Capture the active display (DISPLAY=:10) and return a base64-encoded PNG. Implemented in SCRUM-1397 as the proof action.

Capture and return a magnified PNG of a sub-region of the screen.

Left-click at a coordinate.

Right-click at a coordinate.

Middle-click at a coordinate.

Double left-click at a coordinate.

Triple left-click at a coordinate.

Move the pointer to a coordinate without clicking.

Press the left button at start_coordinate (or the current position) and drag to coordinate before releasing.

Scroll in a direction by an amount at a coordinate.

Press and hold the left mouse button (released later by left_mouse_up). Requires the persistent session.

Release a left mouse button held by left_mouse_down.

Type a string of text at the current focus.

Press a key or chord using xdotool key syntax, e.g. 'Return', 'ctrl+s', 'alt+Tab'.

Hold a key (or chord) down for a duration in seconds.

Read the X clipboard contents (via xclip).

Write text to the X clipboard (via xclip).

Request a session grant for one or more applications (Linux stub: auto-grants and returns screenshotFiltering=false; the real grant model lands with the service engine).

Return the set of applications currently granted desktop access (Linux stub: echoes the granted set).

Launch or focus a desktop application by name.

Switch the display the session targets, e.g. ':10'.

Run a sequence of computer-use actions in order and return one combined result.

Wait for a duration in seconds (then optionally screenshot).

Return the current cursor [x, y] in model coordinates.