unkarelian

Fiction Prose Style — AI skill for Claude Code

AI community

A tested prose standard for AI-written fiction, packaged as an agent skill (Claude Code, Codex, OpenCode).

How to install Fiction Prose Style

This entry records only its repository, not the path inside it, so there is no exact command to give. Open unkarelian/fiction-prose-style and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Fiction Prose Style does

A tested prose standard for AI-written fiction, packaged as an agent skill (Claude Code, Codex, OpenCode).

Alternatives in AI

Karpathy LLM Wiki — Agent Skills-compatible LLM wiki for Claude Code, Cursor, and Codex 618 ★
Engrim — The Universal Cross-Model Episodic Memory Standard 219 ★
Sloptrim — A local detector for AI-writing patterns 196 ★

README

fiction-prose-style

A prose standard for AI-written fiction, packaged as an agent skill (for Claude Code, Codex, OpenCode and anything else that reads `SKILL.md` skills).

It was built from one reader's reactions, not from general advice: a ten-round blind side-by-side test of excerpts, a line-by-line markup of roleplay replies, the reader's own hand edits of model chapters, and measurements of model drafts against real, pre-2022 human fanfiction and against chapters the reader named as a book's gold standard. Prose written with it got called "the best literature written by AI I've ever read".

It is opinionated. Generic "anti-slop" guides tell you to strip stock body language, stated feelings and reflective endings; this one keeps those and goes after the *shapes* instead: "It wasn't X. It was Y.", the soft doubled echo, "he said it in a tone", the witty tail clause, narrated absence, every scene ending in calm, and the storybook narrator who watches the character instead of thinking with them.

What's in it

Path	What it is
`SKILL.md`	The skill: summary of the standard and the working method (drafting, checking, judging, briefing other agents, tell hunts)
`references/PROSE_PREFERENCES.md`	The full standard, in the reader's own voice. This is the file to give any model that writes prose
`references/REVIEWER.md`	Instructions for a fresh reviewer subagent that flags only the shapes a script can't count
`references/PROSE_PREFERENCES_LOG.md`	The evidence behind each rule, version by version
`references/baseline.json`	Rates and word frequencies from about 60,000 words of liked chapters, used by the scanner
`scripts/prose_check.py`	The scanner: rhythm, repeated phrases, overused words, simile and punctuation density, book-length habits, fixed wordings
`scripts/human_compare.py`	Fetches pre-2022 human works from AO3 into a local cache and measures a draft against them
`scripts/gold_judge.py`	Blind compari