Repository navigation
Conversation
… it holds, and later moves don't re-send them (stablyai#30608) * fix(ssh): relinking a stopped server keeps the groups it holds, and later moves reference them Relinking found server-held groups only through the rows in them, so an empty group renamed on the server was re-imported and refused. Delta moves re-exported a linked group whenever a new row sat in it, so a group renamed on the server refused the next Move. Both now leave the server's group as is. * fix(ssh): delta moves keep the server's record of groups the first move imported * test(ssh): cover a deleted moved group and tidy the relink group fixtures * fix(ssh): resume an older build's delta with the groups its manifest exported * fix(ssh): take no group-ownership evidence from server rows on another host * fix(ssh): a relink records only the groups the server's own list holds, with their host stamps
… late (dark, layout-owner build only) (stablyai#30560)
…blyai#29074) * feat(native-chat): choose which agents open as a chat Settings → Chat gets a Providers list. Each agent the host can open as a chat has a switch; off opens that agent in a terminal instead. Only choices the user makes are saved. Unchosen: Claude and Codex on, others on when their CLI is installed on this computer. The renderer's route and the host's launch-mode decision both read one rule, so the tab menu, shortcuts, CLI, orchestration and relayed phone launches agree. * fix: keep chat provider settings with the other chat settings types * test: route input carries this computer's detected agents * feat(native-chat): the + menu shows chat, disabled chat and CLI rows With Chat UI on, each agent whose chat provider is on gets a chat row, disabled with the reason where the workspace's machine can't run that chat (not installed, SSH, WSL, not on this host). Every other installed agent gets a row marked (CLI) that opens its terminal. With Chat UI off, or on an older server that opens no chats, the menu is unchanged. * fix: new tab menu rows use a launch variant instead of restyling the menu item * fix: name every route blocker the menu rows map * fix: menu row subscriptions render on the server too * feat(native-chat): open a chat agent's CLI from beside its + menu row A chat row whose CLI is installed on the workspace's machine gets a › beside it. Its side menu offers '<Agent> (CLI)', which opens that agent's terminal for this launch only; nothing is saved and the row itself still opens a chat. * fix: keep new tab agent surfaces authoritative * fix: handle unknown host support reasons * fix(chat): resolve provider choices from their owning computer * fix: include chat launch feedback translations * test(chat): use checked native inventory fixtures * chore: retrigger PR checks * test: pin pending chat row behavior * chore: retrigger PR checks after cancellation * Fix launch preparation lifetime and execution host startup * fix: keep agent menu choices bound to their launch * Simplify agent menu support reads and preserve explicit surfaces * Keep focused menu behavior coverage within its budget * Keep menu host fixtures typed as execution hosts * Assert notes payload on the current launch action * Combine agent session launch type imports * fix(chat): keep unchosen providers on without discovery * fix(chat): use workspace installation for menu defaults * test(chat): type provider routing fixtures against store contracts * fix(tab-bar): keep + menu chat rows clickable while the host answers Each menu open discarded the host's answers, so every open showed the chat rows disabled with 'Checking chat support' until the host replied. Answers now outlive the menu and refresh in the background, rows use this client's facts until the host answers, and a slow host no longer reads as a refusal. * test(chat): align shared launch fixture with private settings reader * fix(tab-bar): keep known + menu refusals and never treat a silent host as one Closing the menu dropped the workspace's agent list, which changed which agents the menu asks about and threw away the host's earlier answers, so a reopen showed a refused chat as clickable. A support read that could not answer was stored as "No chat on this host", disabling chats the client knows are supported. The menu now keeps the stored agent list while closed (only fetching waits for it to open) and only a real host answer relabels a row. The workspace's detection target is resolved where main already resolves it (QuickLaunchButton and the tab-bar runtime model) and passed in, so no new focus read is added. Tab-bar test mocks now stub the QuickLaunchButton exports the menu imports. * fix: sync runtime English catalog for agent menu * test(chat): give the structured-session RPC fixture the private settings reader The host now reads saved chat choices through requireStore().getSettings(); the shared RPC fixture only had getClientSettings, so the launch-mode reader saw no settings and create support refused Claude. The fixture now follows each test's client settings. * test(chat): pin create support's options argument and kept decline reason * fix(native-chat): + menu chat rows wait disabled for the host and share its answer across tab groups A chat the host refuses showed clickable for up to the 10s host read, then flipped to disabled, once per tab group. Rows now stay quietly disabled until their host answers (or the read times out), and every tab group on the same workspace host reads the same answers. * fix(native-chat): + menu rows also wait before the host's chat list loads, and on every reopen of a host that could not answer Asking the host no longer waits for its chat list, so Claude and Codex no longer show clickable, then disabled, while it loads. A read that times out releases rows for that open only, so the next open waits again instead of flipping on a late refusal. Agents not installed skip the read. * fix(native-chat): typed + menu search puts a launchable row ahead of a disabled chat In a WSL workspace, typing an agent's name and pressing Enter highlighted its disabled chat row, so nothing launched where main opened the CLI. Search now ranks rows that can launch first. * fix: preserve provider choice and cached native inventory * feat(native-chat): hosts publish each workspace's chat availability as one stream A new negotiated stream, agentSession.subscribeAvailability, advertised as agent-session.structured.availability.v1. The host keeps one in-memory resource per workspace, shared by every reader, that asks the existing createSupport check for every registered agent in that workspace and publishes only complete generations. It answers again when a launch setting changes, after an explicit agent refresh, and (for answers that could not be checked) when a workspace appears; the last reader leaving retires it. A check that could not answer is unknown, never a refusal. agentSession.agents stays static and unchanged. * feat(native-chat): + menus read a publishing host's workspace availability The renderer holds the workspace's availability stream while its tab bar is mounted (not while a menu is open), shared by every tab group, with runtime and contact fences. It exposes a readiness model, { ready: false } | { ready: true; byAgent }, from useHostStructuredAgentAvailability and the menu hook. On a host advertising the capability the