UX Edge-Case Checklist
Make every UI feature state what it does when things go wrong before it counts as done.
What it adds
A required list of non-happy-path behaviours — failure, empty, loading, offline, permission, mobile, and accessibility — answered per feature.
What your agent is told to do
5
What your agent is told to do
5-
1
Attach the checklist to the point where work is proposed and to the point where it is reviewed, so the answers exist before the feature is built rather than being reconstructed afterwards.
-
2
Require a concrete answer for each category: what the empty state says and what it offers, what appears while data is loading, what the user sees when the request fails, and what a user without permission gets instead.
-
3
Make each answer an outcome someone else could verify. A feature that says errors are handled has not answered; one that says a failed save keeps the form open with the values intact and a retry has.
-
4
Insist on recovery, not just display. For every failure the checklist names, state how the user gets back to a working screen — retry, undo, a route out, or a preserved draft.
-
5
Do not treat the checklist as a fixed form to be filled identically each time. Cut the categories that genuinely do not apply, say why in a line, and add the ones this particular feature needs — a rigid template invites the same empty answers on every feature.
Edge cases it handles
7
Edge cases it handles
7- Categories that do not apply must be marked as not applicable with a short reason, because a blank line and a considered exemption look identical once the feature has shipped.
- Concurrency belongs on the list: two people acting on the same record, the same person acting in two tabs, and a second action fired while the first is still in flight all need stated outcomes.
- Repeated actions need their own answer — a double-clicked submit, a held-down key, a rapidly retried request — and the answer is usually that the second attempt is absorbed rather than duplicated.
- Recovery is the half most often skipped: showing an error is not handling it, and the checklist should be considered unanswered until the route back to a usable state is written down.
- Vague answers are worse than none, because they pass review and provide no bar to test against — reject anything phrased as handled gracefully or shows a friendly message.
- Offline and slow are different failures with different answers, and a feature that only considers the fully offline case will spin forever on a bad connection.
- The checklist covers the shape of a feature's behaviour; a component's own internal contract belongs to Component Acceptance Criteria, and the two should not both try to specify the same keyboard handling.
Definition of done
9
Definition of done
9- Every UI feature carries answers for failure, empty, loading, offline, permission, mobile, and accessibility before it is built.
- Categories marked not applicable carry a stated reason.
- Each answer describes a verifiable outcome rather than a general assurance.
- Concurrent and repeated actions have explicit stated behaviour.
- Every named failure has a written route back to a working state.
- The checklist is adapted per feature, with categories added and removed deliberately.
- Review treats an unanswered or vague category as unfinished work.
- The feature matches the existing design system.
- No existing functionality is broken.
Related features
Theme Acceptance Criteria
Theme Acceptance Criteria
Set the bar every theme has to clear before it ships to anyone.
What it does
A release gate for light, dark, branded, and customer-supplied themes, covering contrast, first paint, tokens, and print.
How it works
- 1 Write the criteria against the theme layer, not against individual screens. The question is whether a set of token values is fit to ship, and the answer should not require walking every page by hand.
- 2 Require measured contrast for text, icons, borders, focus rings, and disabled content in every interactive state — resting, hover, focus, active, and selected — because a palette that passes at rest often fails on hover.
- 3 Cover the surfaces that sit outside the component library: chart series and their gridlines, map and image overlays, embedded documents, illustrations, and any asset shipped as a fixed-colour file that will not follow the theme.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/theme-acceptance-criteria
Component Acceptance Criteria
Component Acceptance Criteria
Give every shared component a written bar it must clear before anyone uses it.
What it does
A per-component specification of required behaviour, keyboard handling, responsive limits, and theming, written so each line can be checked.
How it works
- 1 Inventory the components other parts of the app depend on — buttons, inputs, selects, dialogs, tables, menus, toasts — and write criteria for those first rather than for every file in the tree.
- 2 Write each criterion as a statement of an observable end state, not a task. A line that says the component supports keyboard use is untestable; a line that says Escape closes it and returns focus to the trigger is not.
- 3 Cover both the controlled and the uncontrolled shape of every component that has one, including what happens when a parent changes the value underneath an interaction already in progress.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/component-acceptance-criteria
Visual Regression Testing
Visual Regression Testing
Catch unintended visual changes before they reach users, not after.
What it does
An automated comparison of rendered screenshots against approved baselines, run on every change.
How it works
- 1 Identify the surfaces where a silent visual change would actually hurt — the shared components, the pricing table, the checkout, the dashboard — and capture those rather than screenshotting every route in the app.
- 2 Freeze everything that varies between runs before capturing: load fonts and wait for them, fix the clock to a known instant, disable animations and transitions, and render from seeded fixtures rather than live data.
- 3 Capture each surface in every theme and at every viewport the app claims to support, and label each capture with that combination so a diff says exactly which one moved.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/visual-regression-testing
How it works
-
1
Copy the link
Grab the Markdown instruction URL for this feature.
-
2
Give it to your AI
Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.
-
3
It inspects, then implements
Your agent reads your existing app first, then adds the feature to fit it.
Works with your stack
These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.
Need it tighter than that? Customize the feature and tell it exactly what you're running.