Expand description
Structural Markdown parser — headings become Section nodes, inline
code-spans and link text that look like identifiers become unresolved
References edges to code symbols. No LLM calls, no HTML rendering: pure
CommonMark structural extraction via pulldown-cmark’s event stream.
Doc→code references are intentionally cross-language (a README can
reference a Python function from a Rust repo’s Python bindings, etc.) —
see EdgeKind::References and resolve_deferred in indexer.rs, which
only scopes resolution by language when the source file’s extension is
recognised. Markdown isn’t, so resolution is a no-op pass-through there.