Chinese Web Novel Writing Skill

Positioning

This is a writing skill for Chinese web novels.

It addresses not 'Is the literary expression sophisticated enough?', but these more practical questions:

Is this idea worth turning into a web novel?
What is this book actually selling?
What will sustain the first 10-20 chapters?
How does the opening convert readers?
How does the middle section not lose focus?
How does the chapter end retain momentum?
How to retrieve similar examples and borrow structures?
How to avoid AI flavor throughout the entire process?

Default Input

The default input is:

A brief description of the target novel

Default behavior:

First, extract the theme, selling points, conflict, story engine, and target length from the synopsis.
Then, determine if the current task belongs to pre-planning, opening, chapter outline, single chapter, local repair, or final manuscript review.
If sufficient information is available, directly proceed to the most needed layer; if not, ask for only the minimal clarification.

Output Principle

By default, only output the most needed layer for the current stage; do not dump everything at once.

Common output layers:

Theme diagnosis
One-sentence hook
Premise
Story engine explanation
Volume 1 plan
Chapter outline for the first 10-20 chapters
Rewrite the opening
Single chapter body text
Chapter ending pressure
Pacing check-up
Imitation reference list

Overall Workflow

First, determine if the problem is in pre-planning, structure, chapter execution, or final manuscript review.
If the problem has clearly narrowed down to a specific writing stage, prioritize invoking the specialized module; do not just give general advice.
By default, first match module examples or local corpora, then summarize structures, then write new content.
'Remove AI flavor' is not a final cosmetic polish; it should be constrained from ideation, scenes, dialogue, and chapter endings.
After each chapter is written, it must by default pass the review module; no direct submission.

Module Responsibility Quick Reference

concept_planning Function: Compress a synopsis into theme, consumer appeal, hook, premise, story engine, length assessment, and direction for Volume 1. Stage: Pre-planning layer.
opening Function: Land the selling points, abnormal situation, and protagonist introduction into the opening conversion zone. Stage: Opening execution layer.
volume_outline Function: Expand the story engine into the golden three chapters, volume divisions, and the first 10-20 chapter outline. Stage: Mid-to-long term structure layer.
plot_logic Function: Fix the cause-and-effect chain of motivation, trigger, decision, consequence, and payoff. Stage: Basic structure module within the body execution layer.
character_consistency Function: Fix five types of character continuity: goals, emotions, relationships, physical state, and voice. Stage: Character state module within the body execution layer.
transition Function: Handle time jumps, space switches, emotional carry-over, perspective shifts, and linking the chapter end to the next chapter. Stage: Scene bridge module within the body execution layer.
dialogue Function: Handle relationship pressure, character voice, information embedding, and sharp dialogue edges. Stage: Dialogue module within the body execution layer.
chapter_ending Function: Handle chapter end pull, lingering effect, backward hook, and connection to the next chapter. Stage: Chapter closure module within the body execution layer.
anti_ai_voice Function: Clean up vague summaries, formulaic atmosphere, instruction-manual-like dialogue, and uniform tone. Stage: Style constraint module within the body execution layer.
consistency_review Function: Conduct a unified review of six types of consistency, preventing problematic chapters from being submitted directly. Stage: Final manuscript check layer.

Module Relationships

Understand modules as three link chains, not ten flat switches:

Pre-planning chain: concept_planning -> opening / volume_outline
Body execution chain: plot_logic + character_consistency + transition + dialogue + chapter_ending + anti_ai_voice
Final manuscript check chain: consistency_review

If a problem spans multiple modules simultaneously, the priority is usually:

plot_logic / character_consistency
transition / dialogue / chapter_ending
anti_ai_voice

Do not mistakenly fix a structural issue as a style issue.

Pre-planning

When these issues are involved, by default invoke the concept_planning module:

Can this idea be written as a web novel?
Are the theme and consumer appeal clear?
Are the selling point and abnormal situation prominent?
Does the hook / premise hold up?
Can the story engine sustain the first 10-20 chapters?
Is it better suited for short story, medium-length, full-length, or serialization?
Is the minimum goal for Volume 1 and the change at the end of the volume clear?

Hard requirements for this step:

First, determine if it's worth writing; then write the body text.
First, provide selling points, conflict, and story engine; then provide the larger setting.
When the synopsis cannot be compressed into a minimal skeleton, do not rush to start writing.

Opening

When these issues are involved, by default invoke the opening module:

The opening lacks grip.
The selling point is revealed too slowly.
The abnormal situation is not established enough.
The golden three chapters start slowly.

Default requirements:

Within the first 300-500 words, provide a gripping point.
Within the first 1500-3000 words, show the selling point.
Within the first 6000 words, let the reader know who the protagonist is, what they encounter, what the selling point is, and what they will want to read next.

Golden Three Chapters, Volume Division, and Chapter Outlines

When these issues are involved, by default invoke the volume_outline module:

The golden three chapters lack momentum.
Volume 1's progression is unclear.
The first 10-20 chapter outline is scattered.
The middle section easily drifts.
The end of the volume lacks a payoff point.

Long web novels must by default have chapter outlines.

Without chapter outlines, these problems are most likely:

Random side plots growing out of control.
Setting bloat.
Weak chapter ends.
The reading incentive chain breaks.

Chapter-by-Chapter Writing

Each Chapter Must Answer Four Questions

What does the protagonist want to do in this chapter?
Who or what stops them?
What does the situation become after this chapter ends?
Why should the reader continue reading?

Minimum Scene Structure

Each scene must have at least:

Goal
Obstacle
Change

Scenes without change should be deleted, merged, or compressed by default.

