| name | retriva-markdown-injector-patterns |
| description | Guidance for building a local Markdown injector for Retriva. |
Retriva Markdown Injector Patterns
Core input shape
The input is a recursive directory containing one or more Markdown files, typically .md or .markdown.
Preferred processing flow
- recursively scan the root directory
- detect candidate Markdown files
- parse raw Markdown text while preserving structural landmarks
- derive document title and section hierarchy from headings where practical
- emit document/chunk objects compatible with the existing indexing pipeline
- keep enough metadata for citations and debugging
Guardrails
- Do not require rendering Markdown to HTML just to ingest it
- Preserve heading hierarchy and section anchors when useful
- Keep code fences distinguishable from prose in metadata or chunking decisions