menu reads its answers from that model and makes no per-agent createSupport reads; an older host keeps the per-agent reads, answer cache and deadline unchanged. Row building and presentation are unchanged. * fix(native-chat): workspace availability checks only what the shown workspace needs Only the shown worktree (or an open menu) holds its host's availability stream, and only with Chat UI on. Display checks no longer force a fresh login shell for an agent that isn't installed, so Claude and Codex answer without waiting on it. A change to the Windows runtime default or a project's WSL choice re-checks, and a workspace still on its first check answers again when one appears. * fix: keep main's bytes for the create-support file the merge reformatted The pre-commit formatter rewrapped main's unformatted lines during the merge and pushed the file past its line limit. * test: type the + menu rows hook mocks without assertions * test: match the availability stream's types in three tests * test(cross-version): declare the workspace availability stream in the agent-session surface * feat(native-chat): + menu holds a placeholder until its agents are known, and learns them while the workspace is shown * feat(native-chat): + menu reads one agent health snapshot per machine The + menu no longer checks chat availability per workspace. Each Orca runtime keeps one agent health snapshot: every agent's location rule on this machine plus a typed observation of its installed release, probed in the server's own home. It is taken on first use and replaced only on a launch setting change, an explicit agent Refresh (after a fresh shell capture), or a restart, and kept without readers. - The version probe reports supported, unsupported or inconclusive; a launch still refuses an inconclusive run, while the snapshot reads it as unknown rather than a refusal. - agentSession.subscribeAvailability takes no params and publishes that snapshot; a replay that closes the transport releases its reader. - The renderer keeps the snapshot per host, runtime and pairing, through lost contact and across workspace switches on the same host. - The menu hook drops the per-open createSupport display reads, their deadline and answer cache; older hosts show client facts and the click's admission, unchanged, states any refusal. * fix: a failed shell recapture still publishes the agent health snapshot * fix(native-chat): capture fresh shell facts for every health snapshot * fix(tab-menu): share settled agent actions across menu surfaces * fix(tab-menu): preserve loading row line height * feat(native-chat): the whole chat row opens its CLI submenu Hovering a chat row in the + menu, or pressing ArrowRight on it, now opens the "<Agent> (CLI)" submenu to the right instead of needing a separate chevron button. Clicking the row, or Enter/Space, still starts the chat and closes the menu; hosts pass closeMenu since a submenu trigger never closes the menu on its own. * fix(tab-menu): line the CLI submenu up with its chat row * feat(native-chat): a chat row's terminal button opens its CLI Clicking a chat row in the + menu starts a chat. Hovering the row (or pressing ArrowRight on it) reveals a terminal button at its right edge that opens the agent in a terminal instead. This replaces the submenu, whose chevron made the row look like it only opened a submenu. The button's click bubbles into the row, so the menu closes as for any item and the hosts no longer need a closeMenu callback. * fix(native-chat): the chat row's terminal button handles drags, arrows and screen readers A press on + dragged onto a chat row's terminal button and released there now opens the CLI: the row reads whether the pointer or focus is on its button instead of the click target, which a drag-release never sets. Arrow, Home and End keys pressed on the focused button move between rows as they do on the row. The row is announced as "Start a <Agent> chat" instead of including the button's label. * fix(native-chat): chat rows read "Start a chat with <Agent>" "Start a OpenCode chat" was ungrammatical for agents whose name starts with a vowel.
…desktop data link (stablyai#30607) * fix(mobile): recover a relay that died during an app switch in one short probe, and retry an attach timeout through the director once Refs STA-11130. Co-Authored-By: Claude <noreply@anthropic.com> * fix(mobile): redial the same cell once on attach-timeout instead of re-resolving Aligns with the relay step 6 design: reason string attach-timeout, no director re-resolve (the director names the cell the host is seated on). Co-Authored-By: Claude <noreply@anthropic.com> * fix(mobile): probe decisively only on a real background return, 4s, and jitter the attach-timeout redial Co-Authored-By: Claude <noreply@anthropic.com> * test(mobile): replace type assertions with a typed fake in the relay probe tests Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
…x config strips command environment variables (stablyai#30406) * fix(codex): a Codex chat's commands keep its Orca identity under a restricted shell environment policy Codex rebuilds each command's environment from the user's shell_environment_policy, so inherit = "core" dropped the chat id, marker, launcher and instance pin Orca gives the app-server. orca orchestration then could not tell which chat was calling. Every thread start and resume now sends those values as shell_environment_policy.set overrides, which Codex applies after its filter and merges per key. * fix(codex): accept the launch's read-only delete list * test(codex): a chat without visuals gets no visuals config, identity overrides aside * refactor(codex): a chat's command env is what the seal wrote, not a second key list The seal now returns the session values it set (identity, a registered worker's own handle, the CLI launcher, the instance pin and the blanked remote selectors) as commandEnv, and the composed launch carries it. The Codex acquire sends that plus the chat's scratch folder from its own visuals. This deletes the hand-kept key list, the handle prefix test and the delete-list re-check, so a key later added to the seal reaches each command too, and an inherited worker handle the seal never named no longer does. * test(codex): wait for the shell command's turn to complete before the cold resume
…lyai#30416) * fix(agent-status): never drop a running helper for lack of room Helper state (subagents, background tasks and commands) was bounded by counts and byte budgets that refused or evicted live records once hit, so the agent kept working while Orca listed it wrong. Every one of these records already dies with its session, pane or chat, so the count and size caps go; identity-history, transcript and transport bounds stay. * fix(agent-status): keep the status field limit, measured without the subagent roster The previous commit dropped the 256 KiB status check outright; it still guards every other field. Only the roster, whose rows are bounded individually, is left out of the measurement. * fix(agent-status): fit helper previews and walk lineage once * fix(agent-status): skip an oversized Pi row instead of stopping, and parse every envelope the store can write Review follow-up: one long row no longer hides later rows that still fit the request, and the transport envelope's reader no longer refuses a size its writer now produces.
