi RPC adapter turns it into
child-work evidence for the existing subagent rows.
2026-11-15 21:03:12 +00:00
OrcaWinandOrcaWin cbbdc26c63 test: remove 16 low-value suites and trim process/stream cases (#42805)
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
2026-11-15 13:02:38 -08:00
Kelvin Amoaba 01eb509036 fix(native-chat): reopen an OpenCode chat in the folder OpenCode moved it to (#42590)
OpenCode 2.x moves a conversation with its session_move tool and then refuses session/load from the old folder. Look the session up with session/list and load it where it lives instead of replacing it with a new session.
2026-11-15 20:59:23 +00:00
Kelvin AmoabaandRadu Toader 3a8b23ab04 feat(mobile): scroll a chat reply to its start from the ↑ under it (#42583)
* feat(mobile): scroll a chat reply to its start from the ↑ under it

Desktop's reply controls have an ↑ that puts the reply's start at the top
and releases the tail follow. Mobile replies had only Copy and Fork. Add the
same action to the page list, which draws every phone chat, reusing
desktop's scrollMessageToTop strings.

Co-authored-by: Radu Toader <4960306+radut@users.noreply.github.com>

* fix(mobile): keep a reply scrolled to the top through growth right after the tap

---------

Co-authored-by: Radu Toader <4960306+radut@users.noreply.github.com>
2026-11-15 20:59:05 +00:00
Kelvin Amoabaandarbourd ccce2543ab fix(terminal): open links on middle-click instead of dropping or pasting (#42579)
* fix(terminal): open links on middle-click instead of dropping or pasting

Middle-click never opened a terminal link. The PTY claim window only
opened for a plain left press, so the link action was always refused.
On macOS and Linux, primary-selection paste also stopped the press
before the link code saw it and pasted instead.

Middle-click now uses the same deferred claim as plain click, and the
paste handler steps aside when a link is under the pointer, the
terminal has no selection, and middle-click isn't set to none.

Co-authored-by: arbourd <7211830+arbourd@users.noreply.github.com>

* fix(terminal): let OSC 8 and handle links take a middle-click

* fix(terminal): skip the native-paste guard when a middle-click opens a link

---------

Co-authored-by: arbourd <7211830+arbourd@users.noreply.github.com>
2026-11-15 20:58:48 +00:00
Kelvin Amoaba bcf6c709a3 fix(native-chat): show a Codex chat with an open async question as waiting on you (#42459)
Codex's request_user_input_async question is journaled as a call block on an assistant message, so the structured status never saw it: the chat read working, then done, while its question card stayed open. The host journal reads now flag such a question (awaitsAnswer) by the card's own rule until the person's next delivered message, and the lead state maps it to waiting. Turn status, prompts, notifications, teardown and saved turns are unchanged.
2026-11-15 20:58:31 +00:00
Kelvin Amoaba d4db6a1d51 feat(mobile): show the desktop's project groups in the workspace list (#42457)
* feat(mobile): nest the workspace list under the desktop's project groups

Shipped phones listed every project flat while the desktop sidebar nests projects under project groups. The desktop's group-tree walk is now shared, and the phone's list uses it with the groups the desktop window's catalog now carries (or projectGroup.list on older desktops). Collapse shares the desktop's project-group keys, and folder workspaces sit under their group header.

* test(mobile): type the parity fixture as the served catalog
2026-11-15 20:58:13 +00:00
Kelvin Amoaba eaef504449 feat(tasks): filter GitHub tasks by several authors (#42450)
* feat(tasks): filter GitHub tasks by several authors

* fix(tasks): keep author candidates across picks

* refactor(tasks): share author filter summary and only-me check
2026-11-15 20:57:54 +00:00
Kelvin AmoabaandGarrett Peppers 54ad5c739a fix(native-chat): run /new as the host conversation clear in Codex chats (#42445)
Codex's app-server has no slash parser, so a typed /new reached the
model as prose and the chat kept its history. The shared structured
composer now resolves Codex's /new to the host clear, on desktop and
mobile alike.

Co-authored-by: Garrett Peppers <garrett@pepperstech.net>
2026-11-15 20:57:37 +00:00
Kelvin AmoabaandDunz Koi 7960957c75 feat(browser): pick a browser profile for each repo's new tabs (#42803)
Each project can pick a browser profile (Default or a named one) for its new tabs, overriding the Browser setting's scope. Also settable with `orca repo set --browser-profile`.

