Build and run agents you can see, understand and trust.
Python agentchatbotlarge-language-modelsllmllm-agentmcpmulti-agentmulti-modalreact-agentrealtime-agentvoice-agentvoice-assistant
import { useCallback, useEffect, useRef, useState } from 'react';
import { hubApi } from '@/api';
import type { SkillCard } from '@/api';
const PAGE_SIZE = 20;
// How many consecutive empty-but-not-final pages to walk past before
// giving up, so a hub that filters everything out cannot spin here.
const EMPTY_PAGE_LIMIT = 5;
/**
* Browse one skill hub's catalog.
*
* Pagination is cursor-based, so pages accumulate — there is no jumping to
* page N and no total count. `hasMore` is simply "the server handed back a
* cursor". Changing `hubId` or `query` starts over.
*
* Responses are matched against the request that is still current before
* being applied, so a slow first page cannot land after a newer search.
*
* @param hubId - Hub to browse, or `null` to stay idle (e.g. while the
* "mine" tab is selected).
* @param query - Keyword filter. Note that hubs may answer a search from a
* separate, unpaginated endpoint, in which case `hasMore` is false even
* when more matches exist upstream.
*/
export function useSkillHubCards(hubId: string | null, query: string) {
const [cards, setCards] = useState<SkillCard[]>([]);
const [cursor, setCursor] = useState<string | null>(null);
// Starts true when there is something to fetch: the first paint
// happens before the effect fires, and a false start would flash
// the empty state before the spinner.
const [loading, setLoading] = useState(hubId !== null);
const [loadingMore, setLoadingMore] = useState(false);
const [error, setError] = useState<Error | null>(null);
// Bumped on every hub/query change; in-flight responses for an older
// request id are discarded rather than applied out of order.
const requestId = useRef(0);
const load = useCallback(
async (append: boolean, fromCursor: string | null) => {
if (!hubId) {
setCards([]);
setCursor(null);
return;
}
const id = append ? requestId.current : ++requestId.current;
if (append) {
setLoadingMore(true);
} else {
setLoading(true);
// Nothing is known about the next page yet; leaving the
// previous query's cursor in place would show "Load more"
// while this one is still in flight.
setCursor(null);
}
setError(null);
try {
// A hub may hand back an empty page that still carries a
// cursor — ClawHub filters after fetching, so its very
// first page is often empty. Taking that at face value
// renders "nothing found" over a catalog that has plenty.
let next = fromCursor;
let fetched: SkillCard[] = [];
for (let attempt = 0; attempt < EMPTY_PAGE_LIMIT; attempt += 1) {
const page = await hubApi.skill.listCards(hubId, {
q: query || undefined,
cursor: next ?? undefined,
limit: PAGE_SIZE,
});
if (id !== requestId.current) return;
fetched = page.cards;
next = page.next_cursor;
if (fetched.length > 0 || next === null) break;
}
setCards((prev) => (append ? [...prev, ...fetched] : fetched));
setCursor(next);
} catch (e) {
if (id !== requestId.current) return;
setError(e as Error);
} finally {
// A superseded response must not end the spinner — the
// request that replaced it is still in flight.
if (id === requestId.current) {
if (append) setLoadingMore(false);
else setLoading(false);
}
}
},
[hubId, query],
);
useEffect(() => {
load(false, null);
}, [load]);
/** Append the next page. No-op while a page is already in flight. */
const loadMore = useCallback(() => {
if (!cursor || loading || loadingMore) return;
load(true, cursor);
}, [cursor, loading, loadingMore, load]);
return {
cards,
loading,
loadingMore,
error,
hasMore: cursor !== null,
loadMore,
refetch: () => load(false, null),
};
}