* fix(sidebar): Sleep stops a workspace's native chat agents Sleeping a workspace stopped terminals and browsers but never its structured (native) chats, so their claude/codex process kept running, and Sleep was disabled for a workspace whose only content was a chat. Sleep now stops each chat's agent through a new agentSession.sleep (the host's close, keeping the tab), routed to the host recorded on the chat's tab; an older host without the capability fails Sleep with a clear message. The Sleep menu counts a chat. * Fix native chat Sleep ownership and delayed teardown * Stop native chat sessions when sleeping their workspace * Fix sleep teardown wake and host-qualified receipt races * Verify mobile Sleep sends the selected desktop host * Keep worktree Sleep RPC within method file limit * Use sleep intent as attempt identity * Gate workspace sleep on structured client support * fix(native-chat): preserve sleep stop receipts and send timing * fix(native-chat): keep superseded workspace sleeps quiet * fix(native-chat): ignore superseded failed sleep stops * Fix sleep lifetime test resolver types * Accept workspace Sleep for already ended chats * Sleep native chats through the shared no-toast sleep core Main split runSleepWorktrees into sleepWorktreesReportingFailures (no toast) and a toasting wrapper. Move the native chat stop, host routing and typed failures into the shared core so the WSL move's Sleep request also stops chats, and report every workspace not verifiably slept (an unconfirmed chat stop included) in failedWorktreeIds. Fold one ACP adapter field onto a line so main's file plus this PR's ownership check stays within the file length limit. * Import isStructuredTab from its shared module * fix(native-chat): leave chat rows to their host through Sleep Sleep's worktree status drop deleted the chat's row, which the host feed then rewrote, so the outcome depended on which write landed last. Open chat tabs own their rows (same rule as terminal-tab close), and a chat only makes a workspace sleepable while its host holds the agent's process. * fix(native-chat): judge the Sleep receipt by facts, not lane occupancy A running chat's stop queues the provider's own requested-close end report onto the chat's lane, so the lane-occupancy check read every proven stop as superseded and dropped its receipt. Later work now counts only once it exists: a moved fence, a child, delivery, child work, a pending submission, a pending create or attach, or an open dispatch. * test(native-chat): report a fake close's end only for an acquired child * Preserve canonical workspace identity when sleeping * Propagate optional sleep host through file command contract * Move the deliberate-sleep PTY tests to main's spec file Main grouped the PTY connection suites (stablyai#29193) and moved this file to a .spec.ts registered from pty-connection-spawn-and-startup.test.ts. Taking main's copy here keeps the following main merge free of a modify/delete conflict; this PR's test changes are reapplied after the merge. * Reapply Sleep's teardown waits to the moved PTY spec; keep acquisition under its line limit The deliberate-sleep spec moved in stablyai#29193; this carries the PR's changes into it: a pane mounted after wake waits for issued teardown, an unqualified pane waits for every same-id host's teardown, and a queued startup waits too. The merge left claude-structured-session-acquisition.ts over 300 lines. resolveClaudeAcquisitionError, which already builds the exit-unproven error, now decides whether to release the acquisition. * test: keep the Sleep receipt checks on the closed handle and kept record * fix(sidebar): a single-workspace Sleep failure says workspace, not workspaces * fix(sidebar): Sleep stops a workspace's chats on every host that has a copy * test: drop redundant casts from the Sleep flow test state * fix(sidebar): stop a workspace's chats after its terminals, leaving main's teardown owner check as is * fix: admit workspace chat stops before teardown waits * fix: retain browser teardown caller sequencing * fix: retire failed acquisitions after observed root exit
…xisting-only (stablyai#30609) A reserve cell keeps no activity rows and its ledger stamps last_activity_at, so the stranded rule read its connected hosts as unserved and re-placed them.
stablyai#30616) * fix(persistence): repo writes take the row's identity, never a bare id updateRepo(target, updates) now takes Pick<Repo,'id'|'connectionId'|'executionHostId'> (RepoHostTarget) and resolves the host from it, so a caller that holds the row can no longer drop its host and write a same-id sibling on another host. The id-only callers (project host setup controller and IPC handler, folder-to-Git upgrade watch, both clone paths, registration adoption, fork backfill) pass the row they resolved. Co-Authored-By: Claude <noreply@anthropic.com> * test: migrate the remaining Store.updateRepo callers to row targets Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
…blyai#30612) * plugins: name the shared catalog and manifest schema parts Pure refactor. Marketplace entry/header fields and the duplicate-category and duplicate-id checks become named pieces, the bounded catalog JSON read is split from schema parsing, and the v1 manifest object is named apart from its refinement, so a second catalog file and manifest version can reuse them. * plugins: share the plugin tree-identity check Pure refactor. Install integrity and instructional-content checks each inlined the full-hash-or-legacy-prefix comparison; both now call pluginTreeHashMatches. The tree limits are exported for reuse.
…blyai#30611) The guest will-navigate guard cancelled every file: navigation, so clicking a relative link on a file:// page did nothing. Allow file: -> file: when the current page is itself a local file; web pages still cannot navigate or redirect into the filesystem.
… (dark) (stablyai#30624) * fix(sidebar): Hide sleeping reads a paired server's own terminal report (dark) With ORCA_SIDEBAR_INPUTS=main, main's activity domain now lists a paired server's rows: their terminals and agents as the server's worktree.ps reports them, plus the browser and chat tabs this desktop's saved session holds there. A server out of contact keeps live terminals unverifiable. Switch-off builds are unchanged. * fix(sidebar): keep server activity out of the published paired rows Review: a strict field on paired rows could drop a whole row for an older reader; the activity domain already carries it.
