• Hardened Markdown escaping for links, images, footnotes, callouts, code blocks, and math, preventing page content from becoming unintended markup or unsafe links (#396).
• Fixed lazy-loaded images using explicit source attributes, preventing filenames and unrelated metadata from replacing image URLs (#388).
• Preserved screen-reader text, including mixed numbers and link labels, while removing redundant link annotations (#390).
• Fixed extraction on pages with special characters in IDs and classes, including Tailwind classes and React streaming content. Kept generated selectors stable after shadow DOM and streamed-content changes (#392).
• Fixed whole-page fallbacks overriding successfully extracted content during retries, and allowed async extractors to run before falling back to the page body (#358).
• Added caught extraction errors to debug.errors and automatic content detection when contentSelector is invalid (#358).
• Shared math processing between core and full builds, fixed element creation in cloned documents, and added direct tests for both variants.
• Removed unused code and added unused-code checks to CI.