article2book

Existing Content Asset Reorganization Skill

Last updated：2026-05-17

Positioning

article2book Used to organize existing content assets in a directory into structured deliverables that can be published, delivered, or sustainably maintained.

This skill still uses "book planning" as the default main path, but it does not force all materials toward a book. It first determines whether this batch of materials is best suited to become:

• Book manuscript
• Booklet
• Course
• Article series
• Practical manual
• Knowledge base
• Material pool not currently recommended for productization

"Existing content assets" here include but are not limited to:

• Published articles, column drafts, WeChat official account drafts
• Course lecture notes, sharing outlines, livestream transcripts, speech drafts
• Video tutorial companion scripts, subtitle scripts, verbatim transcripts
• Interview transcripts, Q&A records, meeting minutes
• Courseware, case materials, practical checklists
• PDF text, Word documents, Obsidian notes
• Cards, methodology drafts, series of essays

Its starting point is not "writing a book from a blank page," but rather:

• First take stock of existing content assets
• Then determine the most suitable delivery format
• Then filter out low-value content and narrow down the core theme, readers, and structure
• Finally reorganize existing materials into a content product more suitable for publication or delivery

The default methodology of this skill is:

• Agent full-read first
• Script index assistance
• First determine the content format, then determine whether it should become a book
• First filter out low-value content, then do structural design
• Default single-file delivery
• After confirmation, proceed to the corresponding product draft

Typical input

• Markdown articles, column drafts, WeChat official account drafts, course lecture notes, subtitle scripts, verbatim transcripts, and Obsidian notes mixed in one directory
• A batch of livestream transcripts, interview transcripts, course transcripts, or meeting minutes, and the user is unsure whether they are suitable for a book, course, or knowledge base
• Many long articles have already been written, but the themes are scattered, there is a lot of repetition, and the expression angles are inconsistent
• The same author continues writing and publishing around a certain methodology, but it has not yet been consolidated into a stable content product
• A mixed directory contains articles, courseware, cases, manual drafts, and scattered notes at the same time

Example：

• 请读取 <素材目录>，判断这些文章能不能整理成一本书，并给出书稿结构方案。
• 我有一批直播逐字稿和配套文章，帮我判断适合做书、课程还是小册子。
• 这个混合目录能不能整理成一个知识产品？如果不适合成书，也请给出替代方案。
• 你先只输出一份策划意见，我确认后你再直接生成对应初稿。

Not applicable scenarios

• The user only wants to summarize a file directory and does not care about content format, reader positioning, or reorganization path
• The user already has a mature directory and only needs to continue writing a certain chapter
• The user wants to conceive a book entirely from scratch rather than organize it based on existing materials
• The folder mainly contains images, scans, recordings, pure video, and other materials that have not yet been converted into readable text
• What the user wants to do is layout, cover design, marketing copy, or distribution process, rather than content asset reorganization

Default output directory

By default, create theunder the source directorycreate the 书稿策划输出/ directory. Even if the final recommendation is not a book, continue using this directory to avoid creating multiple sets of path rules for different formats.

Working modes

Choose a working mode based on the user's request and the complexity of the material:

Mode	Applicable scenario	Default output
Quick assessment	The user only wants to know the direction first, or the material is so large that it is necessary to first judge whether it is worth going deeper	Give a brief judgment in the conversation, and if necessary generate 书稿策划意见.md a simplified version
Standard planning	Default mode; the user asks to judge whether it can become a book, what form it suits, and how to reorganize it	书稿策划意见.md
Deep restructuring	After the user confirms the planning opinion, or explicitly requests direct generation of a complete first draft	全书初稿.md or the corresponding form's first draft

UseStandard planningby default. Enter deep restructuring only after the user confirms the direction, explicitly requests generation of a first draft, or the material is complex enough that it must be split into batches.

Delivery form decision tree

First determine the best content form, then determine whether to enter the book path:

1. Book: The material has a stable core theme, a clear audience, sufficient depth, and sustainable room for expansion.
2. Booklet: The theme is focused but the depth is insufficient, making it more suitable for a short, quick delivery.
3. Course: The material is mainly based on teaching sequence, operational demonstrations, practice tasks, or learning paths.
4. Article series: The viewpoint has value but the theme has not yet formed a complete system.
5. Practical manual: The material is mainly based on processes, checklists, standards, cases, and operating conventions.
6. Knowledge base: Multiple themes coexist in the material, making it suitable for ongoing maintenance and retrieval rather than linear reading.
7. Productization not recommended for now: The material is too scattered, repetitive, outdated, or lacks sufficient original judgment.