* fix(native-chat): identify the owner when resume create is refused * test(native-chat): consolidate transcript refusal cases * test(native-chat): keep ownership coverage at the launch boundary * test(native-chat): update adoption fixture for early ownership admission * test(native-chat): align resume admission assertions with host ownership * test(native-chat): keep RPC coverage focused on ownership proof * test(native-chat): assert blank request presence before optional source * fix(native-chat): admit the actual CLI source during chat creation * fix(native-chat): satisfy source admission lint checks * test(native-chat): provide the waiting hook prompt * test(native-chat): isolate source admission fixtures from host homes * test(native-chat): canonicalize adoption fixture directories * test(native-chat): preserve early ownership admission in source stack * Resume CLI conversations directly in native chat * fix(native-chat): preserve resume refusal behavior and honest transcript feedback * fix(native-chat): retain import read evidence in typed result * Use the header test fixture record type * Consolidate launch owner imports and correct feedback assertion * Keep existing resume translation in runtime catalog * fix(native-chat): preserve admitted source and prepared execution home * fix(chat): retain source evidence for dispatched adoption * fix(chat): keep source capture within module limits * Keep deferred chat publication in the shared launch owner * test(chat): allow host evidence in create routing * Satisfy launch publication checks and scope slot cleanup * Preserve canonical tab membership on projected reorders * Use recorded launch ownership for surface recovery * Reserve deferred chat order through existing group actions * Pass the active layout owner into launch cancellation * Stop terminal writer before adopting conversation in chat * Guard chat adoption completion and satisfy static checks * Switch completed terminal turns without confirmation and test handoff * Recheck terminal writers on reserved adoption retry * Split session entry delegates and fix linted test types * Keep chat opening send below create module line limit * Require explicit terminal stop consent before chat adoption * Use direct process-table test mock * Fix chat adoption types and runtime localization * Localize terminal takeover refusal guidance * Keep CLI stop admission scoped to active source panes * Preserve ordinary history resume through attach * fix(native-chat): admit an unsaved transcript under a linked account folder A missing transcript stayed unresolved while its account root resolved through symlinks, so containment failed and a not-yet-saved conversation showed a load error instead of the unsaved notice. * Update adoption replay test for at-rest creation * refactor(native-chat): move the newest-item import window into its own module Keeps journal-legacy-import.ts under the line limit after the source identity observer landed on top of the streaming import. * fix(native-chat): restore the refusal error import lost in the main merge * Remove duplicate account resolver and condense terminal admission * Refuse conversation adoption until terminal writers exit * Keep conversation inspection within lint limits * Preserve conversation owner in launch feedback * Expect unavailable writer scan to refuse fresh adoption * Keep terminal conversation evidence tied to its execution * Correct writer continuity test refusal and event shapes * Prove original terminal writers exited before chat adoption * Scope saved conversation holds to their execution host * Keep child-inspection fixture verdict literal * Recheck awake saved conversation holds against their owned pane * fix: inspect owned Windows jobs independently of foreground naming * test: use braces in Windows inspection fixture * test: complete ordinary Windows terminal inspection fixture * test: explain partial terminal inspection fixture * fix: inspect the terminal a conversation is switched out of * test: separate OpenCode writer observation coverage * fix: tie terminal conversation links to the PTY execution The conversations a terminal reported now live on its PTY record for that incarnation, so moving the pane or running many other terminals can no longer drop them, and a link whose terminal can't be inspected refuses. - Each link keeps the agent process its hooks last proved; a live or suspended process refuses even when the terminal's tree is shell-only (detached tmux, nohup). - tmux-published events are never pane-owned, and an unmarked binding copy no longer revokes an earlier pane-owned proof. - A restored row or saved checkpoint is censused against the pane's current local PTY instead of requiring this runtime's observed identity. - The refusal now says Orca can't confirm the agent has exited and asks the user to stop what's running there or close the terminal. * test: drop the stub-host admission calls from OpenCode observation coverage * test: census suspended children through the real terminal host * test: drop writer checks the real-terminal flow already covers * fix: let prompt workers outside the terminal's session pass the exit check - The POSIX shell-alone proof skips a session leader below the shell and everything under it: a prompt's zpty worker has left the terminal, while suspended, background, nohup and disowned jobs still refuse. - Windows steady-state polls read the cached process table; every read that decides adoption or close stays fresh. - The pane header and tab menu point back to Session History, since their action is gone once the agent exits. * fix(terminal): count only Windows job members still linked to the shell A pane's job keeps every process it ever started, including daemons whose parent exited (an rc-started ssh-agent), so such panes never read as an idle shell and the chat switch refused forever. Walk descendants of the shell along live parent links within the job, for every shell; the Git Bash launcher keeps its bash-chain allowance and a child older than its parent still refuses. * Revert "fix(terminal): count only Windows job members still linked to the shell" MSYS exec and fork leave intermediate Win32 parents exited, so a live Claude started through a script (and 'sleep 60 &') became unreachable and the chat switch admitted while it ran. Back to the job census until a console-based design is measured. * fix(terminal): set aside detached daemons on a Windows shell-alone decision A pane's job keeps every process ever started in it, so an rc-started ssh-agent made a Git Bash pane read as busy forever and the chat switch refused. MSYS fork/exec leaves live work under exited Win32 parents, so parent links alone can't tell the daemon apart; it is the one that also left the console. On a scanChildProcesses decision read (switch admission, run-terminal close, close guards), a member neither linked to the shell nor attached to its console no longer counts. Polls and shell confirmation keep the job census. The console-list helper now forks through the process-host chokepoint. * refactor(daemon): name the process inspection options once Keeps the daemon router and terminal host within their line budget now that the options carry scanChildProcesses. * fix(terminal): set aside detached daemons only for adoption, never for close Closing a Windows pane kills its whole job, so a close check that ignored a detached process (gitk &, cmd /c start ...) would kill it without asking. Adoption asks whether anything is still in the terminal; close asks whether closing destroys something. Only the adoption read now sends