Co-authored-by: Dunz Koi <258568505+dunzkoi@users.noreply.github.com>
2026-11-15 20:56:14 +00:00
Jinjing a0b2197418 fix(status-bar): color usage bars only in provider detail cards (#42795) 2026-11-15 13:49:26 -07:00
Kelvin Amoaba 42419ed7c0 feat(mobile): one link setting for chat and terminal links, with Ask each time (#42802)
* feat(mobile): open chat links where the link setting says, with Ask each time

The Browser setting now covers web links in terminal output and chat with
one rule: Ask each time shows the link sheet, Orca browser opens a desktop
tab directly (the phone browser where no tab can open), and Phone browser
opens the phone browser. Nothing saved means Ask each time; saved choices
keep the same storage key.

* perf(mobile): keep the chat link handler stable across reconnects
2026-11-15 20:46:41 +00:00
Kelvin Amoaba 28fd7052ad feat(computer): list macOS apps without a Dock icon while they show a window (#42786)
* feat(computer): list macOS apps without a Dock icon while they show a window

list-apps on macOS only listed regular apps, so menu bar and other accessory apps could not be found or targeted by name. Accessory apps are now listed while they own an on-screen window between the normal and status bar levels, which keeps the dozens of windowless system agents out, and each listed app says whether it is an accessory app.

* fix(computer): list only normal-level accessory windows and resolve hidden accessory apps

* fix(computer): prefer a same-named regular app when resolving, not in the list order
2026-11-15 20:33:53 +00:00
Kelvin Amoaba b92c5efe84 fix(notes): keep idle Orca-launched Pi tabs selectable in Send notes to (#42797)
* fix(notes): keep Orca-launched Pi and OMP panes sendable by their title

Orca rebrands an owned pane's π title to Pi or OMP, but the send gate only
read the raw π - prefix, so an idle Pi tab dropped out of Send notes to once
its status row went stale.

* test(notes): pin owned Pi and OMP titles at the send gate only
2026-11-15 20:24:25 +00:00
Kelvin AmoabaandAleksei Tcelishchev 0d07202c9b fix(mobile): take back phone-driven browser tabs with Take back all (#42785)
Take back all only restored the PTYs the phone held, so a browser tab the phone was driving stayed locked at phone size. Reclaim every mobile-driven browser page alongside the terminals; held-fit Restore all still only resizes.

Co-authored-by: Aleksei Tcelishchev <1243074+casualuser@users.noreply.github.com>
2026-11-15 20:21:04 +00:00
Brennan Benson c120b5bfb3 fix(antigravity): skip the hook script and python3 outside an Orca pane (#42790)
agy runs Orca's hook on every message and before and after every tool call.
Outside Orca the command still started two shells and python3 before the
script noticed ORCA_PANE_KEY was empty: about 0.1s per hook, ~0.2s per tool
call, even after Orca was uninstalled (#42601).

The /bin/sh -c command now checks ORCA_PANE_KEY first, like Grok's. Outside a
pane it prints the reply the script would have printed and drains stdin in
the background, so hosts that keep stdin open (ACP) never block.
2026-11-15 12:20:03 -08:00
Brennan Benson 2e77caf0ea fix: point agent-status imports at the moved structural-value-equality module (#42793)
#41793 merged after #42777 moved shared/structural-value-equality into
shared/collections, so six renderer imports point at a missing file and every
test shard fails to load them.
2026-11-15 12:04:50 -08:00
e516efcb61 Keep finished agents marked Done while their panes sleep (#41793)
* Remove finished-agent retention and keep sleeping status with panes

