sumeru-finalize

Web Novel Completion Validation

Trigger Keywords

I've finished writing the novel, help me check it, export a format suitable for Qidian/Fanqie, check for typos, detect sensitive words, organize into a release version, help me export the novel in a publishing format, check the novel for typos, sensitive word detection, novel completion check, export adapted to Qidian format, Fanqie novel format export, novel layout organization, multi-platform format export, novel completion export

Core Features

1. Typo, punctuation, and grammar error checking
2. Sensitive content and violating content screening
3. Format standardization and unification: chapter titles, paragraph formatting, punctuation standards
4. Full-text word count statistics, completion report generation
5. Export adapted to different platform publishing formats
6. Batch replacement function
7. Automatic paragraph splitting function

Skill Boundary Description: Technical text validation (typos/punctuation/grammar) is handled by this Skill,sumeru-polish Skill focuses on optimization at the writing style and content level; the two are complementary and do not overlap. After polishing, users should still perform final technical validation through finalize.

Sub-Agent Parallel Validation Mechanism

When the number of chapters to be validated/exported is greater than 3, the sub-agent parallel processing mode is automatically enabled:

⚠️ Follow global constraints: each sub-agent is responsible for at most 3 chapters(see AGENTS.md "Sub-Agent Parallel Processing Rules" for details)

• Required number of Agents = ceil(total number of chapters / 3)
• Adjacent chapters are assigned to the same Agent to maintain consistency in format processing

Scheduling Logic

flowchart LR
    A[批量校验任务] --> B[读取章节列表与校验配置]
    B --> C[创建任务队列，按每Agent最多3章分配]
    C --> D[计算所需Agent数 = ceil/总章数/3/]
    D --> E[启动N个并行子agent]
    E --> F[子Agent拉取任务 → 校验章节 → 生成错误报告 → 导出格式]
    F --> G{队列是否为空?}
    G -->|否| F
    G -->|是| H[汇总校验结果，合并错误报告，生成完稿报告]

Chapter Assignment Rules

• Assign consecutively in chapter order (e.g., Agent1 handles chapters 1-3, Agent2 handles chapters 4-6)
• For an Agent with fewer than 3 chapters remaining at the end, assign according to the actual number of remaining chapters

Validation Scope of Each Sub-Agent

• Typo, punctuation, and grammar error checking
• Sensitive word detection (according to the three-level standard)
• Format standardization and unification
• Export in the corresponding platform format

Output Content

• Error list: typos, punctuation errors, grammar issues
• Sensitive content prompts: violating content that needs adjustment
• Validated clean full text
• Completion report: total word count, number of chapters, core content summary
• Multi-platform publishing format versions (Qidian, Fanqie, Jinjiang, etc.)

Export Format Rules for Each Platform

Qidian Chinese Network (qidian)

• Chapter title format:第X章 标题内容, center-aligned
• Paragraph first line indented by 2 characters
• One blank line between each paragraph
• Punctuation uses Chinese full-width
• Recommended chapter word count: 3000-5000 words
• Special symbols are prohibited as chapter titles
• Dialogue forms its own paragraph

Fanqie Novel (fanqie)

• Chapter title format:第X章 标题内容
• Paragraph first line is not indented, one blank line between paragraphs
• Keep each sentence as short as possible, suitable for mobile reading
• Recommended chapter length: 2000-3000 words
• Key content can be marked in bold
• Dialogue should be wrapped in quotation marks, with the speaker placed in a separate paragraph or noted at the end of the sentence

Jinjiang Literature City (jjwxc)

• Chapter title format:第X章 标题内容
• Supports HTML format tags
• First line of each paragraph indented by 2 characters
• Recommended chapter length: 2500-4000 words
• Author's note area set separately
• Supports chapter summary

Zongheng Chinese Network (zongheng)

• Chapter title format:第X章 标题内容
• First line of each paragraph indented by 2 characters
• Recommended chapter length: 3000-6000 words
• Supports volume division settings
• Punctuation should use standard Chinese full-width characters

17K Novel Network (17k)

