Repository navigation
Conversation
Ports Synara's Linux admission rules, the task-scoped X11 Escape shortcut and the Linux desktop host (#1294 core hooks). Linux stops failing closed: the Cua host now receives linuxAdmission, and Computer use runs on Linux behind PATHWAY_COMPUTER_USE. Host errors keep the message of plain SurfaceErrors instead of printing [object Object]. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…h app ids A Linux server may not know its desktop yet at startup, so availability gains a transient "checking" member. Preview frames carry their image type (PNG stays the default and the only model-facing format), health can say a backend is dormant by choice, and a launch can name the app id its windows report when a wrapper handed the window off. Ported from Synara pr-1294 (5f3f50397e). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Adds the seams the KWin and Hyprland backends plug into: a switchable backend slot so startup never waits on a wedged session bus, a Linux backend selection with a strict PATHWAY_COMPUTER_BACKEND override, and manager support for desktop-gone reselection, passive status reads, dormant health, luma scroll baselines, per-backend settle policy, observed-paste clipboard restore, launch app ids, JPEG preview frames and pane-input marking. Ported from Synara pr-1294 (5f3f50397e); Synara's tests are the spec. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…views A checking availability shows as checking rather than as a setup to do, on web and iOS, and the live preview decodes each frame as the image type its header names. Ported from Synara pr-1294 (5f3f50397e). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ports Synara's KWin compositor plugin literally: the plugin, its dedicated agent seat, the D-Bus interface, the install and uninstall scripts, the systemd rebuild timer and the fixture tests that compile production functions standalone. Product names are renamed: Synara to Pathway, org.synara.ComputerUse to com.spiritdevs.pathway.ComputerUse, SYNARA_* env vars to PATHWAY_*. The tree lives at native/computer-use-kwin beside Pathway's other natives. Ported from Synara pr-823 (8f8d804828). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ports Synara pr-823's D-Bus layer to Effect: variant unwrapping, typed call/timeout/connection-closed failures, the connection watch that dbus-next itself lacks, the scoped per-bus session token, and the KWin session-bus adapter with owner pinning and throttled-auth cooldown. Connection-level classifiers live in dbusFailures.ts, and the plugin/bus doubles are Effect-returning. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Port Synara's sessionBusNames, evdevInput, pointerSequencing and asarUnpackedPath (pr-823). The probes run on a throwaway watched connection and fail with DbusConnectionClosedError when the bus dies. Sequencing takes Effect sinks and reads time from the Effect Clock; releases run uninterruptibly after the body, whatever its exit. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Port Synara's appLaunchResolution, desktopAppEnvironment and computerHealthState (pr-823). Resolution fails through Effect with AppLaunchResolutionError, a ComputerBackendError; health state is a factory rather than a class. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ports Synara's KWin provisioning helpers: distribution and toolchain probes, KWin compatibility, prebuilt verification, the flock-based provisioning lock, the from-source plugin build, and the pkexec package installer. Cancellation is fiber interruption throughout; runCancellable runs its child in a process group through the Effect spawner, and pkexec is only ever signalled while it is still waiting for its dialog. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Reads, writes and paste-once offers on the seat0 clipboard now run as Effects. The paste-once offer's end is a Deferred with a bounded watch fiber. wl-copy is still spawned directly, because the Effect spawner's scope would take down the fork that serves the selection. The KWin installer-script tests run in the ported sandbox testkit, on Linux only. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Installs the KWin plugin under the user's home, preferring a verified prebuilt and falling back to a source build under the installer lock. Cancellation is interruption; every failure is a ComputerBackendError. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ports Synara's AT-SPI helper (Python, literal with pathway renames), the helper client on the Effect process spawner, tree fusion and the perception layer the KWin engine reads trees through. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ement When the 64-entry operation queue was full, replaceDesktop ran the swap outside the queue and aborted only agent calls. A pane click waiting on the old desktop's window list then sent its input to the replacement desktop. The swap now enters the queue as a barrier the admission limit cannot refuse, and replacement aborts in-flight pane input as well as agent calls. Only a closed queue lets the swap run directly. