books
Pack: books
Reading — what you’ve read, what you’re reading, what you abandoned, and what you might read next.
What it adds
| Rules | agents-domain-books.md — page types, the series threshold, rating precision, dispatch flow |
| Skills | book-capture — capture a recommendation, collision-check it against your library, file it as a ranked candidate |
| Folders | wiki/books/authors/, wiki/books/series/, wiki/books/candidates/ |
| Index | wiki/index-books.md + three sub-indexes (authors, series, candidates) |
| Schema | Yes — adds author, series and candidate page types |
| Subdomains | None |
Install
- Add a dispatch row in
agents-core.md§ 3 pointing atdomains/books/agents-domain-books.md - Add
booksto the domain list inagents-core.md§ 4 - Add
author,series,candidateto thetype:enum inagents-page-conventions.md mkdir -p wiki/books/authors wiki/books/series wiki/books/candidates- Create
wiki/index-books.mdplus the three sub-indexes - Copy
skills/book-capture/into your tool’s skills location
Or ask your agent: “install the books domain pack.”
Why this pack is worth reading even if you don’t want it
It’s the proof that domains can extend the schema. The core four page types genuinely couldn’t hold this: a series isn’t an entity, an author needs a table of works rather than prose, and a book you might read is a different kind of thing from one you’ve read.
It also carries the clearest example in benchbook of a rule that changed because measuring it proved it wrong.
The series-page rule started as “create one when the series has 2+ published volumes.” That generated a 51-page backlog. Measuring the backlog showed the rule over-generated badly — the value of a series page depended entirely on the shelf status of your entry point into it:
| Entry point | Worth a page? | In the backlog |
|---|---|---|
to-read | Yes — reading order is actionable | 18 built |
read, rated 4+ | Yes — continuing is a live question | 4 built |
read, rated ≤3 | No | 16 skipped |
abandoned | No — and actively misleading in a to-read context | 10 skipped |
Two-thirds of what the naive rule would have generated was worthless. The refined rule is
in agents-domain-books.md, with the table.
Three other hard-won rules worth stealing regardless of domain:
- Never write a placeholder. After a bulk import, 169 of 208 author pages rendered a literal “To be filled in.” A placeholder on four-fifths of your pages isn’t a prompt, it’s noise you learn to skip. Either real content or no section.
- Imported and hand-entered values are different claims about precision. A
4from an import and a4.0you gave directly are not the same fact. Never back-fill a decimal to make a column look tidy. - Normalize before you group. Library exports tag the same series inconsistently (“Expanse” / “The Expanse”), which silently splits one series into two pages.