Skip to content

feat(computers): cua-driver 0.47.4 on Omarchy, its Hyprland plugin built by the release's kit - #609

Merged
earakely-scale merged 1 commit into
mainfrom
edgararakelyan/omarchy-trycua-plugin
Nov 11, 2026
Merged

earakely-scale merged 1 commit into
mainfrom
edgararakelyan/omarchy-trycua-plugin

Conversation

@earakely-scale

@earakely-scale earakely-scale commented Nov 11, 2026 •

Copy link
Copy Markdown
Collaborator

Part of FD-5336.

What

Omarchy computers now run cua-driver 0.47.4 with the Hyprland plugin from the same release, which keeps an agent's background clicks, keys, drags and popups on the plugin's own seats.

  • The plugin (computers/disk/omarchy/recipe.sh):
    • The base builds trycua's plugin from the release's source with the release's own build kit, unmodified. makepkg runs as omarchy, and both archives are pinned by sha256.
    • It replaces the 0.33.2 package Omarchy builds, which predates trycua's fixes for background input on Hyprland (any native Wayland app, popups and drags off the user's seat; 0.46.1 and later).
  • What the kit pins:
    • The kit builds only on Arch's hyprland 0.56.2-1 and the GCC 16.1.1 20260728 it was built with. Omarchy 4.5 ships 0.56.2-4 and GCC 16.2.1.
    • The recipe installs those from the Arch Linux Archive's day 2026/08/08 (arch-day.conf, one pacman transaction, signature-checked). With them come that day's aquamarine, hyprtoolkit and hyprland-guiutils, since Omarchy's hyprtoolkit links aquamarine's newer soname.
    • cmake and ninja are installed for the build and removed after it, along with the day's sync databases.
    • The plugin goes in last, after the session toggles, so the Hyprland running in the builder never loads a plugin built for another build of itself.
  • The driver (drivers/envs/cua-driver):
    • On Hyprland, cua-driver comes from the 0.47.4 release's Linux tarball, pinned by sha256 and unpacked by install-hyprland.sh, the way Windows and macOS take theirs. 0.47.4's wheel isn't on PyPI, which refused it for size (perf(cua-driver): strip symbols and fat LTO for release builds trycua/cua#7316).
    • EXECUTABLES["hyprland/amd64"] and mcp-hyprland.json are 0.47.4's.
    • A unit test keeps the plugin's release equal to the driver's.
  • The agent cursor: the daemon already starts with --no-overlay, which turns off cua-driver's agent-cursor overlay, the one that can take the keyboard as it first shows.
    • On the new base, with the flag, get_agent_cursor_state says enabled: false and no cua-agent-cursor layer shows.
    • Without the flag, the layer appears. So nothing changes there.

Verification

On a c8i.4xlarge with KVM (Ubuntu 24.04), against this branch:

  • The base build: Omarchy 4.5's image became a base in about 4 minutes.
    • It ended with hyprland 0.56.2-1, aquamarine 0.14.0-1, the r595 GCC family and cua-hyprland-plugin 0.47.4-1, with no cmake or ninja left.
    • The kit's own checks passed (compositor, compiler, runtime), and its 21 tests passed twice.
    • Booted, hyprctl -j cua:status reports abi.match: true, configured: true, transport.ready: true and no config errors.
  • Integration: tst/integration/computer/omarchy_test.py passed 2 of 2: the probe at full, and the notes app before and after a reset.
  • The bench's BenchLab checks: the 13 scripted checks of omarchy-benchlab, driven through cua-driver, run 4 at a time.
    • 12 ran clean. The 13th's computer was OOM-killed on the 32 GiB host and passed when rerun alone, as the fleet runs trials.
  • Background trials (Sonnet 5.5): omarchy-ir-02 and omarchy-ir-03, r1 to r5 each.
    • All 10 passed both the task grade and gui_only, with user_undisturbed true on every criterion.
    • One trial's first computer was OOM-killed the same way and passed when rerun alone.
    • On the published run, the same 10 trials passed 1 of 10, 9 of them failing user_undisturbed.
  • Unit tests: tst/unit/drivers and tst/unit/env/envs pass.
    • The exception is test_disk_recipe.py's Windows ISO boot test, which fails on macOS on main too.
  • Cause found along the way: a builder whose drop-in landed before the plugin hung Hyprland 0.56.2-4 drawing its config-error overlay. The session now ends first, and a v0.9.1500 baseline on the same host built normally.

🤖 Generated with Claude Code

RetriggerView in GreptileConfidence Score: 5/5

The PR appears safe to merge; no actionable issues were found.

Summary

Updates Omarchy to cua-driver 0.47.4 and builds its matching Hyprland plugin with the release’s build kit.

  • Omarchy bases build the Hyprland plugin from the same release as their driver.
  • Omarchy installs cua-driver's Linux release binary beside its driver code.
  • Hyprland tool notes tell agents to batch known input actions.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Archive[Arch snapshot] --> Base[Omarchy base]
  Kit[Plugin source and release kit] --> Base
  Base --> Plugin[Matching Hyprland plugin]
  Tarball[Driver release tarball] --> Install[Driver installer]
  Install --> Driver[cua-driver 0.47.4]
  Driver --> Plugin
Loading

Reviews (1) · Last reviewed commit: "feat(computers): cua-driver 0.47.4 on Om..." · Reviewed by Greptile

…ilt by the release's kit

Omarchy's base builds trycua's cua-hyprland-plugin from the 0.47.4 release's source with
the release's own build kit, unmodified, on what the kit pins: Arch's Hyprland 0.56.2-1
and GCC 16.1.1 20260728, from that day of the Arch Linux Archive, with that day's
Hyprland libraries. It ends the builder's desktop session before Hyprland's config names
the plugin. The driver takes cua-driver 0.47.4 from the release's Linux tarball on
Hyprland, whose wheel isn't on PyPI, with 0.47.4's manifest.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@earakely-scale
earakely-scale requested a review from a team as a code owner November 11, 2026 01:34
@earakely-scale
earakely-scale merged commit 4c339bb into main Nov 11, 2026
34 checks passed
@earakely-scale
earakely-scale deleted the edgararakelyan/omarchy-trycua-plugin branch November 11, 2026 02:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant