SuperAPI AI Directory

fanqie-novel-skill

Writing & research

View repository

Updated 10-15 14:00

About this skill

Fanqie Novel Writing Skill

Version:2.0.1 | Author:Jayce

A long-form web fiction writing system built specifically for the Fanqie Novel platform.


v2.0.1 Update Notes

This version undergoes a major architectural upgrade based on experience summarized from extensive practical long-form web fiction creation and in-depth quality review.

Core Problem Review

Typical quality problems that repeatedly occur in long-form web fiction creation:

Problem Manifestation Severity
"Talk therapy" proliferation A large number of chapters contain persuasion-type keywords, and multiple BOSSes are resolved through dialogue Fatal
Battle description collapse Battle scenes almost disappear in the later stages, with an extremely high repetition rate of imagery Fatal
Dialogue proportion out of control The proportion of dialogue continues to surge as the length progresses Severe
Pacing control failure The length exceeds the story's carrying capacity, with serious padding in the later stages Severe
Repetitive cyclic patterns Fixed patterns such as "awakening - upgrade - new enemy" repeat without variation Moderate
Villain intelligence reduction The final BOSS's logic collapses and they are easily persuaded Fatal

v2.0.1 Core Improvements

Improvement Item v1.1.1 v2.0.1
Battle scenes No specialized checks Added Dimension 30 "Battle Scene Audit" + Battle Description Guide
Dialogue proportion No monitoring Added Dimension 31 "Dialogue Proportion Monitoring", upper limit 40%
"Talk therapy" detection None Added Dimension 32 "Talk Therapy Pattern Detection", BOSSes prohibited from being persuaded
Cyclic patterns No detection Added Dimension 33 "Narrative Cycle Detection"
Pacing evaluation Word count tracking only Added pacing quality evaluation (climax density, conflict intensity)
Minimum audit set 5 items Expanded to 8 items (newly added battle, dialogue, talk therapy, cycles)

Core Positioning

This Skill solveseightcore problems:

  1. Long-form does not go off track: Maintain worldview, characters, foreshadowing, and status through the Truth Files system
  2. Can sign with Fanqie: Built-in Fanqie platform rules, commercial review, and paid conversion checks
  3. AI has no flavor: Intelligent AI de-flavoring check (replace rather than delete), avoiding platform throttling
  4. Session continuity: Seamless cross-session continuity through handoff_current.md
  5. Plans can be implemented: Progress tracking panel, real-time monitoring of word count/chapter deviation
  6. Battles have tension: Specialized audit of battle scenes, eliminating resolving BOSSes through "talk therapy"
  7. Dialogue not padded: Dialogue ratio monitoring, to prevent the entire chapter from turning into a chat log
  8. Rhythm does not collapse: Climax density tracking, to prevent late-stage padding and repetitive loops

Core Workflow

init -> write-next -> draft -> ai-detox -> audit -> revise -> extract-state -> state-update -> handoff-update

v2.0.1 Workflow Enhancements

写前(强制5项):
  |- 读取 current_state.md -> 确认地点/时间/状态
  |- 读取 outline.md -> 确认本章目标、字数、是否高潮章
  |- 读取 pending_hooks.md -> 确认待处理伏笔
  |- 检查 battle-scene-guide.md -> 如是战斗章,加载战斗描写清单
  |- 检查 dialogue-monitor.md -> 确认当前对话占比趋势

写后(强制8项 -- 最小审计集 v2.0.1):
  |- 1. 字数检查(2000-2500字)
  |- 2. AI去味(智能替换,非机械删除)
  |- 3. 因果链检查(维度29,无跳脱)
  |- 4. 行为-规则一致性检查(维度28)
  |- 5. 时间线检查(无黑洞)
  |- 6. 战斗场景检查(维度30,高潮章必查) [新增]
  |- 7. 对话占比检查(维度31,上限40%) [新增]
  |- 8. 话疗模式检查(维度32,BOSS禁止被说服) [新增]

状态更新(强制):
  |- current_state.md
  |- pending_hooks.md
  |- handoff_current.md
  |- progress_tracker.md
  |- pattern-detection.md [新增,循环模式追踪]

Project Structure

After initialization, the Tomato Novel project contains the following files:

Core Files (Must Maintain)

