point-at
Use when about to ask the person a question, answer one of theirs, or explain something to them, about a screen, a region or a component of the interface: in an interview, a plan, a board, a review or
SKILL.mdskills/ui/point-at/SKILL.md
GitHubPoint at it
A screen is talked about with the screen in front of the person, in the prototype's own names, whether the agent is asking or answering.
- Find the prototype before writing the question, through the
aiviewskill:list --kind prototype --tag <project>, or the one the person named. Read the prototype that draws the screen, the screen's own file and never the library it pulls from, and runcomponentson it. Itspageandvariantslines are the vocabulary: every name the person can be shown on that page. - What the prototype already answers is stated and confirmed, not asked. Point at it all the same.
- Write the question in those names only. A region with no name of its own is placed by the named component it sits in. Where the code differs from the prototype, say both: the prototype's name, then the code's
file:line. A comparison with another screen points at the screen the decision changes, and names the other by its prototype and component. - Point, then ask: one
show <prototype>with--component <Name>once per component, and--variantwhen they exist only in one. Three components at most, the ones the question is about. A name the command refuses is not on that page: go back to 1, never rephrase around it.showwrites each name on the component it outlines, so the question uses the same names, and carries the link the command prints. - One pointer per question: each
showreplaces the last. Questions about different regions are asked one at a time, the one the others depend on first. - When the answer is in,
show --done: the person's tab goes back to what they were reading before the first pointer. Between two questions in a row there is no--done: it closes the run. - No prototype draws it: say so in the question and cite the screen's
file:line. A layout that exists nowhere goes to thedesign-prototypeskill first, never described in prose for the person to choose in.
An answer follows the same steps: the person asks where something is, what a name means or how two regions relate, and the reply points at it before it explains, with the link in the reply, and show --done once they have it. "Which one is that?" is answered by a show, never by a longer description.
tabs: 0 in the command's answer means no viewer is open: the link in the question is how the person gets there.