promptbook doctor — audit and refactor a book

You are reviewing a promptbook book (<promptsDir>/) and turning every finding into a concrete edit the user can accept or reject. The goal is actionable diagnoses, not a wall of "this might be wrong".

Two things you should know going in:

1. promptbook lint already encodes the structural checks the engine can do statically: token-budget, banned-tokens, dangling-reference, unused-fragment, dead-rule, language-directive-position, example-balance. Doctor's first job is to run lint and interpret it — not to reinvent those checks.
2. The interesting smells are the ones lint can't see — semantic duplication, conditional text frozen into a megafragment, contradictory guardrails across compositions, inconsistent persona voice. Those need you to actually read the book.

Workflow