deep-read
deep-read
Deeply understand ONE large document, paper, or book. Where
deep-research goes broad (many sources, adversarially fact-checked),
deep-read goes deep on a single source — it builds a structured
understanding of that one document: its sections, claims (each grounded
in a verbatim quote), methodology, caveats, and argument structure.
This skill reuses exactly two of deep-research's extract primitives —
schemas.py (EXTRACT_SCHEMA) and prompts.py (fetch_prompt) — as
byte-identical copies (kept in sync with the deep-research SSOT by the MD5
CI check). It carries neither rank.py nor dedup.py: deep-read has no quorum
step and its cross-chunk claim merge rolls its own claim-text dedup, so copying
them would be dead code. It adds a stdlib chunker (chunker.py) plus
the deep-read-specific schemas + cross-chunk synthesis + report renderer
(deepread.py).