• Chapter title format:第X章 标题内容
• First line of each paragraph indented by 2 characters
• Recommended chapter length: 2000-4000 words
• Supports chapter preview
• A next chapter preview can be set at the end of each chapter

Sensitive word detection standards

Level 1 sensitive (must be modified)

• Content that violates national laws and regulations
• Involving politically sensitive figures or events
• Pornographic or obscene descriptions
• Violent or terrorist content
• Speech that splits the country or undermines ethnic unity
• Religiously extremist content

Level 2 sensitive (recommended to modify)

• Overly bloody or violent detailed descriptions
• Vulgar language or profanity
• Medical descriptions that may cause discomfort
• Inappropriate content involving minors
• Descriptions related to gambling or drugs
• Content that infringes on others' privacy

Level 3 sensitive (optimization suggestions)

• Too much internet slang affecting reading
• Expressions prone to ambiguity
• Topics that may cause controversy
• Overuse of internet memes
• Repeated and redundant expressions

Error severity prompts

Serious errors (marked in red)

• Typos that completely change the meaning
• Level 1 sensitive word violations
• Chapter title format completely fails to meet standards
• Severely chaotic paragraph structure
• Large-scale punctuation errors

Moderate errors (marked in yellow)

• Ordinary typos
• Nonstandard punctuation usage
• Level 2 sensitive word content
• Inconsistent paragraph formatting
• Grammatical errors affecting comprehension

Minor errors (marked in blue)

• Wording suggested for optimization
• Punctuation usage could be more standard
• Level 3 sensitive word content
• Paragraph layout could be further beautified
• Suggestions for repetitive expressions

Batch replacement function description

Feature characteristics

• Supports global batch replacement of specified terms
• Supports regular expression replacement
• Supports preview and confirmation before replacement
• Supports simultaneous execution of multiple sets of replacement rules
• Supports querying replacement history

Usage scenarios

1. Unified modification of character names
2. Batch adjustment of place names and setting names
3. Batch replacement of sensitive words
4. Unified standardization of punctuation
5. Batch conversion of internet slang

Operation method

• Preset rules: Select a commonly used replacement rule template
• Custom rules: Manually enter the search content and replacement content
• Regex mode: Use regular expressions for complex matching
• Confirm replacement: Review the replacement preview and then confirm execution

Automatic paragraphing feature description

Feature characteristics

• Intelligently identify dialogue and narrative content
• Automatically paragraph based on sentence length
• Support custom word count thresholds for paragraphing
• Maintain the logical integrity of paragraphs
• Dialogue automatically becomes a separate paragraph

Paragraphing rules

1. Dialogue priority: Dialogue content automatically becomes a separate paragraph
2. Word count control: A single paragraph is recommended to be 100-300 characters
3. Logical completeness: Avoid paragraphing in the middle of a sentence
4. Scene switching: Automatically paragraph when the scene changes
5. Psychological activity: Appropriately paragraph long psychological descriptions

Configurable parameters

• Maximum paragraph word count (default 300 characters)
• Minimum paragraph word count (default 50 characters)
• Whether to force dialogue to become a separate paragraph
• Whether to add a separator line when the scene changes
• Whether to retain the original paragraph structure

Data persistence

Completed manuscript data is automatically saved to the .sumeru/finalize/ directory:

• clean/full-text.md: The verified clean full text
• clean/chapters/: Clean files split by chapter
• error-report.json: Error list, including all issues such as typos, punctuation, sensitive words, etc.
• stats.json: Completed manuscript statistical report, total word count, number of chapters, average chapter length, etc.
• export-config.json: Export configuration parameters for each platform

User-visible output (current working directory)：

• publish/: Export versions for each platform, stored by platform name

Coordination with other Skills

• Preceding Skill: Read the final chapter content
  • By default, read from chapters/ to read the latest chapter content (review fixes and polish directly modify chapters/, no additional operation is needed)

Data reuse

• Other platform formats can be re-exported at any time without re-verification
• Error reports can serve as reference for avoidance in subsequent writing
• Support incremental export; after modifying some chapters, only regenerate the platform versions of the corresponding chapters