sumeru-polish

Web Fiction Content Polishing

Trigger Keywords

Help me polish this novel passage, improve the writing style, optimize the chapter pacing, strengthen this satisfying moment, make the dialogue more natural, make this passage more satisfying, optimize the novel's writing style, adjust the chapter pacing, make the dialogue more realistic, help me revise this passage, polish the novel, optimize the satisfying moment, improve the writing style, make this passage more immersive, novel content optimization, writing style polishing

Core Features

Polishing Levels

1. Light Polishing: Optimize sentence expression, remove redundant wording, refine word choice, improve text fluency, retain more than 80% of the original style and expression
2. Moderate Polishing: Restructure paragraphs, optimize narrative perspective, comprehensively improve the texture of the writing, retain 60% of the original core expression
3. Deep Polishing: Polish word by word, refine details, pursue the best reading experience, retain the core plot thread

Style Adaptation Options

• Beginner-Friendly Satisfying Fiction: Mainly short sentences, brisk pacing, direct emotions, accessible and easy-to-understand wording
• Premium Fiction: Varied sentence patterns, delicate writing, emphasis on atmosphere creation, in-depth portrayal of characters' psychology
• Ancient-Style Xianxia: Elegant wording, distant artistic conception, appropriate use of classical vocabulary and rhetorical devices
• Urban Realism: Lifelike language, down-to-earth dialogue, realistic and perceptible scene description
• Mystery and Deduction: Concise language, tight pacing, moderate information density, strong sense of suspense
• Science Fiction Future: Accurate technological vocabulary, rigorous logic, clear expression of the worldview

Targeted Optimization

1. Tighten Pacing

   • Remove ineffective setup: remove scene descriptions and inner thoughts unrelated to the main storyline
   • Compress transitional plot: shorten lengthy transitions and expository text by 30%-50%
   • Speed up information delivery: use dialogue and action to present information, reduce long stretches of narration
   • Adjust paragraph length: split overly long paragraphs into short paragraphs of 2-3 lines to improve reading speed

2. Strengthen Satisfying Moments

   • Front-Loaded Setup: Set up a sense of anticipation 300-500 words before the satisfying moment, through methods such as villain provocation, protagonist's predicament, or others' doubts
   • Emotional Escalation: The explosion of the satisfying moment is divided into three layers: anticipation → tension → release. Each layer of emotion must have specific description
   • Magnify Details: Describe simultaneously from three dimensions: the protagonist's actions, others' reactions, and environmental changes, to enhance the visual sense
   • Powerful Ending: After the satisfying moment, use 100-200 words to summarize the effect, or leave a new hook
   • Classic Structure: [Suppression/Provocation]→[Charging Up/Preparation]→[Explosion/Face-Slapping]→[Aftermath/Impact]

3. Dialogue Optimization

   • Character Consistency: Each line of dialogue matches the character's age, identity, and personality, avoiding a one-size-fits-all portrayal
   • Subtext: Important dialogue contains implied meaning, expressed with the aid of tone and actions
   • Pacing control: Split long stretches of dialogue into multiple rounds, interspersing actions, expressions, and psychological description
   • Information delivery: Naturally convey background and advance the plot through dialogue, avoiding stiff exposition
   • Colloquial style: Reduce written-language phrasing, use more expressions from real life, but avoid grammatical errors
   • Dialogue tag optimization: Replace "said" and "stated" with specific actions, such as "raised an eyebrow and said," "gave a cold snort"

4. Writing optimization

   • Precise word choice: Replace repeated words and vague expressions, choose more accurate verbs and adjectives
   • Sentence pattern variation: Combine long and short sentences, avoid a uniform subject-verb-object structure
   • Sensory description: Engage sight, hearing, smell, touch, and taste to enhance immersion
   • Concrete metaphors: Use things familiar to readers for comparison, avoid abstract or obscure metaphors
   • Remove redundancy: Delete adverbs such as "very," "extremely," and "quite," and express degree with verbs and nouns

Output content

• The complete polished chapter content (directly replace the original file in chapters/ )
• Polishing revision notes: what was adjusted and why (explained by optimization type)
• Optimization suggestions: directions for improving subsequent writing, with specific actionable techniques

Sub-Agent parallel polishing mechanism

When the number of chapters to be polished is greater than 3, the sub-Agent parallel polishing mode is automatically enabled:

⚠️ Follow the global constraint: each sub-Agent handles at most 3 chapters(see AGENTS.md "Sub-Agent Parallel Processing Rules" for details)

• Number of Agents required = ceil(total chapters / 3)
• Adjacent chapters are assigned to the same Agent to maintain stylistic continuity

Scheduling logic

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]

Sub-Agent input context Each sub-agent receives:

1. Polishing level and style parameters
2. The original content of the chapters it is responsible for
3. Worldbuilding settings (condensed version, to keep terminology consistent)
4. Character settings (only relevant characters, to keep personalities consistent)
5. Review issue list (if any, as key optimization directions)

Chapter assignment rules

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

Progress tracking

• After each Agent finishes polishing, it immediately saves the result directly to chapters/(automatically backed up before modification to .sumeru/write/original/）
• Display in real time the status of completed/in-progress/pending chapters
• After all Agents are finished, the polishing results have been directly applied to the chapters/ directory

Data persistence

Polishing process data is automatically saved to the .sumeru/polish/ directory:

• diff/: Revision comparison file, recording the location, original text, revised content, and reason for each revision
• summary.json: Polishing statistics report, including the number of revisions, distribution of optimization types, and word count changes
• style-config.json: Style, level, and focus parameter configuration used for this polish

Working with Other Skills

• Preceding Skill: Reads sumeru-write and sumeru-review output
  • From chapters/ read the original chapter content
  • From .sumeru/review/issues.json read review issues as optimization focus
• Subsequent Skill: The polished content is for sumeru-finalize use
  • Polish results are saved directly in the chapters/ directory, no additional application step required
  • When worldbuilder orchestrates polish, polish results take effect automatically

Data Reuse

• Supports multiple polish rounds, can continue optimizing based on the previous polish result
• Can export diff files for manual review and modification of content
• Style configurations can be reused to keep polish style consistent across the entire book