See the criteria for judging each form in references/content-productization-models.md。

Default delivery

This skill usestwo stagesfor delivery by default, and minimizes the number of files the user needs to read.

Stage one: First produce a planning opinion

By default, output only:

File	Purpose
书稿策划意见.md	The only default deliverable. It centrally provides the best content form, whether it is suitable for a book, the main proposition, target readers, draft structure, deletion and modification actions, transformation path, and next-step outputs

Only in the following cases should supporting files be output additionally:

• The material is very large and evidence of full reading needs to be retained
• The user explicitly requests a detailed table
• The themes diverge obviously and the basis for screening needs to be shown
• The input format is complex and the preprocessing status needs to be recorded

Optional supporting files include:

File	Purpose
processed/agent_reading_notes.md	Internal reading notes after the Agent reads through in batches
processed/article_inventory.csv	The basic checklist index established by the script
processed/article_inventory.jsonl	Structured index
附录-排除与降权清单.md	Output only when the basis for screening needs to be explained

Stage two: After confirmation, generate the first draft in the corresponding form

If the user confirms that the Phase 1 opinion is correct, proceed directly to content restructuring, without first producing a large batch of intermediate files.

The default output depends on the Phase 1 judgment:

Recommended form	Default first draft
Book	全书初稿.md
Booklet	小册子初稿.md
Course	课程大纲与讲稿初稿.md
Article series	系列文章规划与首批初稿.md
Practical manual	实务手册初稿.md
Knowledge base	知识库结构与核心条目.md

Only when the content is too long, requires chapter-by-chapter parallel drafting, or the user explicitly requests splitting, output:

File	Purpose
chapters/01-*.md ...	Chapter-by-chapter or unit-by-unit first draft
全书结构说明.md / 内容结构说明.md	Record chapter boundaries, unified terminology, and merge rules

See the complete template at references/output-template.md。

Workflow

Step 1: Confirm the source directory, file types, work mode, and output location

• Confirm whether the user has provided a single-topic directory or a directory with multiple topics mixed together
• By default, prioritize already text-converted materials, such as .md、.markdown、.mdx、.txt、.srt、.vtt
• For .docx、.pdf, first include them in the list as "materials requiring preprocessing"; enter full reading only when text can be extracted directly
• For non-text materials such as course videos, recordings, and podcasts, first transcribe them into verbatim transcripts before including them in the judgment
• First exclude images, attachments, and image directories to avoid mistakenly identifying illustrations as body content
• Clarify whether to use quick inventory, standard planning, or deep restructuring
• Clarify whether the output goes in the source directory 书稿策划输出/

For how to handle different sources, see references/source-type-handling.md。

Step 2: First have the Agent read through all candidate text materials

This is the main workflow of this skill and cannot be skipped.

• For text materials that can be read directly, the Agent should read through as much as possible, rather than only looking at titles, previews, or script statistics
• If the directory is large, read through it in batches, for example 10-20 materials per batch; after each batch, immediately record notes
• For each material, record at least:
  • One-sentence summary
  • What question it mainly answers
  • Whether it is more suitable as a main chapter, case, appendix, course unit, manual entry, or should be removed
  • Whether there are issues such as strong time sensitivity, repeated expression, insufficient depth, overly colloquial style, or format preprocessing

It is recommended to save the read-through notes to:

• processed/agent_reading_notes.md

However, this file is by default regarded as an internal working draft and should not automatically serve as the user's first reading entry point.

For a more detailed read-through method, see references/agent-reading-protocol.md。

Step 3: After the read-through, first perform "keep / downgrade / exclude" screening

Before starting topic clustering, first screen out low-value content; do not force all materials into the final product.

