Community catalog always empty: scraper looks for <article>, site now serves <div>

Description

getCatalogItems() always returns an empty array, so Skill Finder → Community Skills & Agents shows "0 curated from 0 catalog items" for everyone, on every harness. No error is surfaced — the fetch succeeds and the parse silently matches nothing.

The catalog site changed its markup. Each entry is now a <div class="resource-item">; the scraper in src/webview/panel-catalog.ts:43 still looks for <article>:

const articleRegex = /<article\s+class="resource-item"[^>]*data-path="([^"]*)"[^>]*>([\s\S]*?)<\/article>/g;

Against the live site today:

$ curl -s https://awesome-copilot.github.com/skills/ | grep -oE '<(article|div) class="resource-item"' | sort | uniq -c
 418 <div class="resource-item"

418 entries on that page alone, zero of them <article>.

Steps to Reproduce

1. Open the dashboard → Skill Finder
2. Click Analyze
3. The Community Skills & Agents section reports 0 items

Or directly:

import { getCatalogItems } from './src/webview/panel-catalog';
getCatalogItems().then(i => console.log('items:', i.length));   // items: 0

Expected Behavior

Catalog items are fetched and listed.

Notes toward a fix

Two things need changing, not just the tag name:

1. The entry tag. <article> → <div>, ideally accepting both so it survives a revert.

2. The block delimiter. Entries now contain nested <div>s, so a non-greedy [\s\S]*? up to </div> stops at the first inner close tag. The captured block ends mid-element, titleMatch fails, and if (!title) continue drops every item — so simply swapping the tag name still yields 0. Delimiting on the start of the next entry works:

const articleRegex = /<(?:article|div)\s+class="resource-item"[^>]*data-path="([^"]*)"[^>]*>([\s\S]*?)(?=<(?:article|div)\s+class="resource-item"|$)/g;

With that, all 833 items across the four pages parse, descriptions included.

A note on robustness rather than the bug itself: because the parse failure is silent, this went unnoticed while looking like "no matches for your workflow". A zero-item result after a successful fetch is more likely a parse break than an empty catalog, and saying so in the UI (or logging it) would surface the next markup change immediately.

Separately, hooks 404s — https://awesome-copilot.github.com/hooks/ does not exist, so that fetchCatalogPage call is always wasted. Harmless today since !response.ok returns [], but it may be a leftover.

Extension Version

0.1.0 (reproduced against main as of 2026-09-15)

Read src/webview/panel-catalog.ts at getCatalogItems and run the direct reproduction described in the issue. Update the parser so current resource-item div entries, with nested divs and article compatibility, are captured; done means catalog items and descriptions are returned instead of an empty result.