Editing Principles Used by No AI Slop: A Complete Guide to Human-First Text Editing

No AI Slop applies 17 distinct editing principles defined in skills/no-ai-slop/SKILL.md to remove generic AI-styled language while preserving the writer’s authentic voice, cadence, and intent.

No AI Slop is an open-source editorial skill for ChatGPT and Codex that distinguishes itself through a philosophy of minimal intervention and maximum voice preservation. According to the petergyang/no-ai-slop repository, the editing principles used by No AI Slop are explicitly codified in the source code to strip only "slop"—generic patterns typical of AI-generated content—while leaving human nuance, uncertainty, and personality intact.

The 17 Editing Principles Used by No AI Slop

The editorial framework documented in lines 26-42 of skills/no-ai-slop/SKILL.md centers on three pillars: voice preservation, minimal intervention, and mechanical precision.

Preserving Authentic Voice

The highest priority principles ensure the writer's humanity remains intact:

• Preserve the writer’s real voice: Maintain original vocabulary, cadence, humor, uncertainty, digressions, and polish level. Remove only generic or AI-styled language.
• Keep the user’s meaning: Never invent claims, examples, stats, or opinions; ask for clarification when ambiguous.
• Preserve useful edge and character: Retain strong opinions, blunt language, humor, profanity, self-interruptions, and honest admissions that belong to the writer.
• Make the minimum effective edit: Fix AI patterns, errors, repetition, and unclear passages while leaving strong human sentences untouched.

Structural and Clarity Principles

These guide when and how to reorganize content:

• Lead with the point when the setup adds nothing: Remove generic throat-clearing introductions; keep personal asides only when they add context or character.
• Front-load only when it improves clarity: Move conclusions earlier only if it aids understanding; avoid rigid point-detail-background mandates.
• Keep structure unless it hurts the piece: Maintain the writer’s progression and detours when they add personality; reorganize only with clear justification in the "What changed" section.
• Know the job: Understand the piece’s purpose and audience before adjusting structure or word choice.

Sentence-Level Mechanical Rules

Technical execution standards for individual sentences:

• Use active voice: Prefer direct constructions ("The team shipped it Tuesday") over passive or vague phrasing.
• Make verbs do the work: Replace weak phrases like "made a decision" with direct verbs like "decided."
• Be concrete and specific: Replace abstractions with concrete details, numbers, mechanisms, and examples.
• Untangle sentences without flattening cadence: Split only genuinely confusing sentences; preserve spoken fragments and rhythm when they remain clear.
• Make every sentence earn its place: Remove empty qualifiers and throat-clearing; keep authentic qualifiers ("I think", "maybe") when they convey the writer’s intent.
• Protect the specific fact: Keep concrete details intact; avoid smoothing them into generic statements.

Quality Control Tests

Validation methods to ensure edits enhance rather than homogenize:

• Use the portability test: If a sentence could be moved unchanged to a different context, it’s likely filler and should be cut or replaced with context-specific content.
• Show, don’t tell the reader what to think: Let facts, actions, and examples convey emphasis; delete meta-commentary that labels a point as important, surprising, or obvious.
• Open it up, don’t dumb it down: Retain nuance and precision; strip only jargon, overly long sentences, abstract nouns, and tangled structure.

How No AI Slop Applies These Principles

The skill executes two primary jobs defined in the repository codebase: Edit and Detect.

The Edit Job

When processing a draft, the skill reads the full text, identifies core points and voice traits, then applies the editing principles to remove AI patterns while preserving human elements. Each edit cycle concludes with validation against criteria specified in skills/no-ai-slop/eval.md.

The Detect Job

For identification-only mode, the skill flags AI-styled patterns without rewriting the text, allowing writers to make their own decisions about removal.

Practical Examples of No AI Slop Editing

The following examples demonstrate how the editing principles translate into actual text transformations.

Example 1: Removing Throat-Clearing

Input command:

/no-ai-slop
I think the product launch was a big win. Here's the thing, the team shipped it Tuesday. The best part: it learns.

Result:

The product launch was a big win. The team shipped it Tuesday. It learns.

What changed:

• Removed throat-clearing phrase "Here’s the thing" per the principle to lead with the point.
• Front-loaded the conclusion "It learns" as a plain sentence following the front-load only when it improves clarity rule.
• Preserved the authentic qualifier "I think" in the first sentence as it conveys the writer’s intent.

Example 2: Detecting AI Slop Patterns

Input command:

/no-ai-slop is this slop?
The best part: it learns.

Analysis result:

• Pattern identified: Colon reveal ("The best part: it learns")
• Fix suggested: "It learns."

This demonstrates the Detect job identifying a generic AI construction without automatically rewriting it, respecting the principle to make the minimum effective edit.

Key Implementation Files

The editing principles used by No AI Slop are implemented across these repository files:

• skills/no-ai-slop/SKILL.md: Defines the 17 editing principles, the Edit/Detect jobs, and the complete editorial workflow (lines 26-42 contain the principle definitions).
• skills/no-ai-slop/eval.md: Contains validation checks run after edits to ensure compliance with the principles.
• .codex-plugin/plugin.json: Metadata for publishing the skill as a ChatGPT/Codex plugin.
• scripts/build_plugin.py: Builds and validates the plugin package for distribution.

Summary

• No AI Slop follows 17 specific editing principles codified in skills/no-ai-slop/SKILL.md that prioritize voice preservation over standardization.
• The framework operates two distinct modes: Edit (minimal effective changes) and Detect (pattern identification without rewriting).
• Core philosophy centers on removing only generic AI patterns while preserving the writer's authentic cadence, uncertainty, humor, and edge.
• All edits are validated against criteria in eval.md to ensure compliance with the "minimum effective edit" standard.
• The principles emphasize concrete specificity, active voice, and structural changes only when they improve clarity.

Frequently Asked Questions

How does No AI Slop differ from standard grammar checkers?

Standard grammar checkers enforce standardized, often homogenized prose rules. According to the petergyang/no-ai-slop source code, No AI Slop specifically preserves the writer’s real voice including cadence, humor, and uncertainty, targeting only generic AI-styled language for removal. It applies the portability test to identify filler content that lacks context-specific value, something traditional checkers do not evaluate.

What files contain the editing principles used by No AI Slop?

The primary source is skills/no-ai-slop/SKILL.md, specifically lines 26-42, which enumerates all 17 principles. The validation logic resides in skills/no-ai-slop/eval.md, which checks that edits comply with rules like make the minimum effective edit and keep the user’s meaning. Build scripts in scripts/build_plugin.py package these definitions for distribution.

Does No AI Slop rewrite sentences automatically?

Only when running the Edit job. The skill also supports a Detect mode that identifies AI slop patterns—such as "colon reveals" or passive constructions—without rewriting them, allowing the writer to decide whether to remove flagged content. This dual approach ensures the minimum effective edit principle governs all interventions.

What is the "portability test" in No AI Slop?

The portability test is a quality control principle that evaluates whether a sentence could be moved unchanged to a different context. If the answer is yes, the sentence likely contains generic filler and should be cut or replaced with context-specific content. This test prevents the bland, universal phrasing characteristic of AI-generated text.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}