Now extracts page content as markdown and injects it directly into the LLM's input field for better summaries. Falls back to URL-only when extraction fails. Updated dependencies.