AI Prose-Pattern Detector

AI Prose-Pattern Detector is an agent-readable prose QA toolkit for finding writing patterns that make text sound generic, over-processed, or AI-shaped.

It does not claim to prove who wrote a draft. Instead, it gives writers a practical review system for visible AI-likeness and prose-quality patterns: filler openings, lifeless verbs, decorative adjectives, formulaic contrast, condensed expert prose, mechanical rhythm, generic closers, and formatting that looks more organized than the thinking underneath it.

What It Does

• Runs quick flag checks on pasted prose.
• Produces detailed audits with exact excerpts and rewrite suggestions.
• Rewrites text while preserving the user's facts, claims, and domain detail.
• Helps agents produce prose that sounds clear, specific, and human without becoming casual or vague.
• Centralizes customization in one settings file.

Why It Exists

AI-assisted writing often fails in two opposite ways. Some drafts become generic and padded. Others become compressed expert prose: long noun chains, comma-packed capability catalogs, and scope-setting phrases that sound sharp once but robotic after repeated use.

AI Prose-Pattern Detector treats both as quality problems. The goal is not to make every piece sound informal. The goal is to preserve judgment, specificity, and voice.

Package Structure

AI Prose-Pattern Detector/
  AGENTS.md
  README.md
  detector-settings.md
  references/
    prose-quality-rubric.md
  examples/
    sample-input.md
    sample-audit.md
    sample-rewrite.md

Quick Start

Use the folder as an agent project or copy its contents into a supported AI workspace.

1. Put AGENTS.md, detector-settings.md, and references/ in the project.
2. Paste prose into the agent chat.
3. Ask for one of the supported modes:
   • /ai-prose-check
   • /prose-pattern
   • flag check
   • audit this
   • rewrite using the detector
   • produce this in a natural voice

Examples are optional. They show the intended output style but are not required for normal use.

Modes

Flag Check: Returns risk level, flag count, categories fired, exact excerpts, and one priority fix.

Audit: Ranks findings by severity and gives suggested rewrites for each issue.

Rewrite: Returns revised prose first, preserving meaning and factual claims.

Produce: Uses the rubric silently while drafting new prose.

Customization

All behavior settings live in detector-settings.md:

• default mode
• rewrite strength
• protected content
• condensed prose thresholds
• contrast phrase policy
• output preferences
• severity scale

The rubric file defines the pattern library. The agent guide tells the agent when to load which files.

Example Use

/ai-prose-check

This is not just a reporting process, it is a comprehensive operating cadence
that leverages stakeholder alignment, governance, prioritization, visibility,
decision quality, and execution discipline to drive meaningful outcomes.

The detector should flag formulaic contrast, vague verbs, decorative abstraction, and compressed comma-heavy prose. A better rewrite would name the practical function of the process and split the crowded sentence.

Responsible Use

AI Prose-Pattern Detector should not be used to accuse a writer of using AI. Many flagged patterns also appear in human business writing, especially rushed executive drafts, consulting prose, and technical documentation.

Use it as a writing quality gate: identify weak patterns, preserve the real thinking, and improve the draft.

Suggested GitHub Metadata

Repository description:

Agent-readable AI prose-pattern detector for finding AI-shaped writing patterns and rewriting drafts into clearer, more specific human prose.

Suggested topics:

writing
prose
ai-writing
ai-detection
ai-prose
prompt-engineering
editing
style-guide
agent-workflow
quality-assurance
portfolio-project