computer

computer — look at and poke a screen

computer is the look-then-act loop for every registered screen. Prefer it over v86 type|key|mouse|screenshot|text once a guest is running. v86 start still boots the VM and registers v86:<name>.

Target

computer ls                         # ids, kinds, names, titles
computer use v86:arch               # default for later verbs
computer info                       # descriptor for the current target

Resolution order: -c / --computer, else $COMPUTER, else last computer use, else the only registered computer.

-c takes an id, the -n name the computer was added under, or its current title — in that order. Prefer the id or the name: ls shows them in separate columns because TITLE is whatever the surface calls itself right now (a tab republishes document.title on every capture), while NAME is fixed for the life of the registration.

Look then act

Every poke writes a frozen JPEG and prints target: <id> then screen: <path>. The extension of that path follows the bytes, so it is .jpg in practice and .png only if a backend hands back PNG the runtime could not transcode — read the extension, do not assume it. The target: line is the resolved computer (-c, $COMPUTER, last use, or the only registered one) so a later computer screenshot without -c still binds the bash-row UI. That poke frame is a transcript side effect only — it does not redefine the coordinate space, so never read new coordinates off it (it can come back narrower and your clicks would map to the wrong place). Run a fresh computer screenshot before choosing coordinates for the next click. If the frozen frame cannot be captured, the poke still exits 0 (it already happened); do not retry the same click.

computer screenshot                 # JPEG; prints WxH → wxh (scale s)
computer screenshot --size high out.jpg
computer text                       # text-mode dump when the backend supports it