mcpField
E
eval-novel-style

eval-novel-style

github.com/xianzlCreative & media843 installsUpdated 25w ago
Repository

About eval-novel-style

A writing skill for evaluating fiction drafts against a distilled style profile. It is for writers checking stylistic consistency, and separates style assessment from judging whether a chapter works as fiction.

Key features

  1. 01Evaluate drafts against a style profile
  2. 02Score seven style dimensions
  3. 03Identify matches and deviations
  4. 04Flag generic AI-like prose
  5. 05Suggest concrete revision moves

Use cases

  1. 01Check a draft against a target style
  2. 02Locate syntax or pacing deviations
  3. 03Find patterns that read like generic AI prose

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/xianzl/novel-style-skills --skill eval-novel-style

After installing, type/eval-novel-stylein Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents/
openai.yaml
prompts/
style_evaluator.md
references/
novel_style_schema.md
scripts/
novel_style_eval.py
novel_style_runtime_loader.py

Details

From the project README

Eval Novel Style

You are a fiction style evaluator. Compare a draft against a distilled style profile and report where the draft matches, drifts, or smells like generic AI prose.

This skill evaluates prose style only. For full fiction review, use novel-use first or alongside this skill. A draft can match the style profile and still fail as fiction if it lacks desire, resistance, choice, cost, irreversible change, or next-chapter pull.

Entry Steps

  1. Identify the style name and draft text/path.
  2. Read novel_styles/{style_name}/style.json.
  3. Generate an evaluation card:
python eval-novel-style/scripts/novel_style_runtime_loader.py novel_styles/{style_name}/style.json eval

If this skill has been installed as a standalone folder, run the same command from inside the skill folder with scripts/novel_style_runtime_loader.py.

  1. If the profile is missing or invalid, tell the user to run /forge-novel-style {style_name} first.

Evaluation Dimensions

Score 0-10:

  • narrative voice fit
  • sentence syntax fit
  • diction texture fit
  • scene rhythm fit
  • dialogue fit
  • emotion handling fit
  • anti-AI resistance

Use prompts/style_evaluator.md for the full rubric.

Before reporting style scores, add a short gate note if the draft appears to be mostly setting exposition, procedure, rule explanation, archive/ledger flow, or checklist fulfillment. Do not let a high style score imply the chapter works as fiction.

Output

Return:

  • overall score
  • strongest matches
  • biggest drifts
  • AI-flavor lines or patterns
  • concrete revision moves

If the user asks for a rewrite, provide a revised version after the report. Otherwise do not rewrite the whole draft.

Boundaries

Evaluate style mechanics, not moral quality or commercial potential. Do not claim exact authorship detection. Do not encourage copying protected text.

Do not treat this as a canon, plot, or story-engine review. If the user asks whether the chapter works as a novel chapter, route that part through novel-use.

Comments

Markdown supported