▶
ExecutionTrigger
Automation/Computer/Capture
Observes a window or display for an agent: a screenshot, the same screenshot with numbered boxes on every actionable element, and a compact element list (accessibility tree plus optional OCR text) whose ids Click Screen Element accepts
Scores range from 0 to 10. Higher values mean more impact, exposure, or operational weight.
Trigger
Computer session handle
Browser context info if browser is attached
Current page info if a page is open
Visible text contained in the element; script, style and head content is ignored.
ARIA role, optionally filtered by accessible name: `button`, `button|Sign in` (case-insensitive substring), `button|=Sign in` (exact) or `button|/^sign/i` (regex).
Element ref from the latest Browser Snapshot, such as `e12`.
focused_window: the focused window outside Flow-Like; window: the window titled below; display: a whole display
Title (or part of it) of the window for the window target
Display for the display target
List elements from the accessibility tree
Add recognized text that no accessibility element already covers
Only list controls that can be clicked, typed into or toggled
Upper bound on listed elements (1–500)
Comma-separated OCR languages such as en-US, de; empty lets the OS engine choose
Continue
Computer session handle (pass-through)
Browser context info if browser is attached
Current page info if a page is open
Visible text contained in the element; script, style and head content is ignored.
ARIA role, optionally filtered by accessible name: `button`, `button|Sign in` (case-insensitive substring), `button|=Sign in` (exact) or `button|/^sign/i` (regex).
Element ref from the latest Browser Snapshot, such as `e12`.
Screenshot of the target
Screenshot with a numbered box on every listed element
Desktop rectangle and pixel size of both images
Listed elements with role, name, value, states and desktop bounds
1-based number drawn on the annotated image.
`ax` (accessibility tree) or `ocr` (recognized text).
Visible part of the element.
One line per element for a language model, e.g. [3] button "Save" @ (812,433)
Identifies this capture; Click Screen Element rejects ids from older captures