Local Problem Default Routing

Time jumps, space switches, emotional carry-over, perspective shifts, linking chapter end to next chapter: Invoke transition
High-tension dialogue, character voice, relationship pressure, information embedding: Invoke dialogue
Chapter closure, lingering effect, backward hook, reading incentive pull, next chapter connection: Invoke chapter_ending
Broken cause-and-effect, insufficient motivation, invalid trigger, unsatisfied consequence: Invoke plot_logic
Broken character design, drifting goals, emotional discontinuity, relationship discontinuity, loss of physical and voice continuity: Invoke character_consistency
Vague summaries, formulaic atmosphere, dialogue like instruction manuals, all characters speaking in one tone: Invoke anti_ai_voice

Mandatory Review After Chapter Completion

After each chapter is written, do not deliver it directly; by default, the consistency_review module must be invoked first.

Six types of consistency must be checked for each chapter:

Plot logic consistency: Does each key event have a precondition? Does each key decision have a trigger? Does each key change have a consequence?
Character goal consistency: What exactly does the protagonist want to do in this chapter? Does it drift without reason during the process?
Emotion and relationship consistency: Does the emotion have a source? Does the relational temperature match the previous scene?
Physical and informational state consistency: Are injuries, fatigue, secrets, misunderstandings, and known information lost?
Scene and transition consistency: Will the reader get lost? Is the momentum from the previous scene carried into the next?
Chapter-end consistency: Does the chapter end on a change? Can the first beat of the next chapter pick it up?

If any two are unstable, do not deliver directly; first return to the corresponding specialized module to fix.

General Module Invocation Rules

If the problem has clearly narrowed down to a specific writing stage, do not just give general advice; by default, prioritize invoking the specialized module.

The general order is:

First, determine if this is a specialized problem, not purely a style problem.
First, look at references/modules/README.md, then enter the corresponding module directory.
First, read the module's own README.md, then follow the module's recommended order to read tutorial.md, runtime.md, and the example library.
First, use the module's hierarchy and classification to diagnose the fault point.
Select 2-4 positive examples and 1-2 negative examples from the module's example library.
First, make a local repair plan, then write the diagnosis, rewrite, or body text.

If a problem spans multiple modules simultaneously, first fix the more fundamental, harder layer, then fix the more surface-level issues.

Imitation and Retrieval

This skill requires by default that during writing, you frequently search for similar examples and then borrow structures to write new content.

Search is not only performed when the user says "please imitate." As long as the task involves these contents, prioritize searching the local corpus:

A certain theme
A certain opening
A certain type of dialogue
A certain chapter ending
A certain flow of relationships

If the problem has clearly narrowed down to a specific specialized stage, also prioritize invoking the specialized module, rather than just stopping at general retrieval.

How to Find Modules

First, look at references/modules/README.md.
Then, go directly to the module directory based on the problem keyword: concept_planning / opening / transition / dialogue / chapter_ending / plot_logic / character_consistency / consistency_review / volume_outline / anti_ai_voice
After entering the module, first read the module README.md, then follow the module's own recommended order to read tutorial.md, runtime.md, good_examples.md, bad_examples.md.

Current Local Corpora

Main files:

Retrieval Script

Use:

scripts/search_corpus_examples.py

Common commands:

python3 scripts/search_corpus_examples.py --list-tags
python3 scripts/search_corpus_examples.py --list-types
python3 scripts/search_corpus_examples.py --type 'Opening Hook' --tag 'Crisis Pressure' --limit 5
python3 scripts/search_corpus_examples.py --type 'High-Tension Dialogue' --tag 'Relationship Breakdown' --limit 5
python3 scripts/search_corpus_examples.py --keyword 'True and False Heiress' --limit 10

Imitation Process

First, determine what the user is writing.
Find 2-4 similar examples from the index.
Summarize the common structure of these examples.
Then write new content.

For specialized problems, change to:

First, determine if a specialized module should be invoked.
Determine the hierarchy and type within the specialized module.
Select material from the specialized module's example library group.
First, make a bridge or structural plan.
Then write the body text.

If a chapter has already been written, add one more step:

Invoke consistency_review, pass the six types of consistency check, then submit.

Removing AI Flavor

This is a hard requirement, but the main skill will not repeat the full tutorial.

When the task is concentrated on these issues, directly invoke the anti_ai_voice module:

Too many vague summaries
Too many atmosphere sentences and adjective clichés
Dialogue reads like an instruction manual
All characters speak in the same tone
Sentences are too uniform, too neat, too template-like

Hard rules that are still retained:

Use action, not summary.
Use dialogue, not explanation.
Be specific, not abstract.
Every character's speech must carry a sense of identity.
Do not interpret 'remove AI flavor' as intentionally being crude.

Boundaries

What Not to Do

Do not talk vaguely about "characters should be three-dimensional" or "the story must have soul."
Do not only give worldbuilding without solutions for the first few chapters.
Do not only give character designs without conflict and events.
Do not default to a slow start.
Do not mistake imitation for plagiarism.
Do not let all dialogue sound the same.
Do not output verbose paragraphs that read like lecture notes.

What is Encouraged by Default

Give many specific examples.
Make many positive-negative comparisons.
Provide many actionable writing methods.
Frequently check local examples.
Use structural references to replace vague advice.

Modules and Search

Module main entry point: references/modules/README.md
Module template: references/modules/module_template.md
Corpus search general guide: references/webnovel_corpus_guide.md
Search script: scripts/search_corpus_examples.py

Aoisop/mawoagent

What is mawoagent?

Ask in your favorite AI

Documentation