ghosthand scroll
Usage
ghosthand scroll <up|down|left|right> <ticks>
Options
| Option | Description |
--at [x,y] | Scroll over this point (default: screen center) |
--screenshot [path] | Save a screenshot after the action once the screen stops changing (default: screenshot.png); not available while locked |
--screentext | Print the OCR text of the screen after the action (x,y wxh text, after --screenshot if both); uses --language |
--language [tag] | OCR language, e.g. it-IT; en-US is always included. Env: GHOSTHAND_LANGUAGE |
--scale [n] | Image and coordinate size as a fraction of native pixels (default: screen points, 1/2 or 1/3; 1 = native). Env: GHOSTHAND_SCALE |
--wait-text [text] | After input, poll real OCR for this case-insensitive substring; never repeat input |
--timeout-ms [ms] | With --wait-text: verification deadline, 1...60000 ms (default: 5000); starts after input completes |
Also takes the global options.
Details
down shows content further down (like a finger swipe up); up shows content above.
right shows content further right (carousels, wide tables); left goes back.
- One tick moves about 100 points in Safari, about 115 in Settings-style lists (UITableView), measured on an XR (896 points high, so
scroll down 8is about one screen). These are estimates: check the result with --screentext. - Faster than
swipefor long distances: swipes have no momentum. - --at: the content under this point scrolls (for a list inside a page or sheet).
- ticks 1-1000. At the end of the content nothing moves; compare --screentext.
- Date and picker wheels:
scroll up 2 --at x,yover the wheel moves about 3.3 rows per tick (XR, iOS 18.7). Scroll near the target, tap the row, read the result.
Examples
$ ghosthand scroll -l 42 down 5 --screentext $ ghosthand scroll -l 42 up 3 --at 207,500 --screenshot s.png
MCP tool
The same command is the MCP tool
scroll; flags are arguments.