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…lients Each still-frame publisher numbered its frames from 1, and the manager sent that number on unchanged. After a desktop replacement, web and iOS clients that kept their frame socket open still held the old stream's higher sequence, so they dropped every new frame as stale. The manager now stamps the outgoing sequence from its own counter, so it keeps rising for the computer stream across backend instances. Neither client's sequence gate or the frame protocol changes. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ktop A thread seeded while Linux backend selection was still running kept reporting "checking" after the selected backend arrived. The swap's health and capability events only republish the cached thread snapshot, so the thread showed the placeholder until it asked again. A later replacement left threads showing the gone desktop in the same way. Replacing the desktop now invalidates the physical snapshot, and ignores an in-flight read of the previous occupant. It then pushes the new occupant's passive availability to every seeded thread without establishing or provisioning it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The barrier test forked the barrier lazily, so it asked for admission only after the held operation had been released and the queue had drained. Admitting barriers under the ordinary 64-entry limit still passed. The barrier now starts immediately and must be waiting, not refused, while all 64 entries are still held. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… its probe After a desktop swap, the manager pushes a passive probe of the new desktop to every thread. If a request had already engaged the manager during detection, a later request could establish the new desktop and succeed while the probe was still running. The probe then landed on an empty snapshot, pushed its older "unavailable" verdict, and every later health event republished it. Every establishing availability read now goes through one helper. The first one to succeed after a replacement becomes the threads' verdict and is pushed to them, and the probe still in flight is dropped. With no real use, the probe behaves as before and never starts the desktop. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
After a desktop swap, the manager cached availability through three separate writers, each with its own guard. A cached read failure outlived a later success, so threads kept showing "unavailable". A slow passive or establishing read could overwrite a newer verdict for the same desktop. A push that was still running when the desktop was replaced again read the backend and started the new, unused desktop. Each availability read now takes a revision when it starts. The cache accepts a result only if it is newer than the last one it accepted. A replacement starts the cache at a fresh revision, so reads that began before it land nowhere. A newer success clears an older failure. A failed window or screen read is kept as the thread's error but never hides the availability. Publication only copies the cache, never reads the backend. A push about one desktop stops once that desktop is replaced. The one-time replacement flag is gone. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The cross-replacement test only counted establishing reads on the next desktop. Publication no longer reads the backend, so that count stayed at zero even with the generation check removed. The test now also checks that no push about the first desktop lands after the swap. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…op swaps Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… reads the new desktop The debounced republish after a window change does a full physical read, so a timer armed for the old desktop could fire after a swap and run establishing reads on the new one. The swap now retires the old occupant inside its barrier: it bumps the generation, resets the cache and cancels the pending timer. A timer that fires anyway checks the generation it was armed in and drops itself. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…republish Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Mirrors Synara's KWin prebuild matrix on the Linux fleet runner: one digest-pinned container per distribution, KWin read off the image, and the .so plus a JSON sidecar uploaded. The KWin fixture tests run as their own job. x64 only, with no manifest assembly yet. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Mirrors Synara's Cua Linux workflow: Linux host and protocol tests, the pinned source built with both Pathway patches, and its browser and cursor tests. The Electron X11 smoke is not ported yet. docs/internals/computer-use-ci.md records the three workflows, their gaps from Synara, and what the Hyprland lane will need. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
vouch:trusted
This branch has not been deployed
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
1 participant
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.
Draft, not ready to merge. This parks the Linux (KDE/GNOME) half of Computer Use so it isn't lost. The macOS half shipped in b6e9ce3.
What's here
Hyprland is deliberately left out.
Still to do before this can merge
Checks run on this branch
🤖 Generated with Claude Code
Need help on this PR? Tag
@codesmith-botwith what you need. Autofix is disabled.