* Fix legacy pane projection and sleeping-agent regression fixtures

* Cover sleeping display-only status in parked terminal coverage

* Keep pane conversation identity only across continuous agent observations

* Remove duplicate import and keep sleep fixtures type checked

* Preserve conversation boundaries across replays and inherited clocks

* Cover conversation map stability across a large status batch

* Bound conversation capacity work to changed pane owners

* Pass capacity classifiers their layout input directly

* Finalize conversation provenance once per status batch

* Keep paired snapshot tests focused on observable outcomes

* Type the conversation provenance test fixtures explicitly

* Consolidate agent session imports after main merge

* Preserve provider replacement evidence when resetting conversation continuity

* Separate retention deletion from sleeping agent display

* Restore migration row bucket comparison

* Skip unusable pane identities before title ownership lookups

* Refresh conversation titles when their owning pane changes

* Update completion test sound mock after module move

* Keep finished agent status with sleeping panes and update agents by key

* Forward sleeping pane status and scope paired tab updates

* Preserve keyed selectors and compiler invalidation through final checks

* Keep bound store hooks visible to the compiler and measure published links

* Install provenance measurement before factory capture

* Use a static store hook and name presentation metadata once

* Preserve keyed freshness and atomic agent status publication

* Preserve remote folder dashboard regression after suite retirement

* Model pane wake before live identity reobservation in regression

* Share keyed row transactions and finish main integration

* Align runtime fixtures with keyed agent status reads

* Keep status projections addressed and activity summaries current

* Retire empty workspace status summaries after their source leaves

* Use the standard store fixture for parked terminal graph tests

* Align status fixtures with app types and a fixed test clock

* test: capture stored terminal render failure details

* Read terminal titles and completion prompts from keyed agent status

* Compose foreground agent state in worktree test stores

* test: complete lineage list agent state fixture

* Keep sleeping completion when retiring process authority

* Run durable agent-hook status tests with Node SQLite

* Preserve sleeping Done when renderer retires its process

* Update terminal fixtures for process retirement

* Read sleeping pane resume sources by key

* Bind sleep capture fixture to its live panes

* Pass the pane tree to the sleep capture fixture graph

* Preserve accepted completion identity through sleep

* Pass a nullable title to the status classifier

* Resolve accepted host status once and preserve delivery order

* Drop published sleep status when its host owner retires

* Use the supported remote completion test envelope

* Preserve published activity when filling completed status

* Update imports for main module moves

* Remove an unreachable native chat owner comparison

* Read host and unread card inputs by address instead of per root

A host-scoped sidebar card re-read every root's row and acknowledgement on
any store write, so one card with many sleeping roots paid for other
workspaces' updates. Card host filtering, the summary pill's unread state
and the workspace unread gate now first compare the workspace's projected
row revision and the changed keys of the tab-keyed layout/PTY maps and
the acknowledgement record. Only an addressed change re-reads the roots.
Acknowledge and unacknowledge annotate their writes so readers can skip
keys they don't show.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Correct sleeping pane test imports after module moves

* Combine duplicate subscription test imports

* Record addressed acknowledgement deletions on row drops

* Type subscription fixture key set for string readers

* Keep accepted updates complete after observer failures

* Seed the send picker with its SSH terminal identity

* perf(layout): skip unused normalization reports in pane commands (#42615)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* Clarify cleanup pinning and ignored-file protection (#42616)

* test: remove 55 low-value suites and assertion driver (#42605)

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

* fix(skills): skip Claude sync scratch folders during skill discovery (#42610)

Claude Code parks replaced skill copies in `.trash` and in-progress syncs
in `.staging` under its skills roots. Discovery walked into both, so each
synced skill showed up as duplicates. Prune those directory names in the
native file walk and in the generated WSL `find` command.

Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>

* test: remove 78 low-value suites and trim redundant cases (#42621)

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

* Keep a host card's unchanged agent rows after filtering out another host's pane

Re-lineage clones every row once a host-scoped card filters a row out, so an unrelated pane's status write rebuilt the card's row array and re-ran its signal and rollup memos. Reuse unchanged rows as main's sidebar row selector does.

* Seed the Copy Session ID shortcut test through keyed agent status

* Seed the graph payload partition test through keyed agent status

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: Jinjing <6427696+AmethystLiang@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@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: BingZ <zzb@gxsmjx.com>
Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>
2026-11-15 11:38:40 -08:00
Kelvin AmoabaandYiminnn bf2946164e fix(browser): stop a background paired-desktop page from covering the selected tab (#42787)
Carries over #40602.

Co-authored-by: Yiminnn <47510889+Yiminnn@users.noreply.github.com>
2026-11-15 19:35:50 +00:00
Kelvin Amoaba b8c8d54461 fix(browser): run unwatched serve browser tabs at a low frame rate (#42660)
* fix(browser): run unwatched serve browser tabs at a low frame rate

Offscreen serve pages composite on their own 60 fps frame clock even with
painting stopped, so every unwatched tab kept the GPU and its renderer busy.
Tick unwatched pages at 5 fps and raise them to full rate while a command,
capture or screencast lease needs frames, keeping full rate for 20 s after so
an agent's next step finds a fresh frame.

* fix(browser): reapply the serve page frame rate when a navigation commits
2026-11-15 19:18:59 +00:00
Pr1pandNeil a30dd7d19d fix(new-tab): don't probe the Android emulator's modem listener (#42774)
The new-tab page sends GET / to every scanned workspace port to read a page
title and favicon. The Android emulator's modem simulator is a qemu listener
on [::1]; that request leaves it in CLOSE_WAIT spinning a CPU core at 100%
until the emulator is restarted.