ignoreDetached; close guards and the run-terminal sweep keep the job census. The win32 test also stops the ssh-agent it starts. * fix(native-chat): check the terminal being switched out of before opening its chat When the conversation already had a native chat, the switch hit the ownership conflict first and the header opened that chat, even while the terminal's agent (a later claude --resume) was still running. The switched-out terminal is now checked before the conflict, so a live one refuses as usual and one at rest still opens the owning chat. * test(terminal): assert close reads never call an ssh-agent pane idle A plain read can answer unverifiable while the eval's short-lived processes exit; either way it must not read as no-children. * fix(native-chat): check terminal writers before naming an owning chat for any switch Session History's Resume in New Native Chat carries no source pane, so with an earlier chat still holding the conversation it skipped the terminal check and the owning chat's first send would start a second writer beside a live terminal Claude. Every adoption request now runs the writer check before the owner conflict, as the no-owner path already does. * test(reliability): run the chat-switch writer oracles in the provider-ownership gate
…to find its sessions (stablyai#30469) * perf(session-history): remember each transcripts folder so Workspace scans stop re-reading files * refactor(session-history): keep the scope cwd table in its own module * test(session-history): type the hoisted read log without an assertion * test(session-history): type the collected session folders * chore(wsl-move): list the parse cache scope_cwds table * Keep scoped session selection current with cache authority * Consolidate scope cache imports in regression test * Run scope cache SQLite regression tests under Node --------- Co-authored-by: OrcaWin <alpha-eng@stably.ai> Co-authored-by: Orca Worker <orca-worker@localhost>
…yai#30631) * fix(sidebar): phone feed list leftovers before the flip (dark) Bump the feed's semantics version for P11's desktop facts; mark switch-on phone page bundles so packaging and orcad staging refuse them; show the page's last-listed rows before the first feed read; draw the host's pending creations; pass the listed host's own label into host sections instead of the running device's platform. * fix(sidebar): review fixes for the phone feed leftovers (dark) A dark orcad may stage the dark page (only a marker of an off switch is refused); the bundle tree writer leaves the marker; the pre-feed list keeps the view's filters; a creation names its project only outside that project's header and survives search when its project matches; the feed view and row labels read one resolved desktop platform. * fix(sidebar): second review fixes for the phone feed leftovers (dark) Test the off-switch marker check across switches; resolve the desktop platform once for rows and headers; a section search narrows keeps only matching rows.
…ile (stablyai#30605) * feat(files): guarded file writes that refuse to overwrite a changed file Adds an optional disk precondition to file writes: the writer compares the file on disk with an expected signature inside the write, serialized per file (path then inode), and refuses with a typed changed/missing result instead of writing. Remote hosts serve it as files.writeIfUnchanged behind a capability; older hosts get a client-side compare that fails closed, pinned to the probed runtime and connection. Undo of a delete and the MCP starter config use it to write only into a missing or identical file. Groundwork for guarding editor saves. * fix(files): pin guarded writes on the sending socket; keep refused undo; narrow not-found and pinning * fix(files): pin a guarded save to one pairing; keep only refused undo steps * fix(files): ENOENT by code only; resolve symlink chains fully; pin lost-reply recovery to the save's pairing
…and say why a capture failed (stablyai#30618) * fix(ssh): make room for the pre-activation snapshot, and say why a capture failed Every update left its own full state copy behind, and a failed capture its partial archive, so a small or quota-limited home filled until no update could snapshot. The capture now prunes every pre-* snapshot except the one the activation record names, removes its own partial archive on failure, and prints tar's error. Before stopping the running server, the activation checks that the snapshot fits, so a host without room keeps its server. Refs stablyai#30491 Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): the room refusal says what is free apart from what a prune reclaims It also stops claiming the recorded server is running, and a test pins that the record's snapshot survives both the room check and the prune. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): measure snapshot room before the last terminal recount, within a deadline A slow scan after the recount could let a terminal opened meanwhile be stopped. The scan now runs first, gives up after 12 s per du (UNMEASURED reads as unknown), and a lost probe channel leaves the room unknown rather than failing the update. The newest-mtime probe moves to its own module to keep the snapshot module under its line limit. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): pick timeout's syntax for the room probe, and pin its failure paths BusyBox before 1.30 takes `timeout -t SECS`; plain `timeout 12` disabled the probe there. Tests cover a timeout on the old-snapshot scan and a lost fence from the probe. Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
NarryG
force-pushed
the
feat/omp-thinking-controls
branch
from
October 19, 2026 08:38
7f17777 to
652a8a4
Compare
NarryG
changed the base branch from
review/omp-parity-base
to
review/omp-remaining-base
October 19, 2026 08:41
…ity; delete the deprecated owner resolver (stablyai#30626) * refactor(terminal): one PTY attribution for teardown, close and identity; delete the deprecated resolver pair - strict ssh: route: a proven direct listing beats a same-id server proxy; other contradictions refuse - terminal spawn's git branch takes the strict route when a host is named (rivals refuse) - resolveTerminalPtyOwner + isUnattributedTwinWorktree replace resolveRuntimeOwnerForWorktree in shutdown, pane and agent hibernation, and PTY identity; close logs unroutable runtime ids - delete getRuntimeEnvironmentIdForWorktree/resolveRuntimeOwnerForWorktree (legacy branch moved into the adapter), the alias golden test, the importer ratchet, its baseline and pr.yml step - hasHostEvidence reads the owner indexes instead of scanning repos and groups * fix(terminal): close and identity read spawn's attribution for unscoped remote ids (Opus P2) - an unattributed twin's unscoped remote: id is skipped and logged on close, never closed locally - setup terminals: identity and close follow spawn's runtime; raw-id teardown stays local - drop the last stale getRuntimeEnvironmentIdForWorktree mock (tests/e2e unit suite) * refactor(terminal): one ownership read in resolveTerminalPtyOwner * test(terminal): pin the close-path twin skip and the two-proxy direct listing (Opus P3) * test(terminal): keep the close-provider owner fixture's full union type (Codex P1)
…tablyai#30661) The director's cell-status refuses a step-5 reserve cell with a 404 {error: cell_in_reserve_mode}, so every monitor sample failed while any cell was reserve. Classify that exact refusal as reserve-not-probed, skip the refused migration reads, and judge the cell on probes, logs and the selector instead. Any other 404 still fails the sample.
