Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Desktop for Linux

    1

    Give your AI agent its own isolated Linux desktop. It operates real apps with its own pointer and keyboard while you keep working on yours.

    Free

    Security scanned
    Agent Desktop for Linux

    Agent Desktop for Linux

    Example session with this skill installed

    Open a local form in the browser on your own desktop, fill in the name "Ana López", choose "Portugal", tick "I accept" and submit it. Don't touch my mouse.

    • Read your context and instructions
    • Compiled the agent desktop

    What I did

    Done. I worked on my own private display, so your pointer, keyboard and windows were never touched.

    StepCommandResult
    Start private desktopadc start --nestedDisplay :7, openbox, private accessibility bus
    Open the formadc browser file:///…/form.html --name workChrome with an agent-owned profile
    Read the pageadc snapshot -i --app chrome5 form controls found
    Nameadc type @4cd4:e14 "Ana López"Typed with the real keyboard (accents included)
    Countryadc select @4cd4:e16 PortugalSelected, dropdown closed
    Consentadc check @4cd4:e15Checked
    Submitadc click @4cd4:e17Pressed "Enviar"
    Verifyadc find --name "Enviado""Enviado: Ana López / true / Portugal"
    Finishadc stopBrowser closed gracefully, cookies saved

    Snapshot excerpt

    • textbox "Nombre" @4cd4:e14 [69,172 159x22]
    • checkbox "Acepto" @4cd4:e15 [235,174 14x14]
    • combobox "País" @4cd4:e16 [335,173 73x20]
    • button "Enviar" @4cd4:e17 [411,172 55x22]

    Next steps

    • Add --record ~/Videos/form.mp4 to adc start to get a video of the run.
    • If a site asks for a login or captcha, I'll run adc ask-user and you finish that step yourself in one click.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Manual GUI automation on Linux is high-friction, often stealing focus or clashing with user activity. This skill provides a dedicated, isolated X11 workspace where an agent operates its own pointer and keyboard without interrupting your work.

    What it does

    • Isolated Desktop Management creates a private Xephyr display with its own focus, clipboard and D-Bus session
    • Accessibility-First Navigation uses AT-SPI to snapshot, find, and interact with GUI elements via stable references.
    • Human-Like Input simulates real pointer glides, clicks, drags, and keyboard events that trigger genuine application logic.
    • Visual Verification captures screenshots with coordinate grids and screen recordings for task auditing.
    • Persistent Browsing manages isolated browser profiles so agent sessions and logins stay separate from yours.
    • Safe User Hand-off triggers system notifications for captchas or 2FA, allowing you to temporarily take control.

    How it works

    1. Run adc doctor to verify system dependencies and X11 environment compatibility.
    2. Initialize a private session using adc start --nested to launch the isolated agent workspace.
    3. Use adc snapshot or adc find to identify interactive elements by their accessibility roles and names.
    4. Execute actions like adc click @ref, adc type, or adc drag-xy to complete desktop tasks.
    5. Terminate the session with adc stop to gracefully close applications and save recordings.

    Frameworks & tools

    Built for Linux (X11/GNOME) environments using Python 3, PyGObject, AT-SPI, Xephyr, Openbox, and ImageMagick. It integrates standard CLI tools like xdotool, wmctrl, and ffmpeg for input and media handling.

    Why this beats prompting it yourself

    Standard agents struggle with GUI coordinates and focus stealing, often failing when window positions shift. This skill provides a structured accessibility engine and a sandboxed display, ensuring the agent interacts with actual UI objects rather than guessing pixel locations.

    Use cases

    • Automating form entry in legacy desktop applications that lack an API.
    • Setting up and testing complex software environments in an isolated workspace.
    • Performing long-running web tasks that require persistent sessions and visual verification.
    • Recording step-by-step GUI tutorials or bug reproductions automatically.

    Known limitations

    Does not support Wayland native sessions. Applications that do not expose an accessibility tree require coordinate-based interaction. Requires specific system packages listed by the doctor command.

    Requires X11 environment. No native Wayland support. Apps without AT-SPI trees require coordinate-based interaction. Specific apt packages like Xephyr and Openbox must be installed.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 8 days ago

    • Free with an Agensi account

    Requires

    Python 3 (PyGObject, AT-SPI)
    Xephyr
    openbox
    xdotool
    ImageMagick
    ffmpeg
    Google Chrome (optional, for adc browser)
    Node.js
    Python
    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables
    Browser

    File Scopes

    ~/.local/share/adc/**
    $XDG_RUNTIME_DIR/adc/**
    ~/Videos/**

    Runs a private X display (Xephyr) on its own workspace. Writes screenshots and state to $XDG_RUNTIME_DIR/adc, agent browser profiles to ~/.local/share/adc (mode 0700) and optional recordings to ~/Videos. Network is only used by the browser the agent opens. Credentials are never typed by the agent: sign-ins are handed to the user.

    Listed8 days ago

    What's inside

    Frequently Asked Questions

    Traditional automation tools hijack your mouse and keyboard; this skill creates a private, 'invisible' Xephyr workspace where the agent works in parallel, allowing you to continue your own tasks without interruption.

    The agent uses AT-SPI to 'see' the UI hierarchy, meaning it identifies buttons and fields by their names and roles rather than just pixel coordinates, which makes it much more resilient to window resizing.

    The skill is designed for X11 environments (like GNOME or Openbox) and requires Python 3 along with system packages like Xephyr and xdotool; run the 'adc doctor' command post-installation to verify your setup.

    Native Wayland sessions are not currently supported, and applications that do not expose an accessibility tree will require the agent to fall back to coordinate-based interactions.

    It's free and open source (MIT): the full skill, the adc command-line tool and its accessibility engine. A Pro version with OCR, set-of-marks, macros, session reports, parallel desktops and an MCP server is available separately.

    Yes. adc browser URL --name work opens Chrome with an agent-owned profile that keeps its logins between sessions. Use a different --name for each set of accounts.

    New in Agents & Orchestration