Jack
586776998b
markdown-it runs with html: false, so raw HTML in a message body is escaped and shown as literal text. Agents legitimately use <details>/<summary> to hide the long tail of a reply, and it rendered as a wall of visible tags. MarkdownRenderer now splits the message into plain segments and <details> blocks, renders the plain segments through markdown-it as before, and builds real <details> elements from the blocks. Nested blocks are handled by recursion; an unclosed tag does not swallow the rest of the message. The html option stays false on purpose: enabling it would let model output inject arbitrary markup into v-html. Every fragment still goes through markdown-it, so nothing unsanitised reaches the DOM. Group chat shares this component, so it picks the change up too.