Prioritize excluding or downgrading the following content:

• Old and shallow: Earlier in time, and limited to information listing, tool introductions, and surface-level opinions, without forming a judgment framework that still has value today
• Outdated and invalid: Highly dependent on a certain product update, a certain version feature, or a certain hot event, and difficult to rewrite into evergreen expression
• Repeated but weaker: Expresses the same judgment as another material, but is clearly weaker in argumentation, cases, structure, and expression
• Only excitement, no motif: Looks eye-catching, but has a very weak connection to the main line
• Content density too low: The length is not short, but the information gain is very small, making it hard to support a chapter, unit, or entry

When filtering, note:

• Old content does not equal low quality
• Short content does not equal low quality
• Exclusions must be justified

For filtering criteria, see references/content-screening-rubric.md。

Step 4: Build a material inventory on the basis of filtering; scripts only assist

If you need to fill in paths, check for gaps, or generate a basic index, you can run:

python3 scripts/build_article_inventory.py "<源目录>" --output-dir "<输出目录>/processed"

But note:

• This script is only used to fill in paths, check for gaps, create a basic index, and mark materials that need preprocessing
• You must not make topic judgments directly based solely on script output
• .docx、.pdf is only marked in the script as needing preprocessing, and the body text is not parsed directly
• If the Agent's read-through record conflicts with script fields, the read-through understanding takes precedence, and then manual correction is made

Step 5: Extract themes and content forms, rather than looking only at titles

After reading the inventory, at minimum cover:

• High-density representative materials
• Materials that appear similar but have different arguments
• Materials that may belong to marginal themes
• Materials whose titles look like tool reviews but whose body text carries methodology
• Transcripts that are very colloquial but already contain the prototype of chapters, course units, or handbook entries
• Notes or meeting minutes that appear scattered but can be consolidated into knowledge base entries

Do not cluster based only on titles. The recurring "problem awareness, judgment criteria, methodology vocabulary, operational steps, case structure, narrative tension" in the body text usually determines the final form more than the title does.

Step 6: Identify the "mother theme" and the "most suitable content product"

First classify the materials by theme clusters, then judge whether they can be consolidated into a stable content product.

Focus on identifying:

• Which core question the author repeatedly answers
• What perspective truly differentiates the author compared with similar writing
• Whether the materials are better suited for linear reading, systematic learning, quick reference, or ongoing maintenance
• Whether these contents are more like "different chapters of the same book" or "seeds of multiple content products"

If there are two or more sufficiently strong main lines in the table of contents at the same time, do not force them into one book. You should clearly propose:

• Option A: Focus on one book
• Option B: Split into two books / one book + one course / handbook + knowledge base
• Option C: Do not make a book for now; first consolidate into a series of articles or a material library

Step 7: Assess book feasibility and give alternative form suggestions

Judge item by item according to references/book-viability-rubric.md :

• Theme concentration
• Clarity of the core proposition
• Clarity of the target readers
• Content depth and complementarity
• Perspective differentiation
• Timeliness risk
• Space for sustainable expansion

When outputting, you must also clearly specify:

1. Best content form: book / booklet / course / series of articles / practical handbook / knowledge base / productization not recommended for now
2. Book conclusion: can directly proceed to manuscript planning / can become a book but needs consolidation and rewriting / book not recommended for now
3. Alternative path: if it does not become a book, what product it is most suitable to turn into, and why

Step 8: Consolidate into a 书稿策划意见.md

By default, do not split the analysis process into many user-visible files.

The following content should be written together into one 书稿策划意见.md：

• Best content form
• Whether it is worth turning into a book
• Forms not recommended
• Recommended main proposition
• Target readers and differentiation
• Recommended structure draft
• What content to keep, what to compress, and what to delete
• Core principles for keeping / merging / excluding
• Conversion path and next-step deliverables
• If the user confirms moving forward, how Phase 2 will directly generate the corresponding first draft

Only when the user explicitly requests it, or when the material is especially large in scale and especially contentious, should the screening checklist, master table, and reading notes be additionally output explicitly.

Step 9: If the user confirms, proceed directly to Phase 2 to generate the first draft

If the user approves of 书稿策划意见.md , then proceed directly without lingering on more planning documents.

Default approach:

• First establish the skeleton of chapters, units, or entries according to the recommended structure in the feedback
• Use the existing material as the draft base for merging, deleting, revising, and supplementing
• By default, output a single-file first draft corresponding to the recommended form

When there is a lot of material, you can:

• Use subagents to read through the material in batches
• Use subagents to draft the first draft by chapter, unit, or entry
• But the main Agent must be responsible for:
  • Unifying the main proposition and target readers
  • Unifying terminology and expression standards
  • Removing repetitive arguments
  • Doing the final structural tightening and merging of the full draft

Step 10: Fall back to subdivided files only when necessary

Only in the following situations is it recommended to fall back to a multi-file package:

