No AI Slop Demo: 3 Ways to Test the Codex Skill With Real Examples

The no ai slop demo runs entirely inside ChatGPT or Claude using prompt-only logic, offering two interactive modes—Edit and Detect—plus a local build script to package the skill for distribution.

The no ai slop repository by petergyang implements a Codex-style skill that identifies and removes generic AI writing patterns. Unlike traditional plugins, this demo requires no external runtime code; all behavior is driven by markdown configuration files interpreted by the host AI platform.

Live Demo 1: Edit Mode (Default)

The primary no ai slop demo experience is the Edit mode, accessed via the default prompt defined in .codex-plugin/plugin.json.

Input Example

Trigger the skill by starting your message with the skill command followed by text containing common AI patterns like binary contrast or colon reveals:

/no-ai-slop
I think the future isn’t coming. It’s already here. The best part: it learns.

Output Result

The skill processes the input through the rules defined in skills/no-ai-slop/SKILL.md and returns a cleaned version:

I think the future is already here. The best part is that it learns.

What changed: The demo stripped the binary contrast construction (isn’t coming. It’s already here) and converted the colon reveal (The best part: it learns) into a plain sentence while preserving the author's voice.

Technical Source

According to the source code in petergyang/no-ai-slop, the editing logic resides entirely in SKILL.md. This file contains the editing principles, pattern-cut lists, and transformation rules that the host AI applies to the input text. No external API calls or code execution occurs during this demo.

Live Demo 2: Detect Mode ("Is This Slop?")

The second no ai slop demo mode flags problematic patterns without rewriting the text, useful for manual content review.

Triggering Pattern Detection

Use the alternative prompt specified in the plugin manifest:

/no-ai-slop is this slop?
Here’s the thing, the model does X. The best part: it learns.

Pattern Analysis Output

The demo returns a structured report identifying specific AI slop patterns:

Pattern: Binary contrast – “It’s not X. It’s Y.”
Quote: “Here’s the thing, the model does X.”
Fix: Remove the throat‑clearing opener and state the point directly.

Pattern: Colon reveal – “The best part: it learns.”
Quote: “The best part: it learns.”
Fix: Rewrite as a plain sentence.

This output demonstrates how SKILL.md categorizes violations—flagging throat-clearing openers and binary contrasts—while leaving the rewrite decision to the user.

Local Build Demo: Packaging the Skill

To run the no ai slop demo outside hosted platforms or distribute modified versions, use the build script provided in the repository.

Building the Plugin Package

Execute the Python script scripts/build_plugin.py from the repository root:

python scripts/build_plugin.py

This command performs the following actions:

• Validates the manifest file .codex-plugin/plugin.json
• Copies SKILL.md and eval.md into the distribution bundle
• Packages the icon, license, and metadata into a zip archive
• Outputs dist/no-ai-slop-plugin-1.0.6.zip (version may vary)

Verification with Eval Checks

After the build process, the skill references skills/no-ai-slop/eval.md to perform self-validation. This markdown file contains compliance checklists ensuring that any edit respects the writer's voice and removes only listed AI-slop patterns, functioning as a quality gate before distribution.

Summary

• Edit Mode: Invoke /no-ai-slop followed by text to automatically strip binary contrasts, colon reveals, and throat-clearing phrases according to SKILL.md.
• Detect Mode: Use /no-ai-slop is this slop? to identify specific AI patterns without automatic rewriting, receiving structured feedback based on the pattern list in the skill configuration.
• Local Build: Run python scripts/build_plugin.py to package the manifest, logic files, and validation rules into a distributable ZIP for ChatGPT or Claude plugin stores.
• Architecture: The demo is prompt-only, requiring no external runtime—all logic lives in .codex-plugin/plugin.json, SKILL.md, and eval.md.

Frequently Asked Questions

What platforms support the no ai slop demo?

The demo runs on any AI assistant platform compatible with Codex-style skills, including ChatGPT and Claude. Because the implementation is prompt-only (defined in SKILL.md and plugin.json), it requires no server infrastructure or API keys beyond the host platform's standard plugin interface.

How does the skill decide what counts as slop?

The detection criteria are explicitly listed in skills/no-ai-slop/SKILL.md. This file defines specific patterns such as binary contrasts (It’s not X. It’s Y.), colon reveals (The best part: ...), and throat-clearing openers (Here’s the thing...), along with instructions for removing or flagging each pattern while preserving the author's original voice.

Can I modify the editing rules in the demo?

Yes. Since the core logic is pure markdown, you can edit SKILL.md to add new prohibited patterns or adjust editing principles, then rebuild the package using scripts/build_plugin.py. The eval.md file provides validation checkpoints to ensure your modifications still maintain voice consistency and don't introduce new artifacts.

Is there any external code execution when running the demo?

No. The no ai slop demo is entirely a prompt-only implementation. When you run the edit or detect commands, the host AI (ChatGPT/Claude) interprets the markdown rules directly. The Python build script only packages static files; it does not run during actual text processing or editing operations.

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