File Purpose Update Frequency
story_bible.md Worldview, settings, factions, power system When new settings appear
book_rules.md Tomato platform rules + protagonist lock-in + taboos At project start
outline.md Volume outlines, chapter goals, payoff plan Review at the end of each volume
current_state.md Current world state (most important) Updated every chapter
pending_hooks.md Foreshadowing tracking, suspense management Updated every chapter
handoff_current.md Session handoff file (living file) Updated every chapter
progress_tracker.md Progress tracking panel Updated every chapter
pattern-detection.md Loop pattern detection panel (new in v2.0.1) Updated every 5 chapters

Supporting Files (Recommended to Maintain)

File Purpose
chapter_summaries.md Chapter summaries and state changes
character_matrix.md Character relationships, trust/conflict, information boundaries
character_knowledge.md POV management, character knowledge boundaries
emotional_arcs.md Emotional arc tracking
subplot_board.md Subplot progress management
continuity_issues.md Known issue tracking

Style Files

File Purpose
style_guide.md Style guide (including AI de-flavoring rules)
style_profile.json Quantitative style statistics

Minimal Audit Set (Core improvement in v2.0.1)

After each chapter is writtenMust executeThe following 8 checks cannot be skipped:

Checks 1-5: Inherited from v1.1.1

See v1.1.1 documentation for details.

Check 6: Battle scene check (Dimension 30, new in v2.0.1)

触发条件:
  - 章节标题含"战斗/决战/对抗/最终/击杀"等关键词
  - outline.md 标记为"高潮章"
  - 本章出现BOSS级敌人

检查项:
  1. 是否有实质性攻击/防御动作?(至少3次)
  2. 是否通过"说服/理解/感化"解决冲突? -> 禁止!
  3. 战斗意象是否重复?
  4. 是否有战术博弈?(而非纯力量对轰)
  5. 是否有代价/牺牲?

评级标准:
  PASS 优秀:有战术、有代价、意象丰富
  WARN 及格:有战斗但模式化
  FAIL 不合格:无战斗或"话疗"解决

See:references/battle-scene-guide.md

Check 7: Dialogue ratio check (Dimension 31, new in v2.0.1)

计算方法:
  对话行数 = 统计引号内内容行数
  总行数 = 章节总行数
  对话占比 = 对话行数 / 总行数

标准:
  PASS 优秀:<= 25%(氛围/叙事为主)
  PASS 良好:26-35%(平衡得当)
  WARN 偏高:36-40%(需关注)
  FAIL 超标:> 40%(必须压缩对话)

趋势监控:
  - 记录每章对话占比
  - 若连续3章占比上升 -> 预警
  - 若占比 > 50% -> 立即报警

See:references/dialogue-monitor.md

Check 8: Talk-therapy mode check (Dimension 32, new in v2.0.1)

检测关键词:
  - "我理解你"
  - "说服"/"感化"/"唤醒"
  - "不是消灭,是救赎"
  - "融入"/"共存"/"和解"
  - BOSS被对话解决(三句话内放弃抵抗)

检查规则:
  1. 普通敌人:可通过对话解决(限1次/卷)
  2. 精英敌人:对话解决需铺垫3章+
  3. BOSS级敌人:禁止通过对话解决!必须战斗!

Check 9: Loop pattern check (Dimension 33, executed every 5 chapters)

检测模式:
  1. "遇到新敌人 -> 发现苦衷 -> 对话解决 -> 获得力量"是否重复?
  2. 连续3个章节结构是否高度相似?
  3. 高潮章是否真的有高潮?

预警规则:
  - 同模式出现2次 -> WARN 提醒
  - 同模式出现3次+ -> FAIL 报警,要求变化

See:references/pattern-detection.md


Progress Tracking Panel (Upgraded in v2.0.1)

New: Rhythm quality assessment

高潮密度:
  高潮章数量 / 总章数
  标准:>= 15%(每7章至少1个高潮)

冲突强度趋势:
  统计每章冲突解决方式
  战斗解决占比应 >= 对话解决占比

节奏曲线:
  绘制章节质量曲线
  检测是否"前高后低"崩塌

See:assets/project-template/progress_tracker.md


Session Continuity Workflow

New Session Startup Process

1. 用户请求继续写作
2. AI请求加载 handoff_current.md
3. 用户提供文件内容(或AI从项目文件夹读取)
4. AI确认加载成功,报告当前进度
5. 继续写作或审计

Session Handoff Protocol

AI must say at startup:

"Please provide the contents of handoff_current.md, and I will load the current progress."