…ai#30627) * refactor(ssh): derive a host-key policy from the target owner The verifier and decision took an isEphemeralRuntimeTarget boolean. They now take a HostKeyPolicy derived from the owner by one function, and the owner shape gets a named SshTargetOwner type. No behavior change: on-demand runtime VMs map to the ephemeral policy, everything else to standard, and every existing decision test passes unchanged in substance. * docs(ssh): separate owned-target hiding from the recipe-VM id prefix Hidden-ness is decided by owner in main; the runtime-ssh- prefix also carries per-provision semantics (fresh filesystem, VM cleanup), so only on-demand-runtime owners may use it. Comments only. * refactor(ssh): one id predicate for every Orca-owned SSH target Surfaces that hide owned targets or must never dial them from the renderer (startup re-dial, reconnect hints, automation host catalog, status bar, host registry, state broadcast) now ask isOwnedSshTargetId, which also covers the machine- prefix. isRuntimeOwnedSshTargetId keeps its narrower recipe-VM meaning (address and filesystem replaced every provision), so name retirement and VM cleanup are unchanged. No target uses machine- on main yet, so behavior is identical; tests pin both.
…rrives (stablyai#30628) An optional onStderr hook receives every raw chunk, so long commands can stream progress while still getting the bounded result. Capture, truncation and termination are unchanged; existing callers pass no hook.
…desktop (stablyai#30630) * fix(ssh): a cancelled or revoked host's browser command stops on the desktop orcad now tells a cancel-capable desktop when the CLI gave up on a forwarded browser command, and a running host command on the desktop checks revocation before every side-effecting step (each helper CDP message and each direct input dispatch). * fix(browser): keep local commands out of a host's scope, gate each helper on its proxy * fix(browser): hold a host-driven page at its input sinks instead of per call site * test(browser): checked mock types in the host revocation suites * fix(browser): refuse only a host's own traffic, never the desktop's page commands * fix(browser): a host's proxy never keeps its scope, and a revoked click still lets go of the button * fix(browser): a revoked command may release only what it pressed, on every input path * fix(browser): refuse every revoked step, and let the desktop release what a revoked command left held * fix(browser): retire a revoked command's helper daemon before letting go of the page; release held input with a lease, synced pointer state and the focus guard * fix(browser): deferred proxy input keeps its hold, releases count once they land, and a revoked helper is retired at once with its routes kept * test(browser): keep the mock page cast on its SAFETY line
…through a symlink or subfolder shows its workspaces (stablyai#30657) * fix(projects): a project added from a paired client or the CLI through a symlink or subfolder shows its workspaces * fix(git): keep a literal backslash in a POSIX checkout root
…ommand (stablyai#30635) * test(managed-server): byte budget for every POSIX host-op command Builds each exported POSIX host-op builder at realistic maximal arguments and asserts the raw command stays under 6 KB, under the 8,000-byte compression threshold. Ops already over budget are listed with ceilings that may only shrink until they move onto the host script (round 4 R4-10). Co-Authored-By: Claude <noreply@anthropic.com> * test(managed-server): budget the wrapped, fence-guarded command on Linux and macOS Measures what a gateway sees on the plain fallback (fence guard + uncompressed POSIX wrapper), takes the larger of Linux and macOS, launches the SSH way (nohup), and also discovers the top-level src/main/ssh builders. Exports three builders that are sent as their own exec request so they can be measured. Co-Authored-By: Claude <noreply@anthropic.com> * test(managed-server): count the stdout fence; scan ssh transport subfolders Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
…restart (stablyai#30587) * fix(ssh): stop a closed bridged relay tab's shell from every close route, across restarts The renderer's tab-close kill tombstoned a bridged lease before the bridge's own close ran, so the old relay's shell kept running. Route it through the bridge like the runtime kill. Persist a pending bridged stop on the lease's pendingKill so a restart before reconnecting still sends it, and list such leases for the stop without serving them as panes. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): end a closed bridged tab's lease when its stop can't be proven Clearing the stop left the lease bridgeable, so a process that took the recycled relay id was served under the closed tab. Retire it as any closed tab's lease is. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): keep a closed bridged tab closed across restarts even when its stop can't be fenced An unfenced legacy stop now gets one try and ends the lease, a repeated close keeps the first record's age, and the Docker bridge spec covers a close while disconnected followed by a restart. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): re-list after abandoning a bridged stop so the closed pane drops Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): route CLI and worktree stops of a bridged tab through the bridge, and keep a repeated close fenced to the first identity * fix(ssh): refuse sleeping a bridged pane, settle bridged stop-and-wait on the relay's exit proof, and never attach a pane closed mid-attach * fix(ssh): settle a bridged stop-and-wait on the exit proof alone, for every overlapping waiter * fix(ssh): keep an unfenced bridged stop's pane tracked for its trailing exit report, never serving it * fix(ssh): keep awaiting a bridged stop's exit after a failed reply, never name a retired pane to the relay CLI, and require a real empty listing in e2e * fix(ssh): tell the runtime a bridged terminal exited once its stop-and-wait is proven * fix(ssh): keep an unfenced bridged stop's exit observable after its reply fails --------- Co-authored-by: Claude <noreply@anthropic.com>
