/design-critique
Critique a UI on the visual axis — how it looks. This is a thin launcher over the
design-critic agent: resolve its inputs, spawn it, and relay its findings — the agent owns the
lens and the capture, so don't re-derive them here. (For the usability axis — flows, friction, comprehension — use
/product-critique; for both at once, /critique-committee.)
Do this
Resolve the inputs — ask the user only for what's genuinely missing:
- Target — the Figma URL, an already-captured screenshot path, or a reachable UI (route /
dev-server command). If the caller already holds a screenshot — e.g. a feedback loop
(
web-visual-loop/electron-visual-loop) handing off its annotated capture — pass that PNG path asSCREENSHOTso the agent judges it directly and does not re-drive a loop. - Context — what is this, who is it for, what stage (exploration / refinement / final)?
- How to reach the UI — only when no screenshot is supplied and the UI renders live: the route/state and how to launch it; the agent decides web vs electron vs desktop and captures the screenshot itself.
- Target — the Figma URL, an already-captured screenshot path, or a reachable UI (route /
dev-server command). If the caller already holds a screenshot — e.g. a feedback loop
(
Spawn the agent, foreground:
Agent({ subagent_type: "design-critic", description: "design-critique <target>", prompt: ` TARGET: <Figma URL | reachable UI | screenshot path> SCREENSHOT: <pre-captured PNG path, or "none — capture it yourself"> CONTEXT: <what it is, who it's for, what stage> HOW TO REACH THE UI: <route / launch command> ` })Relay its findings — grouped as the agent returned them. An empty findings array is a valid result; report it as "no visual issues found" rather than inventing problems.
If design-critic isn't installed, say so and stop — don't critique inline; the lens and capture
mechanics live in the agent.