After successful loading, must confirm:

"Loaded handoff_current.md. Current progress: ch{{N}}, next chapter: ch{{N+1}}. The protagonist is currently at [location], holding [item], with [X] pending foreshadowing threads. Total progress: {{actual}} words/{{target}} words ({{percent}}%). Dialogue ratio trend: {{trend}}. Last climax chapter: ch{{last_climax}}."


Tomato-Specific Features

1. Platform Red Line Check

See:references/fanqie-platform-rules.md

2. Life-or-Death Line Check

The first three chapters must pass; see:references/commercial-review.md

3. AI De-flavoring Special Check

See:references/ai-detox-guide.md

4. Commercial Review

See:references/commercial-review.md

5. Combat Scene Guide (new in v2.0.1)

See:references/battle-scene-guide.md

6. Dialogue Proportion Monitoring (new in v2.0.1)

See:references/dialogue-monitor.md


Lesson Library (v2.0.1)

The following are typical problem patterns summarized from extensive practical creation and in-depth review of long-form web fiction, and special attention must be paid to them during audits.

Problems 1-8: Inherited from v1.1.1

(Omitted, see v1.1.1 document)

Problem 9: Combat Description Collapse (new in v2.0.1, severe)

  • Manifestation: Combat scenes gradually disappear in the mid-to-late stages and are replaced by long stretches of dialogue
  • Root cause: No dedicated combat scene check; thematic expression overrides plot tension
  • Prevention: Mandatory check in Dimension 30; climax chapters must contain substantive combat description

Problem 10: Runaway Dialogue Proportion (new in v2.0.1, severe)

  • Manifestation: Dialogue proportion keeps soaring as the length progresses, and later chapters become almost dialogue-only
  • Root cause: No dialogue proportion monitoring; later writing tends to use dialogue to pad word count
  • Prevention: Continuous monitoring in Dimension 31, upper limit 40%, must be compressed if exceeded

Problem 11: Proliferation of the "talk therapy" pattern (new in v2.0.1, severe)

  • Manifestation: Multiple important enemies are resolved through dialogue/persuasion, lacking legitimate combat
  • Root cause: No "talk therapy" detection mechanism; thematic expression is abused as a universal solution
  • Prevention: Detection in Dimension 32; BOSS-level enemies are prohibited from being persuaded

Problem 12: Repeated Cyclical Patterns (new in v2.0.1, moderate)

  • Manifestation: Fixed narrative cycles (such as "awakening-leveling up-new enemy") recur without variation
  • Root cause: No cyclical pattern detection; writing falls into formulaic routines
  • Prevention: Detection in Dimension 33; mandatory alert when the same pattern appears 3 or more times

Problem 13: Villain Stupidity (new in v2.0.1, severe)

  • Manifestation: The ultimate BOSS's logic collapses and they are easily persuaded or defeated by the protagonist
  • Root cause: Eagerness to express the theme sacrifices the villain's behavioral logic and character dignity
  • Prevention: The BOSS must have an independent and self-consistent logical system; the protagonist needs to find logical flaws rather than simply "understanding" them

Problem 14: Pacing Collapse (new in v2.0.1, severe)

  • Manifestation: The first half of the work is stable in quality, while the second half suffers a disastrous decline
  • Root cause: No pacing quality assessment mechanism; later stages are padded and rushed
  • Prevention: Continuous monitoring of the pacing curve; timely alerts for a "high first, low later" trend

Version History

v2.0.1(2026-05-30)

  • Major upgrade: Comprehensive reconstruction based on long-form practical experience and in-depth review
  • Added Dimension 30: Combat scene audit (mandatory for climax chapters)
  • Added Dimension 31: Dialogue proportion monitoring (upper limit 40%)
  • Added Dimension 32: Talk therapy pattern detection (BOSS prohibited from being persuaded)
  • Added Dimension 33: Narrative loop detection (same pattern 3 times + alert)
  • New files:battle-scene-guide.md、dialogue-monitor.md、pattern-detection.md
  • Minimum audit set expansion: 5 items -> 8 items
  • Progress tracking upgrade: Added pacing quality assessment, climax density, conflict intensity trends
  • Lessons library expansion: Added issues 9-14

v1.1.1(2026-05-14)

  • Added: progress_tracker.md, dimension 29, minimum audit set, intelligent replacement table
  • Fixed: disconnection between planning and execution, AI de-flavoring breaking semantics, plot jumps without causality

v1.1.0

  • Added: handoff_current.md, dimension 28, lessons library

v1.0.0

  • Initial version