Skip the page probe for qemu-system-* processes bound to [::1]. qemu hostfwd
listeners bind IPv4 and keep being probed, and the prefix match survives
Linux's 15-character /proc/<pid>/comm truncation.

Fixes #42748

Co-authored-by: Neil <neil@stably.ai>
2026-11-15 11:17:53 -08:00
Neil 9b3737e5e0 refactor: organize shared agent accounts and usage owners (#42781) 2026-11-15 11:06:48 -08:00
MananandNeil 88a5d07046 fix(terminal): hang up an idle login shell on close instead of waiting out the SIGKILL grace (#42767)
On macOS the PTY root is /usr/bin/login and the shell is its only child.
Close sends SIGTERM and SIGKILLs survivors after a 2 s grace, but an idle
interactive zsh or bash ignores SIGTERM, so every idle terminal close took
about 2.1 s.

When the login wrapper's only descendant is a bare interactive shell (argv0
is a shell, args are only -l/-i/--login), the shutdown sweep now also sends
it SIGHUP with its identity-checked SIGTERM, so it exits through the normal
hang-up path. Shells running scripts or -c commands, shells with other
descendants, exec'd programs, unreadable commands, daemon-wide disposal and
retired-daemon recovery keep the SIGTERM grace unchanged, so agents still
run their exit hooks.

Fixes #42675

Co-authored-by: Neil <neil@stably.ai>
2026-11-15 10:52:35 -08:00
Neil ad29834900 refactor: organize shared concurrency and text primitives (#42777) 2026-11-15 10:28:45 -08:00
mia-ambrandNeil c67c705a4c docs: correct Windows verification and contributor script paths (#42766)
Align the Windows update harness docs and CLI help with the implemented
cold-restore/survival profiles, document the daemon-log reader, drop
links to unavailable plan/postmortem docs, and fix relocated script and
source paths in freeze-repro, process-host, and crash-survival docs.
Remove stale French Windows RC claims in favor of the stable latest asset.

Fixes #42765

Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>
2026-11-15 10:22:04 -08:00
Kelvin Amoaba cfd166aec5 fix(browser): stop full-page and emulated-scale screenshots repeating the page in tiles (#42667)
* fix(browser): stop screenshots tiling when the page is taller or denser than its pane

* fix(browser): plan full-page tiles on the pixel grid and read the ratio over CDP

* fix(browser): stop full-page tiling once a tile reaches the page end

* fix(browser): bound the whole full-page capture and its measurement
2026-11-15 18:17:53 +00:00
Brennan Benson 106710b159 fix(native-chat): Claude steers interrupt foreground work and stay in the same turn (#42520)
* fix(native-chat): make Claude steers interrupt instead of waiting on foreground work

A steer sent while Claude ran a foreground subagent or command sat at
"Sending…" until that work finished, because Claude only reads queued
input between tool calls. Steers now move foreground work to the
background first, then send with priority "now" so Claude ends the
current step and answers the steer right away.

* fix(i18n): match the Steer hint fallback to en.json

* fix(native-chat): steer Claude without killing a subagent's work, in the same turn

Background only top-level foreground subagents and commands, by tool
call, so a subagent's own commands stay with it. A step the steer cuts
short now leaves its turn open for the steer instead of ending it as a
finished turn; without a pending steer, an aborted step reads as
interrupted.

* fix(native-chat): keep a steered turn's Interrupted ending if the steer never runs

A Stop or idle before the steer's own cycle starts now ends the held turn
as the cut-off step ended, not as completed, and the cut-off step's root
reasoning ends with it while a subagent's streams continue.

* fix(native-chat): let real Claude results keep their own ending after a steer hand-off

The held ending of a step a steer cut short now applies only when idle
ends the turn without a verdict, and the hand-off ends only root text so
a subagent's streamed text continues.

* fix(native-chat): tidy the Claude steer hand-off

Re-check the turn after the background wait so a turn that ended becomes
a plain send with its own choices; forget only the root stream's blocks
at the hand-off; end leftover work at the real settle time; await the
fake background route.

* fix(native-chat): end a subagent's streamed text when a steered turn goes idle

A steer hand-off keeps subagent text open, so the turn's real settle now
ends and drops it; an idle end takes the held verdict at idle's time.

