narrative-book-summaries
Use when summarizing narrative/argument books into vaults.
- License 许可
- MIT
- 最近更新
- 2026-11-03
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- narrative-book-summaries
- author
- curator
- license
- MIT
- version
- 1.0.0
- metadata
- { "hermes": { "tags": [ "books", "summarization", "obsidian", "mermaid", "markdown-style" ], "related_skills": [ "oralita-book-sum-obs", "obsidian", "research-paper-summaries" ] } }
- platforms
- [ "windows", "linux", "macos" ]
- description
- Use when summarizing narrative/argument books into vaults.
02 资源文件 2
- 参考文档references/narrative-note-anatomy.md4.1 KB
- 脚本scripts/check_chapter_notes.py4.6 KB
03 SKILL.md 正文
Narrative Book → Obsidian Study Vault
When to Use
Argument-driven books where each chapter is a self-contained essay with a thesis: big history, popular science, philosophy, case-study narratives. For technical/practice books, use oralita-book-sum-obs (the execution engine: extraction, manifest, delegation, verification) with its study templates. That skill is user-protected; this companion skill carries the narrative-book FORMATS the user has approved and the style rules that apply to everything generated. For research papers (single-file study notes with quote and number verification), use research-paper-summaries.
Signals: 10–20+ titled chapters with distinct arguments; the user asks to "plan the format first" because the book is "different from other books"; a prior narrative-book vault exists to mirror (e.g. psychology_book/schema_therapy).
Approved Note Anatomy (user-approved 2026-09-03, Sapiens run)
One file per chapter, NN_Title_With_Underscores.md, plus Book Title - Overview.md:
- YAML frontmatter:
tags: [booktag, part-tag, 1-3 topic tags, history]— notitle:field. - Source blockquote:
> *Source: <Title> by <Author> (<publisher, year>), Chapter N "<exact title>", PDF pp. A–B (printed pp. unknown)* - H1:
# NN: <Exact Chapter Title>(colon, never em-dash) ## Key Idea— the chapter's thesis + why it matters, 2–3 paragraphs (~250–350 words), mentor voice.## The Argument— bold-led bullets reconstructing the evidence spine in own words.## Key Concepts—### Nameper durable concept: define + example + why it matters. The book's terminology.## Memorable Quotes— 2–4 short verbatim quotes, each> "..." (PDF p. N).## Connections—- [[Stem]]: one-line reasonwikilinks to sibling notes + the overview last. Colon separator, never em-dash.## Takeaway— one line worth remembering.
Full spec, per-section word budgets, and the Sapiens worked example: references/narrative-note-anatomy.md.
Style Rules (embed from the start — user preferences)
- No em-dashes in generated prose. Use commas, colons, parentheses, or split sentences. En-dashes stay in page ranges (
PDF pp. 84–104). - Verbatim quotes are sacred. Never alter quoted punctuation, including short quoted fragments inside prose (skip them in any bulk punctuation edit).
- Mermaid node labels with special chars must be quoted (
A["text (parens)"]); edge labels with special chars quoted as-->|"label ()/"|. - Match the destination vault's naming/link conventions over any global rule — underscore stems are fine where the local reference vault uses them.
- Overview/MOC: chapter-map table grouped by part with PDF-page sources, one Mermaid argument map, reading paths (goal → start notes), 5–9 core ideas, honest footer with extraction limitations.
Workflow
- Plan first when asked. Inspect the PDF (pymupdf
get_toc(), page count, text-layer vs OCR check), read 1–3 reference-vault notes, then present the concrete plan: per-chapter anatomy, citation scheme, file list. Confirm granularity and quote sections with the user before writing anything. - Extract with page markers. Insert a
[PDF p. N]marker at every PDF-page boundary inside each chapter extract. Writers then attribute quotes from markers, and validation can check attribution mechanically. Verify chapter boundaries by reading head/tail text of each extract, not the TOC alone. - Delegate in live-limited batches. Query
delegation.max_concurrent_childrenat run time. Every writer gets: extract path, staging path, format spec, an exemplar note, and the complete canonical link allow-list. A child's report is NOT proof — validate every staging file (frontmatter, source line, section order, quotes verbatim + page attribution, links) before promotion. Usescripts/check_chapter_notes.py. - Promote, verify, spot-check. Promote validated files, run the vault verifier, then spot-check 5+ factual claims per representative note against the source text (grep with normalized whitespace — extracts are line-wrapped, so phrase search must flatten whitespace first; PDF text layers also hyphenate across lines ("develop-\nment"), so also try the "- "-joined form before concluding a quote is absent).
- Compression pass (optional). The user may ask for shorter notes after the first draft. Rules in
references/narrative-note-anatomy.md.
Mermaid Diagram Rubric
Add a diagram only when the chapter's structure IS a relationship:
- Strong candidates: cycles/feedback loops (luxury trap, capitalist circle), abstraction ladders (money), taxonomies (religion tree), branching futures (three paths), vicious circles as
stateDiagram-v2. - Good candidates: ontologies (objective/subjective/inter-subjective), imperial cycles, self-defeating prophecy (
stateDiagram-v2), science↔politics↔economics loops, three-way alliances. - Skip: plain lists, debates without a spine, psychological findings. Never force a sequence diagram; use
stateDiagram-v2where the structure is genuinely stateful. - Place each diagram inside the Key Concept it illustrates.
- After editing, verify: code fences balanced, square brackets matched, vault verifier passes.
Pitfalls
- Fuzzy patch can over-match. A long
old_stringthat partially matches may swallow a preceding sentence (observed: a patch deleted the opening sentence of a concept paragraph). Check every diff and restore immediately; prefer short unique anchors (heading lines) over long paragraph anchors. - patch escape-drift. Do not backslash-escape double quotes in patch strings; plain quotes match plain quotes in the file. If a match fails with "Did you mean", re-grep the exact glyphs (apostrophes, dashes) before retrying.
- Calibre/PDFDrive conversions have no folio numbers. Scan for digit-only lines to detect printed page numbers; if absent, cite PDF pages and mark
printed pp. unknowneverywhere. Never invent printed ranges. - Endnote texts are usually absent from e-book extracts — never assert footnote content.
- Homo Deus (Sapiens sequel) is queued in
F:/books/human_book/— reuse this anatomy and the Sapiens vault as the format reference.
Reference Map
references/narrative-note-anatomy.md— full per-section spec, compression-pass rules, Sapiens case study incl. the 14-diagram map.scripts/check_chapter_notes.py— validates staging/vault notes: frontmatter, source line, H1, section order, verbatim quotes + page attribution, wikilink allow-list.
04 同仓库更多 Skills
project-spec-docs
Fill/update project_spec docs (templates, MM handoffs).
penpot-mcp
Penpot MCP: connect, verify, and fix streaming via nginx.
nginx-streaming-proxy
Configure nginx for streaming endpoints (SSE, MCP, WS).
branding-ci
Use for brand/CI identity, style guides, and brand docs.
yuanbao
Yuanbao (元宝) groups: @mention users, query info/members.
slidev
Create and present web-based slidedecks for developers using Slidev with Markdown, Vue components, code highlighting, animations, and interactive features. Use when building technical presentations, conference talks, code walkthroughs, teaching materials, or developer decks.