v0.6.1

full 98-test suite (markdown + html + index) stays green.

• All 44 existing markdownToFormattable tests still pass, and the

  including the original LLM-style enumerated answer repro.

• 8 new regression cases in the "block separators" describe block,

Tests:

logic, so it's removed.

The previous workaroundForHeading is a strict subset of this

paragraph+blockquote, paragraph+code, heading+anything, etc.

the current block's trailing newlines. Covers paragraph+list,

token isn't already a space, emit a synthetic space carrying

normalizeBlockSeparators. For each non-space block whose next

Fix: generalise the previous heading-only workaround into

markdownToFormattable.

assistant reply render wrong in Telegram bots that use

which LLM-generated text almost never has — making every enumerated

Only happened when the source had no blank line between blocks,

"Agenda:- one\n- two" ← wrong

an empty separator glued adjacent blocks together:

drops that trailing whitespace, so joining top-level tokens with

processToken reads the structured tokens/text fields and

list raw="- one\n- two"

paragraph raw="Agenda:\n" (trailing \n = 1)

For example, "Agenda:\n- one\n- two" parses to:

fix(markdown): preserve newlines between adjacent block tokensMarked stores a block's trailing newlines on its own raw field.

chore: bump version to 0.6.1
Full Changelog: v0.6.0...v0.6.1