Agent skills, MCP servers, and the configuration needed to run them.
| Package | Purpose | Origin |
|---|---|---|
| agent-desktop | On-demand Hyprland desktops, MCP server, CLI helper, skill, and setup | This project |
| shot-page | Temporary private screenshot and recording pages, skill, helper, and hosting setup | This project |
| unslop | Edit AI writing habits out of prose; skill only | Lauren Tan’s pstack, MIT; pinned unchanged upstream copy |
From this checkout, run python3 install-skill.py unslop --client codex or --client claude. Restart the CLI and invoke $unslop in Codex or /unslop in Claude Code. The installer refuses to replace an existing skill. Source attribution and T3 notes.
Requires Linux, a running Hyprland session with Lua configuration, and systemd user services. The package was tested on Hyprland 0.56.2. On Omarchy, install dependencies from a terminal:
omarchy pkg add nodejs npm python jq grim wtype wlrctl xorg-xwaylandClone the repository and install as your regular user:
git clone https://github.com/not-compromised/skills-mcp.git
cd skills-mcp
python3 packages/agent-desktop/install.pyTo choose a monitor, get its output name from hyprctl monitors, then pass it to the installer:
python3 packages/agent-desktop/install.py --monitor DP-1Restart your Claude Code or Codex CLI and ask it to use the agent-desktop skill. The skill can call the local MCP through agent-desktop tool without native MCP registration. New desktop windows do not take initial focus; click one to interact with it yourself.
The MCP listens on loopback and requires a private token. Desktop input and browser profiles are separate, but commands still run with your user’s filesystem permissions. This is not a security sandbox.
See the package guide for optional native MCP registration, operation, removal, and T3 Code integration. Direct CLI use is the default documented setup.
The same package was submitted in Omarchy PR #10594. That PR also adds omarchy install ai agent-desktop and the Omarchy manual entry. The command is proposed upstream; use the standalone installer here until your installed Omarchy version includes it.
The initial export matches default/agent-desktop in that PR’s commit 579ee99. The Omarchy integration wrapper and manual entry belong to that repository.
Install Bubblewrap for test isolation, then run:
./testThe runner installs locked npm dependencies with lifecycle scripts disabled, then runs the package tests in a separate process and network namespace. It supplies a temporary home and no host desktop sockets, session bus, or GPU devices. Tests cover the actual HTTP authentication, client calls, desktop ownership/lifecycle behavior, and installer preservation/rollback.
GitHub Actions runs the same ./test command on every PR and main push, using
Node 22 and Ubuntu 22.04 with Bubblewrap and FFmpeg installed. Tests fail if isolation cannot
start; CI never falls back to accessing the runner's desktop or network. The
runner's system Python runs the Python suites. Test logs are retained for seven
days. The final CI gate requires an explicit successful test result, including
when the test job fails or is skipped. Configuring that check as required in
repository protection is a separate step. CI does not install the desktop
service or publish packages.
Do not use the broad Omarchy test suite as a live-session smoke test. GUI verification belongs in a disposable graphical environment.
Our code is MIT. Third-party packages retain their original copyright and license notices; Unslop carries Lauren Tan’s MIT license.