sumeru-polish

Online Novel Content Polishing

Trigger Keywords

Core Features

Polishing Levels

1. Light Polishing: Optimize sentence structure, remove redundant expressions, refine word choice, improve text fluency, and retain over 80% of the original style and expression
2. Moderate Polishing: Restructure paragraph frameworks, optimize narrative perspectives, comprehensively enhance writing quality, and retain 60% of the original core expressions
3. Deep Polishing: Polish every word, refine details, pursue the best reading experience, and retain the core plot context

Style Adaptation Options

• Casual Power Fantasy Novels: Dominated by short sentences, brisk rhythm, direct emotions, and easy-to-understand vocabulary
• High-Quality Literary Novels: Varied sentence structures, delicate writing style, focus on atmosphere building, and in-depth portrayal of characters' psychology
• Ancient Fantasy & Xianxia Novels: Elegant vocabulary, distant artistic conception, appropriate use of classical words and rhetorical techniques
• Urban Realistic Novels: Lifelike language, down-to-earth dialogues, and realistic scene descriptions
• Mystery & Detective Novels: Concise language, tight rhythm, moderate information density, and strong sense of suspense
• Sci-Fi & Futuristic Novels: Accurate tech-related vocabulary, rigorous logic, and clear worldview presentation

Targeted Optimization

1. Rhythm Tightening
   • Remove ineffective foreshadowing: Delete scene descriptions and psychological activities irrelevant to the main plot
   • Compress transitional plots: Shorten lengthy transitions and explanatory texts by 30%-50%
   • Accelerate information delivery: Use dialogues and actions to present information instead of long paragraphs of narration
   • Adjust paragraph length: Split overlong paragraphs into short paragraphs of 2-3 lines to improve reading speed
2. Highlight Enhancement
   • Pre-foreshadowing: Build anticipation 300-500 words before the highlight through villain provocation, protagonist's dilemma, others' doubts, etc.
   • Emotional progression: The highlight outbreak is divided into three layers: anticipation → tension → release, with specific descriptions for each emotional layer
   • Detail amplification: Simultaneously describe the protagonist's actions, others' reactions, and environmental changes to enhance sense of imagery
   • Powerful conclusion: Use 100-200 words to summarize the effect after the highlight, or leave a new hook
   • Classic structure: [Suppression/Provocation] → [Accumulation/Preparation] → [Outburst/Slap in the Face] → [Aftermath/Impact]
3. Dialogue Optimization
   • Consistent with character settings: Every line of dialogue matches the character's age, identity, and personality, avoiding one-size-fits-all expressions
   • Subtext: Important dialogues contain implied meanings, supplemented by tone and actions
   • Rhythm control: Split long dialogues into multiple rounds, interspersed with actions, expressions, and psychological descriptions
   • Information delivery: Naturally explain background and promote plot through dialogues, avoiding rigid explanations
   • Colloquialization: Reduce written language, use real-life expression habits, but avoid grammatical errors
   • Dialogue tag optimization: Replace "said" with specific actions, such as "raised his eyebrow and said", "snorted"
4. Writing Style Optimization
   • Precise word choice: Replace repeated words and vague expressions with more accurate verbs and adjectives
   • Sentence variety: Combine long and short sentences, avoid uniform subject-verb-object structures
   • Sensory description: Mobilize vision, hearing, smell, touch, and taste to enhance immersion
   • Concrete metaphors: Use things familiar to readers for comparison, avoid abstract and obscure metaphors
   • Remove redundancy: Delete adverbs like "very", "extremely", "quite", and express intensity with verbs and nouns

Output Content

• Complete polished chapter content (directly replaces original files in

  chapters/

  )
• Polishing modification instructions: Adjusted parts and reasons (classified by optimization type)
• Optimization suggestions: Specific actionable techniques for improving subsequent content writing

Sub-Agent Parallel Polishing Mechanism

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

flowchart LR
    A[批量润色任务] --> B[读取章节列表与润色配置]
    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 → 润色章节 → 保存结果]
    F --> G{队列是否为空?}
    G -->|否| F
    G -->|是| H[汇总润色结果，生成统计报告]

1. Polishing level and style parameters
2. Original content of the responsible chapters
3. Simplified worldview settings (to maintain consistent terminology)
4. Character settings (only relevant characters, to maintain consistent personalities)
5. Review issue list (if any, as key optimization directions)

• Allocate continuously in chapter order (e.g., Agent 1 is responsible for Chapters 1-3, Agent 2 for Chapters 4-6)
• Agents with less than 3 remaining chapters at the end are assigned according to the actual number of remaining chapters

• After completing polishing, each Agent immediately saves the results directly to

  chapters/

  (automatic backups are made to

  .sumeru/write/original/

  before modification)
• Real-time display of completed/in-progress/to-be-polished chapter status
• After all Agents are completed, the polished results are directly applied to the

  chapters/

  directory

Data Persistence

.sumeru/polish/

• diff/

  : Modification comparison files, recording the position, original text, modified content, and modification reason for each change

• summary.json

  : Polishing statistical report, including number of modifications, distribution of optimization types, and word count changes

• style-config.json

  : Style, level, and key parameter configurations used in this polishing

Coordination with Other Skills

• Preceding Skills: Reads outputs from

  sumeru-write

  and

  sumeru-review

  • Reads original chapter content from

    chapters/

  • Reads review issues from

    .sumeru/review/issues.json

    as key optimization directions
• Subsequent Skills: Polished content is used by

  sumeru-finalize

  • Polished results are directly saved in the

    chapters/

    directory, no additional application steps required
  • When worldbuilder orchestrates polish, the polished results take effect automatically

Data Reuse

• Supports multiple rounds of polishing, allowing continued optimization based on previous polishing results
• Diff files can be exported for manual review of modifications
• Style configurations can be reused to maintain consistent polishing style throughout the entire novel