Update to World News Sources

{
  "description": "Find configured publisher RSS feeds or sitemaps by country, name, domain, language or category, and explain their cached endpoint health when requested.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 275
    }
  ],
  "name": "find-news-sources",
  "skill_md_contents": "---\nname: find-news-sources\ndescription: Find configured publisher RSS feeds or sitemaps by country, name, domain, language or category, and explain their cached endpoint health when requested.\n---\n\nUse the World News Sources MCP tools for source discovery and cached endpoint metadata. Follow the user's requested scope and output format.\n\n- Use `search_sources` for a source shortlist. Country codes, `endpoint_type` (`rss` or `sitemap`), query, language and category can narrow it. `enabled` is a configuration flag, not a health result.\n- Use `list_countries` when supported countries or configuration counts are requested. Continue with `next_offset` when another page is needed; counts are not unique publishers or live feeds.\n- Use `get_source` with an ID returned by search for details. Keep RSS and sitemap endpoints separate.\n- For requested health or working-source comparisons, use `get_endpoint_health` with returned endpoint IDs, at most 20 per call. If `remaining_endpoint_ids` is nonempty, request those separately. Do not invent IDs, URLs or results.\n\nPresent the source name, country, endpoint type and configured URL. Add the recorded status and actual `checked_at` when discussing availability. `healthy` means a recent structural XML check succeeded; inspect `audit_status`, `entry_count` and `entries_with_url` for empty documents or absent URL outputs. A cached result is not a new live check.\n\nKeep `newest_content_at` and `content_freshness` separate from `checked_at`. Missing dates cannot establish freshness. Report stale, unreliable, blocked, failed, unknown and incomplete results plainly. A sitemap success cannot establish RSS health. `sitemapindex` counts and sample URLs describe child sitemap entries; they do not establish article availability or child-file validity.\n\nUse `get_country_source_inventory` for active registration and distinct typed endpoint counts. Use `get_source_article_activity` or `get_country_article_activity` for cached first-discovered unique article-candidate URL counts. Country counts deduplicate across sources and must not be summed from source counts. Every new endpoint/source binding has an excluded baseline. Always report tracking start, initialization, latest collection, partial/untracked coverage and timezone. Before real records, null means not tracked; it is not zero. Read retained history in ranges of at most 31 calendar days; older stored dates remain queryable. Check history_available_since and snapshot_coverage. Missing/corrupt archive data is an error, not zero or never tracked. A legacy truncated inline snapshot can still report history_not_in_snapshot; do not describe that as never tracked. Calendar days use the requested IANA timezone (default UTC); rolling 24h is a separate mode. Counts are discoveries, not published articles or feed entry counts. Sitemap-index children and known uncertain navigation URLs are excluded from candidate NEW. No semantic duplicate detection is performed.\n\nThese tools provide source metadata and observation aggregates, not article search or full text. They do not fetch publishers during calls, crawl children or article links, change the registry, or run the operator audit. For unsupported requests, explain this capability boundary and offer relevant source discovery or cached results without claiming the unsupported action occurred.\n"
}