extension.js · part 24

60 text occurrences from desktop/Cursor.app/Contents/Resources/app/extensions/cursor-computer-use/dist/extension.js, part 24. Every entry preserves the shipped literal and its saved verdict or selection reason.

File contents and all parts · All files

Drag ${e} point in latest-screenshot pixels.

Or pass path (2–64 points) to drag through them in order in one call (a whole outline); a path of over two points must stay inside the window it starts in.

{x, y} points, in order; replaces from and to.

pid from computer_apps or computer_resolve_app.

target_id from computer_apps or computer_resolve_app.

Default: {"scope":"app","pid":…,"target_id":…} from computer_apps or computer_resolve_app. That drives one macOS app without taking screen control or moving the real cursor (a cursor overlay is still shown), and needs no computer_start_control. Omit (or {"scope":"screen"}) only to escalate to full-display remote control.

The app to act on.

Required with an app target: the coordinate_token from the latest computer_screenshot of this app. It stays valid until the next screenshot or until the window moves or resizes; screenshot again when an action reports it stale.

Id from a computer_zoom result: x and y are then pixels of that zoom image.

Id from a computer_zoom result: the points are then pixels of that zoom image.

The snapshot_id from computer_app_state, or from a screenshot that reused a live handle. It stays valid across actions until you read a new tree, the window or app changes, or an edit opens a sheet or replaces the tree. Do not call computer_app_state after a screenshot that reused snapshot_id; read computer_app_state when a screenshot omitted the tree or an action reports the snapshot stale.

An element id in brackets from computer_app_state.

Computer-use screenshot must be ${xi.width}×${xi.height}, received ${n.width}×${n.height}

If this click was meant to open or switch to an app, its window may take a second: call computer_wait (1000–3000 ms) and look again before trying another route.

App targets need the macOS computer-use helper. Use the screen target.

APP pid=${e.pid} target_id=${JSON.stringify(e.targetID)} active=${e.isActive}

 app=${JSON.stringify(e.localizedName)}

 bundle=${JSON.stringify(e.bundleIdentifier)}

Nothing was actuated. Correct the argument the message names and call the same tool again.

No tool by that name exists here; use one of the advertised computer-use tools.

Invalid arguments for ${e.name}: unknown argument${1===a.length?"":"s"} ${s(a)}; 

${e.name} takes no arguments.

the arguments of ${e.name} are ${s([...n])}.

Invalid arguments for ${e.descriptor.name}: ${Gl(r.error)}

${e.path.join(".")||"arguments"}: ${e.message}

Run up to ${Dn} actions as one call with one screenshot at the end (${Mn} when every step is computer_key or computer_type), for a sequence already planned from the current screenshot: pick a tool, then a color, then draw; type a value, then press Return. Each step names a tool and its usual args; coordinates and zoom_ids describe the screen as it is now. The batch stops at the first refused step, at an expect_change step that changed nothing visible, or when the foreground window or its title changes; the result reports each step's change and target and names the steps that did not run. Do not batch past a step that opens a window, menu or dialog the next step needs to see. Delete, Backspace, Alt+F4, Ctrl+W, Ctrl+Q, Ctrl+X and Win+L need allow_destructive.

That tool's arguments, as for a lone call.

Stop the batch if this step changes nothing visible.

Allow keys that delete, cut, close or lock.

The batch needs the Windows computer-use sidecar. Run the steps as separate calls.

Invalid arguments for computer_batch: steps[${t}] names ${n.tool}, which is not a tool of this toolset.

Invalid arguments for computer_batch: steps[${t}] (${n.tool})

${s}: unknown argument${1===e.length?"":"s"} ${t(e)}; the arguments of ${n.tool} are ${t([...a.advertised])}.

${s}: ${Gl(e)}

Unhandled batch step tool ${String(e)}

Ran the batch.

 With an app target the result is text only; observe again before the next action.

Take a screenshot of the primary display

, or of one app's window with an app target.

 ${kc} Take one before your first coordinate-based action and whenever the screen may have changed.

 An app-target screenshot is the image. When a live snapshot is still fresh it reuses that snapshot_id and does not walk. If snapshot_id is present, do not call computer_app_state. Otherwise the image returns without a tree; call computer_app_state when you need element ids.

Captured a screenshot.

Captured a screenshot.

Click at a screen coordinate.

Locate the target in a fresh screenshot first.

the click

Mouse button. Defaults to left.

Click count; 2 double-clicks. Defaults to 1.

${s.text} ${a}

Move the mouse cursor to a screen coordinate without clicking.

Useful for revealing hover UI.

the move

Moved the cursor.

Moved the cursor.

Press the mouse button at one coordinate, drag to another, and release.

Locate both points in a fresh screenshot first.

the drag

Defaults to left.

Mouse button to hold. Defaults to left.

Type into a macOS app. Prefer computer_set_value or computer_type with element_id. Without element_id, inserts into the click-selected field via AX, or the focused field. A missed click target is refused. If nothing is focused, click a field first. Does not raise or move the real cursor. Screen-scope typing is refused. The result is text only.