theclaymethod/unslop

An agent skill to de-AI your writing

What it solves

unslop identifies and removes formulaic, AI-generated writing patterns (often called "slop") while preserving the original meaning, voice, and factual accuracy of the text. It prevents the "AI tell"—the predictable rhythms, phrases, and structures that make machine-generated prose feel artificial.

How it works

The system uses a three-layer detection stack to find problematic writing:

  1. Phrase Layer: Scans for specific "banned" phrases and triggers (e.g., "rich tapestry," "let's dive in") with contextual protection to avoid flagging legitimate technical or legal jargon.
  2. Structure Layer: Analyzes document-level metrics like sentence burstiness, paragraph openers, and the use of "moralizing codas" to catch rhythmic tells.
  3. Silhouette Layer: Evaluates the high-level arrangement of ideas, flagging things like recap loops and template-driven paragraph roles.

Once detected, the tool can either flag issues as reviewable suggestions (/unslop cleanup) or rebuild the draft under strict guards (/unslop rewrite). It also includes a "voice" system (/unslop teach and /unslop mimic) that analyzes a user's own writing samples to create a stylometric fingerprint, allowing the AI to draft in a specific human voice without triggering the "slop" detectors.

Who it’s for

  • Writers and editors who want to remove AI-generated markers from their prose.
  • Developers integrating AI writing assistants into their workflows via agent skills.
  • Users who want to maintain a consistent, personal writing voice when using LLMs.

Highlights

  • Contextual Gating: Avoids blunt word bans by ensuring technical, medical, or legal terms are not flagged when used in their proper domain.
  • Stylometric Voice Profiling: Uses character 3-grams and sentence-length distribution to mimic a specific user's voice based on approved samples.
  • Strict Preservation Guards: Employs magnitude-aware checking to ensure facts (like numbers and dates) are not altered during rewrites.
  • Agent-Ready: Ships as a skill that can be installed into coding agents like Claude Code or Cursor.

Related

  • Project
    kermitt2/delft

    A Keras and TensorFlow framework for text processing that supports rich text layout information for sequence labeling and text classification.

  • Project
    dreadnode/rigging

    A lightweight LLM framework that simplifies production code integration by providing unified model access, structured Pydantic outputs, and type-safe prompt definitions.

  • Project
    usepr/eva

    A lightweight, single-file AI agent that executes shell commands and writes scripts, featuring directory-scoped sessions and safety reviews.

  • Project
    zmzhace/SeedWorld

    An emergent world simulation engine that uses LLMs to drive autonomous characters to create unscripted, evolving civilizations and social dynamics.