Markdown intelligence
Owns committed Markdown indexing and discovery within one selected project. Documents retain heading hierarchy, bounded chunks, frontmatter metadata, GitHub anchors, and explicit source links. Search uses local weighted FTS, not vectors.
AI/JSON CLI tools expose search, inspect, and expand only. Parsing runs through explicit app maintenance or enabled post-commit hooks. Unindexed link targets keep their status and source locations without offering a mutating parse follow-up. Durable context belongs in the owning README/docs only when useful; routine edits do not need post-task notes or duplicate implementation details.
Source map
schemas.ts— Typed tools, results, documents, sections, links, and diagnostics.parser.ts— CommonMark/GFM parsing, heading identities, source chunks, YAML metadata, GitHub anchors, and explicit links.snapshot.ts— Committed inventory, project ownership/exclusions, dirty-input and revision guards, and Git blob reading.store.ts— Versioned snapshots, SQLite/FTS, and atomic generation guards.queries.ts— Ranked local search, bounded section inspection/traversal, and live indexed-target link/backlink resolution.service.ts— Incremental parsing and independent attempts.
The plugin's discovery.ts combines code and Markdown discovery. Shared search
cursors preserve separate offsets; revision changes invalidate pagination.
Source Markdown remains in its owning repository. Memory stores a searchable
committed snapshot, not a competing editable documentation source.
See Code intelligence, the Memory overview, and the design/progress record.