Skip to content

Agent skill & CLI

Drive a running app from the terminal: navigate, click, type, and take screenshots. Built for AI agents and tests.

Every vinland-ui app can expose a local control port. vui-ctl speaks to it, which gives agents and tests the same feedback loop a person gets from looking at the screen.

# render offscreen at a fixed scale; nothing appears on your desktop
VUI_HIDDEN=1 VUI_SCALE=2 VUI_SIZE=1440x900 ./build/showroom &
./build/vui-ctl info
./build/vui-ctl nav page=box-shadow
./build/vui-ctl dark on
./build/vui-ctl screenshot shadow-dark.png
./build/vui-ctl click-text "Show all"
./build/vui-ctl scroll 600
./build/vui-ctl tree 6 | jq '.nodes[] | select(.t)'
Class Effect
info viewport, scale, frame time, node count, unknown classes
screenshot <png> [x y w h] writes the last frame (optionally cropped)
click <x> <y> [right|double] synthetic click in logical px
click-text <text> [nth] clicks the element whose text matches
move / scroll / type / key pointer, wheel and keyboard input
resize <w> <h> / dark [on|off] window size and color scheme
tree [depth] element tree with rects, classes and text as JSON
settle / wait [frames] block until animations finish
<app command> key=value forwarded to App.on_command (nav, pages, search...)

The repository ships skills/vinland-ui/SKILL.md, a skill for coding agents. It teaches the element API, the utility vocabulary, the token system and this CLI loop, so an agent can build a screen, screenshot it and iterate without a human in the loop.