…crawl (stablyai#30704) * fix(ssh): keep SFTP uploads pipelined so a high-latency link doesn't crawl ssh2's WriteStream acks each chunk before writing the next, so uploadFile moved one 64 KiB chunk per round trip: about 200 KB/s on a 300 ms link, which made the ~90 MB managed-server deploy (Node runtime plus orcad bundle) take over seven minutes. A 2 MiB write buffer lets ssh2's _writev keep that many bytes of WRITE requests in flight, as fastPut's default does. Co-Authored-By: Claude <noreply@anthropic.com> * test(ssh): type the fake session's stream options with ssh2's WriteStreamOptions Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
… refuse a reserve roll target by name (stablyai#30707) * fix(relay-ops): let same-cap rolls run while another cell is reserve; refuse a reserve roll target by name The director refuses cell-status for a step-5 reserve cell with a 404 {error: cell_in_reserve_mode}. The same-cap headroom check read it for every general cell, so any roll failed while c7 was reserve. A reserve cell that is not being rolled now counts as no free slots (its seats are not in the database) and is listed with reserve: true. A reserve roll target, in the headroom check, the same-cap job's first cell-status read and the candidate tool, now fails with the design's procedure: flip it to admitMode=db first, roll, then flip it back. Any other 404 still fails. * fix(relay-ops): pin the headroom gate's imports as evidence code; name both cells in a two-cell reserve refusal
…d runtime-terminal-inspection (D2, D3) (stablyai#30684) * refactor: de-barrel remote-runtime-client and runtime-terminal-stream (D2) importers name the defining modules; the two vi.mock factories that named a re-export move to the defining module (pty-id spy) or drop the dead key (RemoteRuntimeClientError) * refactor(renderer): de-barrel runtime-terminal-inspection (D3) importers name runtime-terminal-input-recording, runtime-terminal-verified-input and shared/remote-runtime-pty-id; 22 vi.mock factories move the re-exported keys to those modules (pty-id keeps its other exports via importOriginal) * test: drop the duplicated verified-input mock (Opus P3)
…store (stablyai#30697) * plugins: extract the verified artifact-pack registry from language packs Pure refactor. Verify, bounded read, parse and per-plugin error handling move into PluginArtifactPackRegistry; language packs become one instance, so later data families reuse it instead of copying the file. * plugins: reconcile content families from one list Pure refactor. Reconcile and error reporting walk one family list in the same order as before, so a new family is one entry. * plugins: one line per family for declared artifacts and contribution defaults Pure refactor. Declared artifact paths are a list of per-family rows, and the contributes default comes from the schema (prefault) instead of a second copy of the field list; the existing defaults test pins the result. * plugins: extract the renderer plugin content list store Pure refactor. The fetch generation, startup join window and content-change subscription move into createPluginContentListStore; language packs become one instance, and store fields are renamed to entries/fetchEntries.
…uns out of space (stablyai#30679) * fix(ssh): say the host's disk is full when a managed-server install write fails for lack of space OpenSSH's sftp-server reports ENOSPC and EDQUOT as the generic SSH_FX_FAILURE, so a deploy onto a full home failed with the bare detail "Failure" under "Try connecting again". After a failed install write, probe the host's free space where the write was going; when it cannot hold the write, fail with an error that names the filesystem, how much is free and how much is needed. Lost or aborted transfers and Windows hosts keep their original error. Co-Authored-By: Claude <noreply@anthropic.com> * fix(ssh): diagnose only SFTP's generic failure on an exhausted filesystem Codex review: a partial upload may spend the room the rest needed, so compare against an empty filesystem, not the bundle size, and keep explicit refusals such as permission denied. Co-Authored-By: Claude <noreply@anthropic.com> * refactor(ssh): reuse formatKb and size an upload the way uploadDirectory walks it Opus review: the shared formatKb already exists, and following symlinks inside the tree could recurse; uploadDirectory follows the root and skips links below it. Co-Authored-By: Claude <noreply@anthropic.com> * test(ssh): measure the free-space probe in the host-op byte budget --------- Co-authored-by: Claude <noreply@anthropic.com>
…stablyai#30710) * fix(security): put every trust dialog on a host-owned token surface Plugin themes write CSS tokens on the document root, and dialogs portal to body, so any trust surface without plugin-security-chrome inherits them; a theme could hide what is being approved. The orca.yaml hook trust dialog, notebook run confirmation, plugin install, remove and rollback dialogs were missing it. DialogContent gains surface="security", which applies the host-owned tokens, and a scrollable prop for the plugin dialogs' existing viewport-height scrolling; every trust dialog uses the surface and a test lists them. The style guide says when to use each. * fix(security): keep inline trust decisions on the host-owned surface The agent command approval card (Allow/Deny) and the browser certificate override sit outside a dialog, so plugin theme tokens still reached them. A SecuritySurface primitive puts its child on the same host-owned token set as DialogContent surface="security"; both use it, and the security surface test now lists inline trust surfaces as well as dialogs.
