| name | gstack-plan-design-review |
| description | Use when reviewing a wireframe, mockup, prototype, or finished UI; when launching a new surface; or when the user asks "is this design good?". Triggers: "review the UX", "is this design good", "pressure-test the mockup", "is the 200ms hook right", "is the empty state designed", "is the primary CTA clear", "what does the user see first". |
Plan — Design Review
The third leg of the plan-review tripod (CEO, Eng, Design). Design here means the user's experience of the thing, not "is the typography good."
When to run
- When a new surface / screen / workflow is being introduced.
- When the user is second-guessing a UI.
- When a competitor has a clear visual / interaction advantage and you don't.
- When the agent is about to generate UI from a description and you want a checklist.
The 9 questions
1. What does the user do here, in one sentence?
If you can't say it in one sentence, the screen is doing two things. Most bad UIs are screens that can't pick a job.
2. What does the user see in the first 200ms?
Above-the-fold, pre-scroll, before they read. If the most important thing isn't there, the design has failed before the user decides to engage.
State: "In the first 200ms, the user sees X, Y, Z. The most important of these is X because..."
3. Where is the one primary action?
Every screen has one action the user should take. If there are five equal-weight CTAs, the user takes none.
State the primary action. The other actions can exist, but they should be visually secondary.
4. What is the cost of being wrong here?
If a wrong click destroys data, the confirmation flow is part of the design, not an afterthought. If a wrong click does nothing, the affordance can be loose.
Design for the cost of the action, not the type of the action.
5. Is the system feedback synchronous?
When the user does something, do they see a response immediately? Spinner, optimistic update, transition, sound — anything that says "I heard you."
The failure mode of AI agents is silence. The user clicks, the system thinks for 4 seconds with no signal, and assumes it's broken. The design's job is to fill the silence.
6. Does it scale across screen sizes?
A design that only works on a 27" monitor or only on a phone is half a design. State the breakpoints and the behaviour at each.
7. Is the empty state designed?
What does the user see when there's no data? "No results" is not a state. The empty state is the first state for new users — it should be inviting, instructive, and demonstrate the product's value.
8. Does it respect the user's expertise?
If the user is a power user, don't hide the keyboard shortcuts. If the user is new, don't dump 14 settings panels on them. The design should match the user's stage, not the designer's.
9. Can the user undo?
If the user can destroy work, the design needs an undo. If the user can confuse themselves, the design needs a back button. If neither is true, the design doesn't have a permanence problem and can be bold.
Anti-patterns to call out
- Modal for everything — modals interrupt. Use them only for "stop and confirm" moments.
- Settings as a feature — every visible setting is a decision the user has to make. Default aggressively.
- Animation that lies — if the spinner stops, the work is done. If the work is done but the spinner is still going, the user loses trust. Be honest with motion.
- Loud and quiet in the wrong places — the brand shouldn't compete with the content.
- Inconsistency — same shape button, same colour hover, same padding everywhere. Inconsistency is the design tax users pay without knowing it.
Output format
DESIGN: <name / screen>
JOB: <1 sentence>
200MS: <what the user sees first>
PRIMARY CTA: <the one thing the user should do>
COST OF WRONG: <what happens on a bad click>
FEEDBACK: <how the system responds in <100ms>
SCALE: <behaviour at phone / tablet / desktop>
EMPTY STATE: <what new users see>
EXPERTISE: <new / mixed / power user>
UNDO: <how the user reverses an action>
VERDICT: <ship / cut / re-scope / re-design>
Source / license
Adapted from JZKK720/gstack · plan-design-review, MIT.
