SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsbook2skill

book2skill

General & other

View repository

Updated 10-15 23:44

About this skill

book2skill — a meta-skill that distills a book into a set of executable skills

Mission

Break down the methodology accumulated in a book into a set ofatomic skills that an agent can invoke in real scenarios, so that readers truly put them to use.

Boundaries:

  • ✅ Do: distillation of methodology / decision frameworks / checklists / principles / conceptual systems
  • ❌ Don't do: book excerpts / reading reflections / author persona role-play (for the latter, use nuwa-skill)

Core methodology: RIA-TV++

A pipeline of four stages + parallel extraction + triple validation + darwin compatibility testing. See methodology/00-overview.md。

阶段 0: Adler 整书理解     → BOOK_OVERVIEW.md
阶段 1: 5 个 agent 并行提取 → 候选方法论单元池
阶段 1.5: 三重验证筛选       → 通过的单元
阶段 2: RIA++ 构造 skill     → 每个 skill 的 SKILL.md
阶段 3: Zettelkasten 链接    → INDEX.md
阶段 4: 压力测试 (darwin 兼容) → test-prompts.json + 回炉淘汰

When to invoke this skill

The user says something like:

  • "Help me break down 《Poor Charlie's Almanack》"
  • "Distill the Selected Works of Mao into a skill"
  • "distill this book into skills: "
  • "I want to turn this book's methodology into usable skills"

Input requirements

Before startingyou mustconfirm with the user:

  1. The book's text source: PDF / EPUB / TXT file path, or accessible plain text.Do not"break down a book from memory" without the text — rather stop and ask the user for it.
  2. Book title + author + publication year: used for directory naming and auditing.
  3. Whether it is a first pilot: if the user is using book2skill for the first time, it is recommended to first break down 1 book to validate the process before doing it in bulk.

Output structure

books/<book-slug>/
├── BOOK_OVERVIEW.md           # 阶段 0 产出: 主旨/骨架/术语/批判
├── INDEX.md                   # 阶段 3 产出: skill 总览 + 引用图
├── candidates/                # 阶段 1 产出: 原始候选池 (审计用)
├── rejected/                  # 阶段 1.5 淘汰的单元 + 原因 (审计用)
├── <skill-slug-1>/
│   ├── SKILL.md
│   └── test-prompts.json      # darwin-skill 兼容格式
├── <skill-slug-2>/
│   └── ...

Execution flow (strictly in order)

Stage 0 — Whole-book understanding

  1. Read the book text provided by the user. Read large files in chunks.
  2. Execute methodology/01-stage0-adler.md Adler's four steps (structure / interpretation / criticism / application).
  3. Fill in according to templates/BOOK_OVERVIEW.md.template , write to books/<slug>/BOOK_OVERVIEW.md。
  4. Show the output to the user for confirmation: "Did I understand the skeleton correctly? Is there any direction you want emphasized?" Once confirmed, proceed to Stage 1.

Stage 1 — 5 sub-agents extract in parallel

In parallel spawn 5 Task sub-agents (using the Agent tool, launching 5 in a single call):

sub-agent Prompt read Output
Framework extractor extractors/framework-extractor.md Decision frameworks / mental models
Principle extractor extractors/principle-extractor.md Principles / checklists / rules
Case extractor extractors/case-extractor.md Instances the author personally used in the book
Counterexample extractor extractors/counter-example-extractor.md Failure modes warned about in the book
Term extractor extractors/glossary-extractor.md Dictionary of key concepts

Each sub-agent reads the book independently, extracts independently, and outputs independently to books/<slug>/candidates/<type>.md。

Stage 1.5 — Triple validation screening

Read methodology/03-stage1.5-triple-verify.md, and for each candidate unit execute:

  • V1 Cross-domain: Is there corroboration from at least 2 independent passages in the book?
  • V2 Predictive power: Can it be used to answer a new question not explicitly stated in the book?
  • V3 Uniqueness: Is it not common sense that any smart person would say?

Those that pass enter stage 2. Those that do not pass are written to books/<slug>/rejected/ with reasons attached — preserving an audit trail, and also allowing users to retrieve them later.

Stage 2 — RIA++ construct skill

For each unit that passes, fill in according to templates/SKILL.md.template :

  • R (Reading): original text quote ≤150 characters/paragraph
  • I (Interpretation): rewrite the methodology skeleton in your own words (avoid copying the translation)
  • A1 (Past Application): cases the author used in the book
  • A2 (Future Trigger) ★: in what situations the user would need this → the skill's description field
  • E (Execution): 1-2-3 executable steps
  • B (Boundary): when it does not apply / the author's blind spots from the stage 0 critique stage

For detailed rules see methodology/04-stage2-ria-plus.md。

Stage 3 — Zettelkasten links

According to methodology/05-stage3-zettelkasten.md:

  1. Identify reference relationships between skills (A depends on B / A contrasts with B / A combines with B)
  2. Add a "related skills" section at the end of each SKILL.md
  3. According to templates/INDEX.md.template generate INDEX.md (including a reference graph mermaid)

Stage 4 — Stress testing (darwin compatible)

For each skill, according to methodology/06-stage4-pressure-test.md:

  1. Design 5–10 test prompts, according to templates/test-prompts.json.template write to test-prompts.json
  2. Include at least 3 categories: should invoke / should not invoke (decoy) / boundary ambiguous
  3. Run it locally once,those that fail go back to stage 2 for rework — do not do "surface patching"
  4. After all pass, notify the user: "Completed, can be fed to darwin-skill with one click for automatic evolution"

Quality red lines (violation blocks output)

  1. Each skill must passalltriple verification
  2. Each skill must have the complete six sections R / I / A1 / A2 / E / B
  3. Original text quote ≤150 characters/paragraph
  4. Each skill must have test-prompts.json, and include decoy tests (scenarios that should not invoke it)
  5. description The field must clearly define the trigger conditions, not just "a skill about X"

Ecological positioning with nuwa-skill / darwin-skill

  • nuwa-skill: distills people (ways of thinking / expression DNA)
  • book2skill (this skill): distills books (methodologies / frameworks / principles)
  • darwin-skill: evolves any skill

The three interlock: the output of this skill test-prompts.json strictly follows the darwin-skill format, so that the produced skills can be directly connected to darwin for automatic evolution.

Invocation conventions

  • Always pilot with 1 book first — unless the user explicitly says "batch"
  • Proactively report progress between stages — do not run silently to completion and then dump the results
  • Do not deconstruct books from memory — if there is no text, stop and ask
  • Preserve the audit trail — keep both candidates/ and rejected/