…ablyai#30693) The Git runner sends a command into WSL when the working directory is a \wsl.localhost UNC path, even if no WSL distro is set in the options. literalPathspec only looked at the options, so Windows-style backslashes in file paths reached Git inside WSL unconverted and stage, unstage and discard failed to match the file. literalPathspec now also checks the worktree path, and every caller passes it. Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
…stablyai#30659) GitHub orgs using an SSH certificate authority get an SCP remote whose user is org-<id> rather than git. The three GitHub remote parsers only matched a literal git@ user, so those repos were classified not-github and lost PR, issue and check support. Co-authored-by: Yi-111-a <yi-111-a@users.noreply.github.com>
…er pages everywhere, headless sleep stops terminals (stablyai#30715) * fix(runtime): one owned-session teardown policy; removal closes host browser pages everywhere, headless sleep stops terminals A declared table says what each entry point (worktree, folder and project removal, sleep, disconnect) does to each owned kind (PTY, native chat, host browser page, launch record), so a new kind is a compile error until decided. Removal over desktop IPC and repo.rm now close the host's browser pages and launch records like the runtime removal; worktree.sleep with no renderer sleeps the host's terminals instead of doing nothing. Sleep->chat is declared pending stablyai#28445; disconnect stops nothing. * fix(runtime): project removal finds headless pages and failed launches without a client page; stub the new runtime method * fix(runtime): headless sleep rejects an unproven stop; read headless page owners through the bridge, not Electron
… remote contradicts (stablyai#30717) * fix(projects): refuse to set a folder up under a project its own GitHub remote contradicts Setting an existing folder up for a project stamped the project's identity as the folder's upstream even when the folder's own remote named another GitHub repository, so the folder was silently relabelled into the wrong project (fabricated upstream, wrong projectId on its setup and workspaces). Fixes stablyai#30663 * test: type the folder-remote spec store without casts * fix(projects): judge a registered folder by its current remotes; ignore host aliases * test: cover the desktop re-read of an already-registered folder
…the desktop (dark switch) (stablyai#30721) * fix(sidebar): key folder workspace rows by host in the shared builder so the phone matches the desktop Co-Authored-By: Claude <noreply@anthropic.com> * test(sidebar): pin the SSH folder row key encoding Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
…all an older version" (stablyai#30724) * fix(ssh): name a noexec home instead of calling the host unsupported * test(ssh): measure the stage exec probe in the host-op byte budget * fix(ssh): the connect error names a noexec home too * chore(i18n): add the noexec status line to the English catalog * fix(ssh): only the shell's exec refusal reads as noexec; other probe failures reach the self-test * fix(ssh): a stage the probe can't write skips the probe, so a full disk still reads as full * fix(ssh): read the probe's exit 126 as noexec in any locale
… of nothing (stablyai#30725) * fix(runtime): say why a new terminal on a host failed, with a retry * fix(runtime): never offer a retry or say refused when the host's reply was lost * fix(runtime): only a host error reply counts as refused; the wake flow keeps its own toast * refactor(runtime): one 'couldn't confirm' notice carrying the reason, no failure classification * fix(runtime): one create notice per workspace, so retried seeding can't stack toasts
… unproven ones (stablyai#30722) * refactor(pty): every PTY retirement names its exit proof; ratchet the unproven ones * refactor(pty): count raw ownership drops, route an abandoned relay stop to unverifiable, honest proof labels
…esktop sidebar does (switch on) (stablyai#30718) * test(e2e): a phone lists the desktop's managed SSH workspace as the desktop sidebar does (switch on) * fix(sidebar): key folder workspace rows by host in the shared builder so the phone matches the desktop Co-Authored-By: Claude <noreply@anthropic.com> * test(sidebar): pin the SSH folder row key encoding Co-Authored-By: Claude <noreply@anthropic.com> * test(e2e): route the SSH phone feed list spec from its sources Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com>
…r yet" terminal list (stablyai#30743) * test(e2e): mount-authority polls retry the host's "can't answer yet" terminal list First-mount adoption briefly makes terminal.list refuse with terminal_surface_ownership_unavailable. Thrown out of an expect.poll callback, that ended the poll instead of retrying it, so the spec failed whenever a slow runner's probe landed in that window. * test(e2e): say the refused read returns undefined
…ck main's copy (stablyai#30745) The pane-close consistency spec closed panes with a bare PaneManager.closePane, the drawing-layer call that never asks main. In a layout-owner build that close reached main only because killing the pane's terminal made main retire the pane; right after a reveal the terminal has not attached yet, so main kept the pane. Every user close path already sends the close to main first. The helpers now close through the pane close request the menu and shortcut use, and every scenario also checks that main's stored panes match the window's.
…h) (stablyai#30746) Each stamp write made every client re-list the repo's worktrees; a pane whose agent title flips while streaming turned that into hundreds of re-lists a second. The write stays immediate; the notice is coalesced per workspace (50 ms quiet window, 250 ms cap) through the existing keyed coalescer.
… in shell; the state lock sends its script once (stablyai#30744) * refactor(orcad): every snapshot state op fits the gateway exec budget in shell; the state lock sends its script once * test(orcad): a restore that stops part way reports FAILED and reruns from the archive
…ages and every terminal on the host (stablyai#30728) * fix(projects): removing a project on this computer stops its sessions on the host, like repo.rm * fix(runtime): a spent removal budget still asks every project terminal to stop; paired residue rows are no twins
…s the one layout delta (stablyai#30734) No host ever read intent: 'gesture'. The client stops sending it, the params schema strips it from older clients, and docs/reference records the rule that a snapshot never carries intent.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
ELI5
OMP workers could accept a thinking preference but leave it out of the actual launch. Orca now passes the requested setting to OMP, including Off and Auto; leaving it unset still uses your OMP configuration.
What Changed
--effortto native--thinking, preserving Off, literal Auto and opaque model selectors.Why
A native thinking policy is not always an effort tier. Keeping omission, Off and Auto distinct avoids silently changing the user's configuration or acknowledging a preference the host drops. This reuses the existing option and launch paths rather than adding another setting store.
Linked Issue
Fixes stablyai#23388
Overlap: PR28438 handles ordinary effort mapping; PR29798 adds native discovery. Both remain open. This candidate retains the gap around configured-default omission and explicit Off/Auto; it does not copy either PR or add another discovery path.
Visual Proof
Both launches requested Off, with OMP's configured default set to Low. Windows before:
51530bb; after:652a8a4.Before — request ignored
After — request honored
Actual native policy rows captured with OMP still running, not after returning to the shell. No completed model turn or per-turn Auto classification is claimed.
Testing
pnpm testshards passed: run.AI Disclosure
Implementation, integration and final review: OpenAI Codex GPT-6.1 Sol. Earlier independent review also used Claude Fable 5.1. Checks were agent-run, not human testing.
Review
Agent skill upstream boundary
Notes
No dependency or release-version changes.
Author
GitHub: @NarryG. X: @NarryOnTheNet.
Checklist