| name | book-note-quality-ops |
| description | Evaluate and improve public-ready book recommendation notes before they are published to a catalog, reading radar, newsletter, or agent skill. Use when checking whether a book summary, recommendation rationale, reader fit, personal-note synthesis, or copyable markdown is substantive, public-safe, and distinct from raw import metadata. |
| memory_tags | ["domain:books","workflow:public-note-quality","skill_role:evaluator","repo_boundary:tools","inputs:markdown","outputs:quality-audit","risk:medium"] |
Book Note Quality Ops
Trigger
Use when:
- evaluating public-ready book recommendation markdown
- deciding whether imported/private reading notes are ready for publication
- checking a book catalog, reading radar, newsletter, or agent skill for thin or generic book descriptions
- rewriting a book note after a user says the summary, recommendation, or notes are not good enough
Do not use when:
- the user wants raw Apple Books/iBooks extraction; use the Apple Books export skill
- the user asks for a full literary review, school essay, or exhaustive book report
- the task requires quoting long copyrighted passages
Inputs
- Required: one or more book note drafts, generated catalog records, or rendered page excerpts.
- Optional: private source notes, import summaries, desired audience, catalog style, and target output format.
Public Note Contract
Each public book note should contain separate layers:
- Book summary: what the book is about, in plain language, without explaining why it was imported.
- Recommendation lens: why this reader or curator thinks the book matters.
- Reader fit: who should read it and what problem, taste, or context it serves.
- Personal-note synthesis: what the source notes or highlights suggest mattered, paraphrased safely.
- Next action: one concrete way to read, revisit, compare, or apply the book.
- Copyable markdown: includes the same canonical sections and is not a separate stale summary.
Quality Rubric
Score each note pass, revise, or block.
Pass only if:
- the summary is specific enough to distinguish the book from nearby books
- the recommendation lens is not just "I highlighted this a lot"
- personal-note synthesis is paraphrased and public-safe
- the note avoids raw import language such as annotation counts, private draft markers, asset IDs, and local paths
- sections are consistent between the source markdown, generated catalog data, and copyable markdown
- the note says something a reader could act on
Revise if:
- the summary is accurate but generic
- the recommendation lens repeats the summary
- notes are useful but too abstract
- the copyable markdown is missing a section or has stale wording
Block if:
- raw private notes, long copyrighted excerpts, account details, local paths, or credentials appear
- the title/author is uncertain but presented as confirmed
- the public page describes import mechanics instead of the book
- the recommendation relies mainly on private counts or metadata
Review Workflow
Bulk Audit Rule
For large audits, preserve the main context by splitting the review across
subagents whenever subagent tooling is available.
Use subagents when any of these are true:
- auditing 10 or more books
- auditing more than roughly 10,000 words of note content
- comparing source markdown, generated catalog data, rendered pages, and
copyable markdown across many records
- the task is likely to require several independent passes for privacy, quality,
and consistency
Parent-agent responsibilities:
- define the audit rubric and pass/block criteria once
- assign bounded batches, normally 5-8 books per subagent
- tell each subagent to return only status, issues, suggested fixes, and file or
record identifiers
- keep raw private notes out of subagent prompts unless they are required for
the assigned review
- aggregate the subagent findings into one final decision list
- run deterministic validators or grep checks after any edits
Subagent responsibilities:
- inspect only the assigned batch
- score each note
pass, revise, or block
- flag public-safety risks, thin summaries, stale copyable markdown, and generic
recommendation text
- avoid rewriting the whole note unless explicitly assigned a rewrite batch
If subagent tooling is not available, process the audit in small batches and
report that the subagent split was unavailable.
- Identify the source layer:
- raw/private import
- reviewed public markdown
- generated JSON/catalog record
- rendered public page
- Check the section contract:
- Summary
- Why/recommendation lens
- Best for/reader fit
- Notes/personal synthesis
- Next action
- Copyable markdown
- Flag public-safety issues:
- import metadata
- raw quotes
- private notes
- local paths, credentials, email addresses, account IDs
- uncertain metadata presented as fact
- Evaluate substance:
- Does the summary explain the book's core argument, story, or premise?
- Does the recommendation explain the curator's lens?
- Do the notes add specific, non-generic reading value?
- Is the next action concrete?
- Return a compact audit and, when asked, a revised public-safe version.
Output Contract
For each note, return:
Status: pass | revise | block
Main issue: <one sentence>
Fixes:
- <specific edit>
- <specific edit>
Public-safety concerns:
- none | <concern>
When rewriting, return only public-ready text unless the user asks for an audit
table. Do not include raw source excerpts in the rewritten note.
Boundaries
- Keep raw highlights and private notes out of public output.
- Do not mention import counts, annotation counts, asset IDs, or draft mechanics in public recommendation copy.
- If private source material is used, synthesize patterns rather than quoting it.
- Keep this skill reusable: no hardcoded personal handles, private repo names, account IDs, local paths, or user-specific defaults.