* fix(native-chat): keep a subagent's text streaming when a new prompt supersedes the turn

Only a real settle ends and drops leftover streamed text; superseding a
turn ends reasoning only, as before, so a background subagent's block
keeps its row.
2026-11-15 10:12:23 -08:00
Neil 31b23b84d4 refactor: organize shared input and terminal protocols (#42772)
* refactor: organize shared input and terminal protocols

* refactor: align input cleanup validation selectors
2026-11-15 09:33:02 -08:00
BingZandNeil f7a7305c76 fix(plugins): return null workspace context when no workspace is focused (#42762)
On the start screen the stored session names no workspace, but
resolveActiveWorktreeContext still borrowed the first renderer tab with an
active pane, so workspace.readContext and workspace.changed reported a
workspace the user had not opened. The stored session always exists when a
store does, so the session's focus is now the only answer and the start
screen reads null, matching the plugin API contract.

Fixes #42752

Co-authored-by: Neil <4138956+nwparker@users.noreply.github.com>
2026-11-15 09:16:13 -08:00
Brennan BensonandClaude Opus 5.5 eec5ebfe7a Free idle hidden desktop browser pages (#40653)
* Bound retained desktop browser guests by idle page count

* Type the browser guest retention regression fixtures

* Update document visibility after React commits

* Preserve current page recency and private document audio protection

* Avoid observer handshakes on child-frame load completion

* Complete the browser certificate replay WebContents fixture

* test(browser): complete guest protection fixture contracts

* fix(browser): retain the saved page catalog after guest reclamation

* fix(browser): preserve inventory ownership and global tab order

* fix(browser): scope inventory ownership work to browser workspaces

* Complete browser cleanup mocks after main sync

* fix(browser): retain edited documents and earlier history

* test(browser): complete document guest history fixture

* test(browser): assert history transition publication

* Restore reclaimed browser pages before commands and retain active owners

* Fix native idle-page eligibility and retain canonical reclaim order

* Keep browser inventory recovery fixtures faithful to saved pages

* fix(browser): retry eligible guests and avoid default debugger acquisition

* fix(browser): publish owner protection before discard retries

* test(browser): isolate protection acquisition before release

* test(browser): keep preview input fixture native-shaped

* Fix browser lifecycle fixture typecheck coverage

* fix(browser): finish guest discard admission before replying

* fix(browser): retire only the hidden page surplus

* fix(browser): retain cold recovery and modal ownership boundaries

* test(browser): complete successful native load before stopping

* Preserve browser page ownership across profile replacement

* Typecheck connected browser profile fixtures in their owning project

* Import browser disposal operations from their owning module

* WIP: interrupted command-target repair (stopped 11-14 for Brennan's no-cap ruling)

* Keep browser commands on the selected page after idle discard

* Limit page-close routing cleanup to its current references

* Guard page close against unrelated routing work

* Type the routing retention probe by its owner

* Add a Memory Saver setting for idle browser pages

Settings > Browser gets a Memory Saver switch (on by default, kept under the
existing kill-switch key) and an "Unviewed for" hours field (1-168, default 4).
Main re-keys idle deadlines live from when each page hid, so turning it off
stops pending frees and changing the hours applies without a restart.

* Let main alone gate Memory Saver and key idle deadlines by hide time

Changing the hours no longer walks every tracked page: the heap is keyed by
when each page hid and the shared offset is applied at the head. The renderer
no longer vetoes on the setting or reinstalls its listener on toggle, so a
toggle sends no visibility replay and can't strand an in-flight page.

* Document the fixture casts in the browser inventory publication test

* Await the now-async window graph sync in browser inventory tests

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-11-15 09:14:11 -08:00
Neil a616643671 refactor: organize shared project and workspace modules (#42757) 2026-11-15 08:35:07 -08:00
NeilandClaude Opus 5.5 2cd3f168f9 fix(orchestration): add --run to orchestration inbox (#42747)
`orca orchestration inbox` rejected --run even though the
orchestration.inbox RPC already accepted a run filter. Accept --run in
the CLI spec and forward it; the server now refuses an unknown Run
(hinting when a run:<id> mailbox address was passed instead of a bare
id), and the CLI refuses rows from other Runs when an older server
ignores the filter.

