Keep a details fold written in Markdown
An agent can now fold the work behind an answer from Markdown. A <details> block typed in Markdown reaches Basecamp as a fold, instead of showing its tags as text.
Basecamp's rich text allows details and summary (bc3 doc/api/sections/rich_text.md), and posts written as HTML already fold. The Markdown path escaped both tags, so the only way to fold was to hand-write the whole post as HTML, which the basecamp skill tells agents not to do. In 392 agent posts on 37signals' Basecamp, only 5 folded, all of them HTML, and one post typed in Markdown showed <details> as literal text.
What changes:
richTextTagskeepsdetailsandsummary. The Markdown between them still converts, the way it does in any other kept HTML block.- The basecamp skill says how to write a fold. The skill is still under its 16 KiB cap: I shortened the Mermaid paragraph to make room.
- The connector's subagent instructions say how, and when: only when a reply combines its answer with the details of the work behind it, never folding the answer, an ask or a blocker.
Tests: two new cases, one fold with Markdown inside and one written inline. Both fail with the tag change reverted. The old "unsupported inline tag is text" case now uses <abbr>. bin/ci passes locally.
One thing to know: the separators the Markdown path adds put an empty line just inside the top and bottom of an open fold.
Paired with basecamp/coworker#325, which tells agents when to fold. For a person to review and merge.