# Linux Desktop

Control Agent Zero's Docker/Xpra desktop and GUI apps; not the user's host computer.

- Repository: https://github.com/agent0ai/agent-zero (19630 stars)
- Category: AI and agents
- License: see repository
- Last updated: 2026-11-01
- Runs scripts: yes, review before use

## Install

Claude Code: `npx skills add agent0ai/agent-zero --skill linux-desktop -g -a claude-code -y`
Codex: `npx skills add agent0ai/agent-zero --skill linux-desktop -g -a codex -y`
Cursor: `npx skills add agent0ai/agent-zero --skill linux-desktop -g -a cursor -y`
Gemini CLI: `npx skills add agent0ai/agent-zero --skill linux-desktop -g -a gemini-cli -y`

## Description

Control Agent Zero's Docker/Xpra desktop and GUI apps; not the user's host computer.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Linux Desktop Interface

## Boundary With Host Computer Use

This skill is only for the built-in Desktop surface running inside Agent Zero's Docker/runtime Xpra session. It does not control the user's host OS, real monitor, local Ubuntu/Wayland session, macOS desktop, or Windows desktop.

If the user asks for `/computer-use`, "my computer", "host machine", "local desktop", screenshots of the user's screen, hiding/minimizing a window on their screen, or any connected A0 CLI Computer Use task, do not use this skill, `desktopctl.sh`, Xpra, `code_execution_tool`, or Docker shell commands. Load `host-computer-use` and use `computer_use_remote`; if unavailable, tell the user to enable `/computer-use on`.

`desktopctl.sh` only targets the internal Agent Zero Xpra display. It cannot observe or control the user's connected host screen.

Use the Desktop as a full Linux GUI when the user explicitly needs a visual workflow, an installed desktop app, or manual layout polish that is awkward through structured file edits alone. Agent Zero may warm the persistent Desktop runtime during initial startup, but visible Desktop surface use remains opt-in. The Desktop is opt-in at the UI level: do not open a surface just because the user asks for a document. Use structured tools first for deterministic content changes, then use the Desktop for inspection, GUI-only actions, and final visual confirmation.

## Operating Model

The Desktop is a structured-first X11 control surface. Use this decision hierarchy:

The rest: https://github.com/agent0ai/agent-zero/blob/HEAD/plugins/_desktop/skills/linux-desktop/SKILL.md