Fixes #42732

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-15 07:45:48 -08:00
Neil 10c0892c21 fix(native-chat): show plugin skills as chips before the first reply (#42754)
Plugin skills were scanned under their bare folder/frontmatter name, so an
unclassified 'fwx:discuss' command from Claude's pre-init catalog never
matched a scanned skill and was inserted as plain text. Carry the plugin
name on scanned plugin roots/skills and name them '<plugin>:<folder>' in
the picker, the way Claude invokes them. The plugin name is the
installed_plugins.json id prefix; native discovery also reads the manifest
name from <installPath>/.claude-plugin/plugin.json. WSL discovery keeps the
id prefix and launches no extra wsl.exe.

Plugin workflows (commands/*.md with no SKILL.md) are out of scope.

Fixes #42746
2026-11-15 07:40:30 -08:00
NeilandClaude 2f6dd43ff5 feat(editor): add PowerShell to rich markdown code block languages (#42753)
List PowerShell in the code block language picker and register the
highlight.js PowerShell grammar with the rich markdown lowlight instance,
since lowlight's common set omits it.

Fixes #42744

Co-authored-by: Claude <noreply@anthropic.com>
2026-11-15 07:35:13 -08:00
Neil c3a681f5b1 refactor: group process and platform modules by ownership (#42742)
* refactor: group process and platform modules by ownership

* refactor: preserve moved module workflow routing

* refactor: preserve native loader and embedded bundle origins
2026-11-15 07:23:28 -08:00
Iwo SmuraandNeil 684d36274a fix(stats): keep large usage costs inside their overview tiles (#42725)
The Stats & Usage summary grids switched to four columns on xl viewports,
but the settings column is capped at 784px, so each tile was ~160px wide and
a five-digit estimated cost spilled out of its tile. Use container queries
(@md:grid-cols-2 @4xl:grid-cols-4) so four columns only apply when the pane
itself is wide enough, and let StatCard values wrap as a last resort instead
of overflowing.

Fixes #42724

Co-authored-by: Neil <neil@stably.ai>
2026-11-15 07:09:46 -08:00
NeilandClaude 18e24b2e79 fix(orchestration): make agent-facing CLI hints point to commands that work (#42745)
- Mention routing notes and terminal_busy / chat-session refusal pointers
  now include the required --subject and --body flags for
  `orchestration send`.
- `orchestration send --to <ctx_id>` refusals for a bare id of an existing
  Dispatch now suggest `--to dispatch:<id>`.
- `orchestration check --help` documents its own --format meaning instead
  of the screenshot image-format text.
- Remove an unused agent-reference test fixture.

Fixes #42735

Co-authored-by: Claude <noreply@anthropic.com>
2026-11-15 06:59:18 -08:00
brthapa-maitriandNeil ee2009e271 fix(jira): treat Atlassian-hosted sites saved as self-hosted as Cloud (#42728)
A Jira Cloud site connected through the self-hosted form was saved with
authType 'server', so issue lists hit /rest/api/2/search, which Atlassian
removed from Cloud and now answers with 410. Resolve the auth type from the
host: atlassian.net and jira.com sites are always Cloud, both when
connecting and when reading saved site records. Emailless server records
(PAT sites) stay server so they are not dropped and their tokens orphaned.

Fixes #42722

Co-authored-by: Neil <neil@stably.ai>
2026-11-15 06:33:58 -08:00
NeilandClaude 8c958bb9a7 fix(native-chat): stop the Add to chat offer from blocking composer focus (#42729)
The selection Add to chat popover is a visible overlay, and the composer
refuses focus while an overlay is open. After selecting transcript text,
a Cmd/Ctrl+V or plain typing was routed into the composer's text but focus
stayed on the transcript, so later keystrokes such as Space scrolled the
transcript instead of reaching the editor.

Mark the popover content data-focus-behavior="page" so overlay focus
ownership ignores it (it never takes focus itself), and drop the flushSync
close that only existed to work around the same guard.

Fixes #42714

Co-authored-by: Claude <noreply@anthropic.com>
2026-11-15 06:23:35 -08:00
Neil fce620d991 refactor: organize shared telemetry, education and platform contracts (#42713)
* refactor: organize shared telemetry, education and platform contracts

* fix(agent-status): retain transcript polling when agent IDs resolve
2026-11-15 05:44:50 -08:00
Noah RijkaardandNeil 942b5ff553 feat(terminal): add a rebindable Copy Session ID shortcut (#31814)
Adds a terminal.copySessionId action under Terminal Panes in Settings >
Shortcuts (no default chord). Pressing it in an agent pane copies the
same session ID as the context menu item via a shared resolver, swallows
held repeats, and shows a localized toast when no live session exists.

Fixes #29729

Co-authored-by: Neil <neil@stably.ai>
2026-11-15 05:03:18 -08:00