Problem

The Markdown stdlib (md_parse, md_render, md_has_heading, md_has_code_block, md_extract_code_blocks) covers the boolean-question and code-extraction use cases well, but leaves structural traversal entirely to the user. Extracting a heading's text, iterating document blocks, or collecting all links requires deeply nested pattern matches over the full ADT tree:

MdDocument([MdBlock...]) → MdHeading(Int, [MdInline...]) → MdText(String)

This is the same boilerplate pattern as the JSON field-extraction problem — every Markdown-consuming program will rediscover it.

Proposed additions

-- Unwrap MdDocument into its block list; identity on other constructors
md_blocks(@MdBlock.0)                 -- Array<MdBlock>

-- Extract plain text from a single MdInline node
md_inline_text(@MdInline.0)           -- String

-- Extract plain text from a single MdBlock (concatenates inline text)
md_block_text(@MdBlock.0)             -- String

-- Extract all headings as (level, text) pairs
md_extract_headings(@MdBlock.0)       -- Array<Tuple<Int, String>>

-- Extract all links as (url, display_text) pairs
md_extract_links(@MdBlock.0)          -- Array<Tuple<String, String>>

-- Filter a document to blocks of a given constructor name
md_filter_blocks(@MdBlock.0, "MdHeading")   -- Array<MdBlock>

Motivation

The three dominant Markdown agent tasks are: (1) extract structured content (headings, links, code); (2) check content properties; (3) render to string. The stdlib covers (2) and (3) well. md_extract_code_blocks shows the right pattern for (1) but only covers one constructor. The proposed additions complete the set for the most commonly needed constructors.

Implementation

Pure Vera prelude functions — recursive traversal over the MdBlock/MdInline ADTs. No new WASM host imports. The tricky case is md_inline_text for MdEmph/MdStrong (which wrap Array<MdInline>) — needs array folding to concatenate.