[ShadowfetchStudio]

Omarchy could run the installed engine, but the 2.3.1 AppImage window failed because WebKit used the Debian build host's absolute helper path. Relocate the known helper paths without changing ELF offsets, retain a before/after hash receipt and patch program, and verify the shipped library. On compatible non-Debian layouts, select a complete host WebKit/GTK stack to avoid mixing newer Mesa with bundled Debian libraries. Missing dependencies/helpers and unknown relocation layouts fail closed; sandbox protections remain enabled.

Also fix two interaction problems reproduced during qualification: the model picker could steal focus after selection had focused the composer, and background ACP discovery could initiate Grok's interactive browser login. Picker close now respects focus moved outside it. ACP discovery uses known existing-login methods or checks session availability; it does not select an arbitrary interactive authentication method. Integration coverage distinguishes an existing session from authentication-required responses. Sidebar hints and editor browser scenarios use the viewing platform's keyboard conventions.

Prepare version 2.3.2 and add portable installed-engine endurance qualification with binary hashes, platform resource measurements, cancellation and persisted-state recovery. The existing approved signing verification module remains byte-identical; new WebKit regression coverage is an additional mandatory step on both Linux release architectures.

Validation for candidate 7f1958aa6e949f882368881874a7aa0bef2bb167:

• UI: 875 unit tests across 155 files and all 136 browser scenarios passed in CI. Typecheck passed. The focus regression failed before the fix; all affected workspace/composer scenarios also passed 111 executions across three repetitions.
• Windows CI: workspace compile, engine unit/integration tests, preview installer, version/no-HOME checks and official clean-PATH installed-runtime checks passed. macOS compile/Clippy/core checks passed.
• Linux x64 and ARM64: workspace tests and built-project cleanup passed. Each native engine passed 200 concurrent tasks, 100 sequential tasks, five cancellations and ten subprocess-output floods. Native WebKit smoke passed 31 checks; the dedicated window endurance run passed 100 explicitly approved UI tasks on each architecture. These are X11 source-build results, not physical Wayland or final public packages.
• Physical Windows preview: the unpacked CI artifact passed native find/replace, undo, both redo shortcuts, save-to-disk and Settings checks with no page errors. The same executable passed 100 sequential tasks, 80 concurrent tasks, five cancellations and persisted-state recovery. This is an isolated preview preflight, not installed public-release qualification. A real Gemma E4B CPU coding preflight also passed in a disposable plain folder: the agent ran the requested tests, all five original tests were preserved and independently passed, and the run completed in 444.8 seconds. This does not exercise Git change review.
• Physical Omarchy packaging experiment: the baseline executable with candidate launcher/relocation passed 12 focused Wayland checks, including onboarding, test-model selection, approved write/exec task, summary and app/child shutdown. The full suite's floating-size assertion failed under Hyprland tiling; the focused smoke explicitly excludes resizing/recovery/endurance and the known-bad baseline Grok probe. The actual Grok CLI separately confirmed that initialize/session-new without authenticate reports sign-in required without starting browser authentication.
• Installed 2.3.1 engine baselines on Windows, Omarchy and Intel iMac each passed the portable endurance probe. Scope and raw hashes are in the qualification record.

Remaining release work: finish Linux native packaging CI (Flatpak CI has passed), build and publish through the existing release workflows, install and qualify the exact final downloads on the three named machines, and publish verified GitHub/website notes and current screenshots. The staged website images are unedited captures of the native Linux candidate with an example project/test model. Real cloud-provider accounts, GPU endurance and broad model-quality coverage are not established by scripted-provider tests.

[coderabbitai]

Important

• Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: f7ad616e-c8d0-4ec6-9b9c-751c1061c271

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

Share

• X
• Mastodon
• Reddit
• LinkedIn