Add-on: JSONL Chunking + Citation Metadata

Use this skill to create the canonical intermediate representation:

• JSONL records (archival artifact in object storage)
• document_chunks rows (queryable in Postgres)

This is the foundation for extraction, validation, and audit-ready citations.

Inputs

Collect:

• CHUNK_TARGET_CHARS: default 1200.
• CHUNK_MIN_CHARS: default 400.
• CHUNK_OVERLAP_CHARS: default 120 (only if using sliding windows).
• CHUNK_STRATEGY: paragraph (default) or window.

JSONL Record Shape (Required)