Print the following help text exactly, then stop. Do not add commentary.
Book Review Pipeline
Two skills, one feedback loop:
/book-review — review any part of the book
/book-refine — improve the review system based on your feedback
Reviewing
/book-review review uncommitted changes
/book-review book/src/guide/gadgets/ review a whole section
/book-review book/src/protocol/core/nark.md review a specific chapter
/book-review book/src/protocol/core/nark.md:40-80 focus on specific lines
/book-review book/src/protocol/prelim/bulletproofs.md book/src/protocol/core/nark.md
review multiple files + cross-file consistency
/book-review "accumulation schemes" find and review a topic wherever it appears
/book-review "the transition from preliminaries to core construction"
review a conceptual boundary across chapters
Launches one parallel agent per policy file in .claude/book-review/. Currently:
grammar, structure, math. Results are synthesized and presented as
must-fix issues first, then suggestions.
Refining
After a review, give feedback in natural language, then:
/book-refine use feedback from the conversation /book-refine "stop flagging passive voice in definitions" /book-refine "I'd write 'the prover' not 'a prover'" /book-refine "add a reviewer for code example accuracy"
This enters plan mode — it reads all policies, generalizes your feedback into
a principle, audits for duplication/contradictions, and shows you the proposed
changes before executing. New policy files are auto-discovered by /book-review.
Policy files
.claude/book-review/standards.md master standards (all reviewers) .claude/book-review/grammar.md prose quality, voice, terminology .claude/book-review/structure.md organization, flow, progressive disclosure .claude/book-review/math.md notation, correctness, accessibility
Edit these directly or let /book-refine manage them.