• The user explicitly requests item-by-item verification of where the material goes
• The screening basis needs to be shown to a third party
• The topic branches severely and multiple content form options need to be compared
• Phase 2 has already started and a more detailed editing ledger needs to be retained

Regardless of whether the output is one file or multiple files, the following judgments should still be completed internally:

• 直接保留
• 局部吸收
• 合并重写
• 拆分改写
• 仅作案例/附录/练习/条目
• 移出本产物

Judgment principles

1. Content form takes priority over the urge to make a book: Not all material should be written into a book; first determine the form best suited to the user's goals and the state of the material.
2. Core theme takes priority over quantity of material: Having a lot of material does not mean it can become a book; the key is whether it continues to advance around the same question.
3. Distinctiveness takes priority over coverage: The most important thing for a book or course is a unique judgment framework, not covering every topic once.
4. Reader questions take priority over the author's timeline: The structure should be organized according to the reader's order of understanding, not arranged according to the order in which the material was produced.
5. Evergreen content takes priority over topical heat: When encountering strongly time-sensitive content, evaluate whether it can be rewritten into a long-lasting case or argument.
6. Evidence can be traced back: Topic judgments, structural mappings, and deletion/change suggestions should all refer back to the original file paths as much as possible.
7. Allow the conclusion that it is "not suitable to become a book": If the content naturally branches, do not force integration.
8. Full-text reading takes priority over script conclusions: The final judgment must be based on the premise that the Agent has covered all readable candidate text material.
9. Subtract first, then structure: After reading through, first remove low-value content, then do the structural design, to avoid the final product being dragged down by low-quality old content.
10. Reduce the user's reading burden by default: If one main opinion document can explain it clearly, do not split it into 6-8 user files.

Output quality requirements

• Conclusions must be clear; do not give only vague descriptions
• Must simultaneously state the "best content form" and "whether it is suitable as a book"
• For each major judgment, attach the corresponding material or topic cluster basis whenever possible
• Distinguish between "content already exists" and "must be supplemented"
• Distinguish between "suitable for publication as a book" and "more suitable for a course, column, handbook, or knowledge base"
• Mark missing information 未提及 / 待补充 / 需作者确认

Collaboration tools

Tool	Purpose
Agent / file reading capability	Main process: read through all readable materials in batches and form reading notes
Subagent / parallel Agent	When the amount of material is large, read through in parallel and draft by chapter/unit, but the final judgment and unification of the overall draft must return to the main Agent
python3	Optional assistance: run scripts/build_article_inventory.py to build an index, supplement paths, check for omissions, and mark materials requiring preprocessing
Markdown output capability	In the first stage, generate a single file 书稿策划意见.md; in the second stage, generate a draft in the corresponding form

Reference files

File	Description
references/agent-reading-protocol.md	Recommended process for the Agent to read through materials, take notes in batches, and form preliminary judgments
references/content-screening-rubric.md	Screening criteria for judging "keep / downweight / exclude" after reading through
references/content-productization-models.md	Criteria for judging content forms such as books, booklets, courses, columns, handbooks, and knowledge bases
references/source-type-handling.md	Handling methods and preprocessing rules for materials from different sources
references/output-template.md	Default output template
references/book-viability-rubric.md	Dimensions for assessing book feasibility and suggestions for alternative forms
scripts/build_article_inventory.py	Material directory scanning and indexing script (optional assistance)

Dependencies

System dependencies

Dependency	Installation method
python3	macOS: built into the system or brew install pythonLinux: sudo apt-get install python3

Python packages

Package name	Purpose	Installation command
No additional third-party dependencies	The script uses only the Python standard library; the script is optional assistance	No installation required

Notes

1. If the directory contains a large number of images, image directories, or automatically generated materials, exclude them before analysis.
2. If the article spans two strong themes, such as "legal AI methodology" and "general AI philosophical essays," priority should be given to splitting the book or splitting it into a book + course / knowledge base.
3. If the article title looks like news or a review, but the main text carries a stable methodology, its destination should be judged by methodological value rather than title style.
4. If the user already has a clear publishing direction, this skill should serve convergence and integration; if the user does not yet have a direction, this skill should first make a form judgment rather than rushing to create a table of contents.
5. By default, do not deliver too many intermediate files to the user. If one 书稿策划意见.md can explain it clearly, do not split it into 6-8 results.
6. In the second stage, by default, a draft of the corresponding form is generated directly, and users are no longer asked to first read a large number of transitional documents.