sumeru-deconstruct

Web Fiction Deconstruction and Imitation Writing Analysis (deconstruct)

Trigger Keywords

Help me deconstruct a book, deconstruction draft, analyze why this book is popular, deconstruct this book's structure, learn this book's writing method, imitate this book, deconstruct the golden three chapters, analyze the pacing of payoffs, benchmark book analysis, deconstruct viral hits, how are this book's hooks designed, deconstruct character relationships, analyze foreshadowing recovery, deconstruct pacing, deconstruction report

Inapplicable Scenarios (Explicitly Excluded)

❌ Directly continue/rewrite the original work, generate fan fiction → redirect to sumeru-write(and original settings must be used)
❌ Change the original work's names and place names and pass it off as your own work → refuse, and explain the infringement risk
❌ Request to "write a book exactly like this one" → can only provide a structural template and differentiation plan
❌ Analyze pirated resources, provide original text downloads → do not participate

Core Principles

Deconstruct structure, not sentences; learn methods, not names.

What book deconstruction outputs is a "structural template + verifiable parameters + differentiation plan", not "reskinned text". Any continuous excerpt of the original text that enters the report shall not exceed 12 characters, and the original work's character names/place names/proper nouns must not appear in the report body.

Workflow

输入（单文件/章节目录/片段）
    ↓
[1] 机械统计（脚本，必跑）
    → analyze-book.py → .sumeru/deconstruct/structure.json + 拆书数据报告.md
    ↓
[2] 结构拆解（模型，基于数据）
    → 十二维分析：开篇 / 人物功能位 / 节奏爽点 / 结构伏笔 / 语言视角
    → 每条结论必须有数据或章节位置支撑
    ↓
[3] 抽象化改写
    → 角色替换为 [主角] [对手] [导师]，地名替换为 [初始势力] [高层地图]
    → 情节替换为「动作 + 结果 + 代价」的功能描述
    ↓
[4] 输出拆书报告
    → 拆书报告_《[作品代称]》.md
    ↓
[5] 原创性检查（脚本，必跑）
    → check-derivative.py
    → 未通过必须改写后重跑
    ↓
[6] 生成仿写方案（可选）
    → 保留结构比例，替换世界观/人物关系/冲突类型

Command

# 1. 结构统计
python3 skills/sumeru-deconstruct/scripts/analyze-book.py --file 参考书.txt
python3 skills/sumeru-deconstruct/scripts/analyze-book.py --chapters reference/chapters

# 2. 原创性检查（报告写完后必跑）
python3 skills/sumeru-deconstruct/scripts/check-derivative.py \
  --source 参考书.txt \
  --output 拆书报告_某书.md \
  --forbid "原作主角名,原作地名"

Output Content

User-visible output (current working directory)

拆书数据报告.md: mechanical statistics (chapter word count, dialogue ratio, hook distribution, point of view, repeated phrases)
拆书报告_《作品代称》.md: complete structural deconstruction + abstracted template + imitation writing plan

Intermediate data (.sumeru/deconstruct/）

structure.json: per-chapter statistical data
derivative-check.json: originality check results (passed must be true）
derivative-check.md: readable check report

Deconstruction Report Structure

One-sentence conclusion: what keeps readers (≤80 characters)
Structural skeleton: stage / chapter range / core goal / main conflict / change at stage end
Golden three chapters deconstruction: initial state → disruption → variable → small trial → hook
Payoff and pacing curve: supported by hook distribution, word count distribution, payoff intervals
Character function slots: mentor/companion/opponent/love interest/tool character, appearance and transformation nodes
Foreshadowing ledger: planting chapter / recovery chapter / span / recovery method
Reusable structural template: all use placeholders, no proper names from the original work appear
Imitation writing plan (differentiation): what to keep, what to replace, what is prohibited
Risks and points not recommended to imitate

For detailed dimensions and templates, see references/deconstruct-framework.md。

Data Interpretation Criteria

Metric	Low	Healthy range	High
Word count per chapter (Qidian)	<2500	3000-5000	>6000
Word count per chapter (Fanqie)	<1500	2000-3000	>4000
Dialogue ratio	<10%	15%-35%	>45%
Average paragraph length	<30 characters	50-150 characters	>300 characters
Chapter-end cliffhanger ratio	<30%	50%-80%	100% (fatigue)

Data is used to identify problems, not to score works; different genres have different healthy ranges.

Connection with downstream Skills

→ sumeru-outline: Only pass "structural template + pacing parameters + functional slot configuration", not the original work's proper nouns and specific plot points
→ sumeru-topic: Can serve as market validation reference for the same genre, but cannot be the sole source of topic-selection ideas
→ sumeru-write: Only pass "hook type ratio, dialogue ratio, paragraph and sentence length parameters"
→ sumeru-review: The book-breakdown conclusions can serve as a reference baseline for pacing review

Prohibited: Using the book-breakdown report directly as an outline; directly using the original work's character relationship chart after renaming; reusing the original work's chapter titles after changing one character.

Deliverables and Acceptance Criteria (Definition of Done)

Acceptance item	Standard	Verification method
Data complete	Chapter count, total word count, per-chapter word count, and hook distribution all complete	`structure.json` Fields complete
Conclusions well-founded	Every structural conclusion is supported by a chapter position or data	Spot-check the report
Thoroughly abstracted	The report body contains no character names, place names, or proper nouns from the original work	`check-derivative.py --forbid` No matches
No verbatim copying	Longest continuous overlap ≤12 characters	`check-derivative.py` Pass
Template executable	The imitation-writing template can guide writing a different story	Manual check of placeholders
Differentiation clear	At least 3 settings that must be replaced are provided	Check the report's "Imitation-writing plan"
Risk warning	Clearly point out the points not recommended for imitation	Check the report's "Risks"