Compound Component Pattern
Let related parts of a component share state while the caller keeps control of the arrangement.
What it adds
Groups of subcomponents that coordinate shared state internally while leaving structure and ordering to the call site.
What your agent is told to do
5
What your agent is told to do
5-
1
Identify the components currently configured through long lists of props for their internal parts — a tabs component taking an array of labels, a dialog taking title, body, and footer as strings — and let the caller compose those parts instead.
-
2
Keep the shared state inside the group. Selection, open state, active index, and registration of the parts are the group's responsibility, and callers should not have to wire them together to get correct behaviour.
-
3
Support both a self-managing mode and a caller-controlled mode for the shared state, and make them mutually exclusive rather than allowing a half-controlled configuration that fights itself.
-
4
Give each subcomponent a clear error when it renders outside its group, naming the parent it requires. A part that silently renders inert because it found no shared state is very hard to diagnose from a screenshot.
-
5
Do not couple the parts by inspecting children, cloning them, or reaching for a particular position in the child list. That breaks the moment someone wraps a part in a conditional, a layout element, or a component of their own — which is exactly the flexibility this pattern exists to allow.
Edge cases it handles
7
Edge cases it handles
7- A subcomponent rendered outside its required parent must fail with a message naming that parent rather than rendering as an inert element. Silence here produces a control that looks right and does nothing.
- Reordering or wrapping the parts must not break accessibility. The relationships that assistive technology relies on — which label belongs to which panel, which trigger controls which region — must be established through explicit association rather than through the order the parts happen to appear in.
- Shared state passed implicitly is easy to depend on by accident. Keep what the group provides deliberately small and documented, so a caller cannot end up relying on internals that change in the next release.
- Both a self-managing and a caller-controlled mode must be supported, and switching between them mid-life must be refused or handled explicitly rather than leaving two sources of truth disagreeing about which item is selected.
- Parts rendered conditionally, in a loop, or through an intermediate wrapper must still register with the group, so keyboard navigation and roving focus cover exactly the parts that are present.
- Nesting one group inside another — a menu within a menu, tabs within a panel — must resolve to the nearest enclosing group, not the outermost one.
- Duplicated or missing required parts, such as two panels claiming the same identifier or a trigger with no matching panel, need a development-time warning.
Definition of done
9
Definition of done
9- Related components are composed at the call site rather than configured through arrays or slot-shaped prop lists.
- Shared state is owned by the group, and callers get correct behaviour without wiring it themselves.
- A subcomponent used outside its parent raises an error naming the required parent.
- Self-managing and caller-controlled modes are both supported and cannot be mixed.
- Wrapping, reordering, or conditionally rendering the parts leaves accessibility relationships intact.
- The group's implementation never inspects, clones, or indexes into its children.
- Nested groups resolve to the nearest enclosing parent.
- The feature matches the existing design system.
- No existing functionality is broken.
Related features
Avatar Group
Avatar Group
Show who is involved in a compact space without a pile of broken images.
What it does
A stacked set of member avatars with deterministic ordering, an overflow count, and access to the full list.
How it works
- 1 Find the places that show several people at once — assignees, collaborators, workspace members, comment participants — and render them through one grouped component.
- 2 Order the stack deterministically by a stable rule such as role then name, so the same group renders in the same order on every load and every device.
- 3 Cap the number shown and collapse the remainder into a single count that is announced meaningfully, not as a bare number sitting after a row of pictures.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/avatar-group
Screen Reader Only Utility
Screen Reader Only Utility
One tested way to add text for assistive technology without it appearing on screen.
What it does
A single visually hidden utility used for labels, instructions, and context that assistive technology needs and sighted users do not.
How it works
- 1 Find every place the app currently hides text from sight — off-screen positioning, zero dimensions, transparent colour, tiny font sizes, aria-label strings duplicating nearby copy — and consolidate them onto one utility.
- 2 Build it by clipping the element to nothing while keeping it rendered and in the accessibility tree: it must occupy no visual space and contribute nothing to layout, yet still be read in document order.
- 3 Provide a second variant that becomes visible when it receives focus, and use it for skip links and any other hidden content that is itself focusable.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/screen-reader-only-utility
Landmark Shell
Landmark Shell
Give every page the same navigable regions so people can jump straight to the content.
What it does
A reusable application shell providing labelled banner, navigation, main, complementary, and contentinfo regions on every route.
How it works
- 1 Find the layouts the app already renders and identify what each region actually is — the top bar, the primary navigation, the page body, any inspector or sidebar, the footer — then express them as landmarks once in the shell rather than per page.
- 2 Give the shell exactly one main region per rendered page, and make it the target of a skip link that is the first focusable thing in the document.
- 3 Label every landmark that appears more than once so they can be told apart — primary navigation, breadcrumb, pagination — using a short distinct name rather than the word navigation repeated.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/landmark-shell
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.