agentsclimarketplace

Novel character consistency

Skill ThanhTNV/Novel-Generator/.cursor/skills/novel-character-consistency

Retrieve and maintain character consistency in novel generation by querying Milvus for all passages mentioning a character. Ensures character traits, appearance, relationships, and voice remain consistent across thousands of tokens. Use when continuing a story, introducing recurring characters, or when the model generates new scenes involving existing characters.From its SKILL.md

Install
npx -y skills add ThanhTNV/Novel-Generator --skill novel-character-consistency

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.3 KB, 602 tokens by cl100k_base, as published. Nobody here has run it

Novel Character Consistency

Maintain accurate character representation by retrieving relevant passages from Milvus before generating new content.

Workflow

Before writing new character scenes:

  1. Identify characters in the prompt

    • List all characters that will appear in the new section
    • Note primary vs. secondary characters
  2. Query Milvus for character passages

    • For each character, query: "[character name] appearance, personality, relationships"
    • Retrieve top 5-10 most relevant chunks
    • Combine results into a character profile
  3. Extract key details From retrieved passages, document:

    • Physical description: eye color, height, distinguishing features
    • Personality traits: temperament, speech patterns, quirks
    • Relationships: how they relate to other characters
    • Role in plot: their motivation and story arc status
  4. Inject into prompt context

    • Include extracted details in the system prompt
    • Format as: "Character: [name]\n- Appearance: ...\n- Personality: ...\n- Relationships: ..."
    • Place character profiles in explicit context section before generating
  5. Generate with grounding

    • Model now writes new scenes with retrieved context
    • Consistency automatically enforced through prompt injection
    • New passages automatically embeddable into Milvus for future retrieval

Character Profile Template

Use this format when injecting character data into prompts:

Character: [Character Name]
- Appearance: [physical traits from retrieved passages]
- Personality: [temperament and speech patterns]
- Relationships: 
  - [Character A]: [relationship type and status]
  - [Character B]: [relationship type and status]
- Current arc: [their unresolved story threads]
- Contradictions to avoid: [traits that shouldn't change]

Query Strategy

Effective Milvus queries for character retrieval:

  • "[character] appearance, eyes, hair, clothing" → physical consistency
  • "[character] said, spoke, thought" → voice and dialogue patterns
  • "[character] relationship with [other character]" → interpersonal dynamics
  • "[character] backstory, origin, history" → motivations and depth
  • "[character] personality, emotional, reaction" → behavioral consistency

Common Contradictions to Prevent

Track and verify these don't change mid-narrative:

  • Physical traits (eye color, scars, distinguishing marks)
  • Speech patterns and dialect
  • Core personality traits (temperament, values)
  • Relationship status with other characters
  • Character's death/absence (major plot points)
  • Skills and abilities established earlier

Tips

  • Query before each major character appearance
  • Include secondary characters mentioned peripherally
  • When characters interact, retrieve profiles for all involved
  • After generating, embed new passages to update character knowledge in vector DB

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,422. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.