Skip to content

Releases

v0.2.3

Choose a tag to compare

@github-actions github-actions released this 18 Nov 00:42
  • Screenshot providers registered with register_screenshot receive the request's Starlette QueryParams instead of a plain dict, so a repeated query parameter such as frame=a&frame=b reaches them in full through getlist(). get() works as before.

v0.2.2

Choose a tag to compare

@github-actions github-actions released this 17 Nov 23:27

The code is unchanged from 0.2.1. This release republishes the image so the Docker Hub page carries the corrected README.

  • The README and documentation site present the image as usable on its own for Linux desktop automation as well as a base for derived images.
  • run_command and launch_app are described as running any command or program they are given, instead of as a shell.
  • The README lists every action. The actions reference documents every action's parameters, defaults, limits and return fields, including screenshot sizing, the events stream, recording modes and clipboard limits. The MCP page lists how each tool differs from its action. The installation, script mode, plugins and development pages fill in the remaining environment variable values, startup checks, script failure cases, registry options and Make targets.

v0.2.1

Choose a tag to compare

@github-actions github-actions released this 17 Nov 22:57

v0.2.0 was tagged but never published: a ShellCheck finding in the new VNC cursor test failed the lint job, so no image or release was built. v0.2.1 is the first published release with everything listed under 0.2.0.

  • The VNC cursor test passes ShellCheck.

  • paste_text action and MCP tool: enters text the way people do. It puts the text on the clipboard, clicks the field when x and y are given, then presses ctrl+a, backspace and ctrl+v through the humanized send_key, with irregular pauses between the keys and before the first key after the click. replace: false pastes at the caret. It is meant for GUI text fields; terminals bind those keys differently.

  • set_clipboard_data MCP tool for HTML, PNG and file-list clipboard data, matching stealthy-auto-browse.

  • The MCP mouse_move and human_pause tools take every option stealthy-auto-browse's do (curvature, tremor, drift_interval, drift_distance, speed), and system_double_click, system_right_click and system_drag take the mouse overrides.

  • The integration suite now covers every system-level check from stealthy-auto-browse's suites: users and ownership, recording, auth, the VNC cursor over a raw RFB client, resolutions, desktop screenshots, script mode and run_script control flow, the clipboard with a real X client, the request lock, the generic MCP tools, and humanized mouse, typing, scroll and pause metrics against human thresholds.

  • The MCP instructions and the system_type, send_key and set_clipboard descriptions tell agents to enter text with paste_text and keep system_type for a few characters. The actions reference and MCP docs say the same.

  • The noVNC viewer kept showing the pointer where it was once viewers had sent no input for about 30 seconds, so someone watching automation saw a frozen cursor. x11vnc's idle napping stopped it from redrawing the cursor; it now runs with -nonap. An integration test idles past that point and checks the cursor still moves.

  • desktop_list_windows returned a wrong title and WM_CLASS for windows whose WM_CLASS contains a space, such as a second Tk window (tk #2), so finding those windows by title failed. Titles now come from a listing without WM_CLASS, and WM_CLASS is read separately.

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 17 Nov 16:40
  • Run a virtual Linux desktop in Docker: Xvfb at XVFB_RESOLUTION, Openbox with one desktop, x11vnc and noVNC on port 5900, and an HTTP API with an MCP endpoint on port 8080.
  • Humanized OS-level mouse, keyboard and wheel input with per-request timing overrides, in screen coordinates. The input planners and executor come from stealthy-auto-browse v3.6.0.
  • Window control through desktop_list_windows, desktop_focus_window, desktop_move_resize_window, desktop_set_window_state and desktop_close_window.
  • Openbox key bindings for a terminal (ctrl+alt+t, win+enter), half-screen tiling, maximize, fullscreen, minimize and centering, plus desktop_keybindings, which lists every binding from the live config with its send_key value.
  • Run shell commands with run_command and start programs with launch_app, list_launched and kill_launched. DESK_EXEC_ENABLED=false turns both off, and the container warns at startup when they are on without AUTH_TOKEN.
  • Screenshots, screen recording, the X clipboard, events, human handoff, run_script and YAML script mode.
  • A plugin interface for derived images: AUTO_LINUX_DESK_PLUGINS loads modules that add or replace actions, MCP tools, routes, screenshots, state, health checks, lifecycle hooks and the input coordinate surface. Root entrypoint hooks run from /etc/auto-linux-desk/entrypoint.d/.
  • Documentation site, unit tests and a Docker-backed integration suite that measures input through a real X client.
  • A Buy Me a Coffee funding link and a README Sponsor section.