How OpenAI Agents Are Configured for No AI Slop: A Complete Technical Guide

OpenAI agents are configured for No AI Slop through declarative YAML interface definitions that map the /no-ai-slop command to a default prompt, requiring no code changes to the underlying model.

The No AI Slop skill in the petergyang/no-ai-slop repository exposes text editing capabilities to OpenAI-based agents through a lightweight configuration layer. Understanding how OpenAI agents are configured for No AI Slop reveals a declarative approach that separates skill logic from agent integration, allowing the same editing rules to work across different deployment contexts. The configuration leverages YAML interface files and plugin metadata to register the skill with the ChatGPT/Codex ecosystem.

YAML Interface Definitions

The configuration relies on two complementary YAML files that define the agent interface. Both files follow the same schema to ensure consistency whether the skill operates as a standalone agent or as part of a larger package.

Global Agent Configuration

The agents/openai.yaml file serves as the global entry point for OpenAI runtime discovery. This file declares the skill's display name, description, and the default prompt that prepends user requests:

interface:
  display_name: "/no-ai-slop"
  short_description: "Remove AI slop while preserving the writer's personal voice."
  default_prompt: "Use $no-ai-slop to remove AI patterns from this draft while preserving my personal voice."

The $no-ai-slop placeholder signals the runtime to process the following text through the No AI Slop skill engine rather than standard completion logic.

Skill-Specific Configuration

The skills/no-ai-slop/agents/openai.yaml file mirrors the global definition, ensuring the skill can be loaded both as a top-level agent and as an embedded package component. This redundancy allows flexible deployment scenarios while maintaining identical interface contracts.

Plugin Metadata Registration

The .codex-plugin/plugin.json file completes the integration by registering the skill for the ChatGPT/Codex ecosystem. This metadata file references the YAML configurations and establishes the skill's capabilities within the OpenAI plugin architecture. The plugin system scans this file during agent startup to discover available skills and their associated prompt templates.

Runtime Execution Flow

When OpenAI agents are configured for No AI Slop, the runtime processes user requests through four distinct phases:

1. Agent startup – The OpenAI runtime scans the agents/ directory and loads any YAML files matching the interface schema, populating the available command registry.

2. Command matching – When a user types /no-ai-slop …, the runtime looks up the display_name field in the loaded configurations and retrieves the associated default_prompt.

3. Prompt injection – The runtime expands the $no-ai-slop placeholder into a call to the underlying skill, combining the default prompt with the user's input. This ensures the model applies slop-removal rules rather than generating new content.

4. Skill execution – The skill processes the text using the logic defined in skills/no-ai-slop/SKILL.md and validated by skills/no-ai-slop/eval.md, then returns the cleaned version to the agent for relay back to the user.

Implementation Examples

The configuration supports both interactive chat usage and programmatic API access.

Chat Interface Usage

Users invoke the skill directly through the command interface:

User: /no-ai-slop
It’s not X. It’s Y. Here’s the thing, you should listen.

The agent processes this through the configured default prompt and returns:

It’s not X. It’s Y. You should listen.

Programmatic API Access

Developers can leverage the same configuration through direct API calls:

import openai

prompt = "/no-ai-slop\n" + """It’s not X. It’s Y. Here’s the thing, you should listen."""
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[{"role": "user", "content": prompt}]
)

print(response.choices[0].message.content)

Summary

• Dual YAML structure – Both agents/openai.yaml and skills/no-ai-slop/agents/openai.yaml define the interface, supporting standalone and embedded deployments.
• Declarative configuration – The .codex-plugin/plugin.json registers capabilities without requiring model modifications.
• Placeholder system – The $no-ai-slop token bridges agent configuration and skill execution.
• Validation pipeline – Execution relies on SKILL.md for logic and eval.md for output validation.

Frequently Asked Questions

What files are required to configure OpenAI agents for No AI Slop?

The configuration requires three core files: agents/openai.yaml for global agent definition, skills/no-ai-slop/agents/openai.yaml for skill-specific settings, and .codex-plugin/plugin.json for plugin metadata registration. These files work together to expose the /no-ai-slop command to OpenAI-based agents.

How does the $no-ai-slop placeholder function in the configuration?

The $no-ai-slop placeholder in the default_prompt field acts as a runtime marker that tells the OpenAI agent to route the following content through the No AI Slop skill engine rather than processing it as a standard completion request. This mechanism connects the declarative YAML configuration to the imperative skill logic defined in SKILL.md.

Can the No AI Slop skill work without the YAML configuration files?

No. The OpenAI agent runtime specifically scans for YAML interface definitions in the agents/ directories during startup. Without these configuration files, the runtime cannot recognize the /no-ai-slop command or inject the appropriate default prompts required to activate the skill's editing capabilities.

Where is the actual editing logic defined if not in the YAML files?

The editing logic resides in skills/no-ai-slop/SKILL.md, which contains the detailed rules for pattern matching, replacements, and voice preservation. The skills/no-ai-slop/eval.md file provides validation tests to ensure output quality. The YAML configuration files only handle the interface layer, while these markdown files contain the operational intelligence.

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