How to Integrate GPT-Image2 Skill with Claude Code or Codex: Complete Setup Guide

You can integrate the GPT-Image2 Style Library skill into Claude Code or Codex using the skills CLI, npm scripts, or marketplace commands, enabling AI agents to generate optimized image prompts directly from the repository's style definitions.

The GPT-Image2 Style Library in the freestylefly/awesome-gpt-image-2 repository is packaged as a portable agent skill that works across multiple AI coding environments. Located at agents/skills/gpt-image-2-style-library/, this skill reads from data/style-library.json to provide Claude Code and Codex with structured prompt templates for GPT-Image2 generation.

Understanding the GPT-Image2 Skill Architecture

Before installing, it helps to understand how the skill components fit together. The integration relies on a YAML-based skill manifest and a JSON data source that stays synchronized with the web UI.

Skill Definition and Data Sources

The skill architecture centers on two primary files:

• SKILL.md – Located at agents/skills/gpt-image-2-style-library/SKILL.md, this YAML front-matter file defines the skill's name, purpose, workflow, and output defaults. Claude Code and Codex parse this manifest to understand how to invoke the skill.

• data/style-library.json – This master data file contains all style entries, categories, tags, and example case IDs. At runtime, the skill reads this JSON file to assemble prompts dynamically, ensuring agents use the same style definitions as the web interface.

Generation and Installation Scripts

The repository provides automation scripts defined in package.json:

• npm run generate:style-skill – Executes scripts/generate-style-skill.mjs to rebuild SKILL.md and related assets from style-library.json whenever the library changes.

• npm run install:skill – Copies the generated skill files into local agent directories (~/.claude/skills for Claude Code and ~/.codex/skills for Codex).

Installation Methods for Claude Code and Codex

You have three primary ways to integrate the GPT-Image2 skill into your agent environment. Choose the method that best fits your workflow.

Method 1: Using the skills CLI

The skills NPM package provides a universal installer that works for both Claude Code and Codex simultaneously.

Install for specific agents:

npx skills add freestylefly/awesome-gpt-image-2 \
  --skill gpt-image-2-style-library \
  --agent claude-code codex \
  --global --yes --copy

Install for all supported agents at once:

npx skills add freestylefly/awesome-gpt-image-2 \
  --global --all --copy

Explanation: The npx skills add command fetches the repository, extracts the skill definition, and copies it into the global skill directories. The --agent flag targets specific installations, while --all expands to every supported agent including Claude Code, Codex, and Cursor.

Method 2: Claude Code Marketplace Commands

If you prefer working within the Claude Code interface, use the built-in marketplace commands:

/plugin marketplace add freestylefly/awesome-gpt-image-2
/plugin install gpt-image-2-style-library@awesome-gpt-image-2

These commands register the repository in Claude Code's plugin marketplace and install the specific skill version. This method is ideal when you want to manage skills through Claude Code's native interface rather than command-line tools.

Method 3: npm-Based Installation

For developers who prefer global npm packages, the repository provides a thin CLI wrapper:

npm install -g gpt-image-2-style-library
gpt-image-2-style-library install all

This approach installs the package globally, then uses the gpt-image-2-style-library binary to copy the skill into both ~/.claude/skills and ~/.codex/skills directories in one operation.

Regenerating the Skill After Library Updates

When you modify data/style-library.json or add new style templates, you must regenerate and reinstall the skill for changes to take effect in your agents:

npm run generate:style-skill   # Rebuilds SKILL.md from style-library.json

npm run install:skill          # Copies updated skill to agent directories

Run these commands from the repository root after any changes to the master style data. The generation script converts the JSON entries into the structured format that Claude Code and Codex expect in the skill manifest.

How the Skill Works at Runtime

Once installed, the skill operates through a four-phase workflow:

1. Invocation – You prompt the agent with a natural language request like: "Use gpt-image-2-style-library to create an infographic prompt about machine learning."

2. Template Selection – The agent loads style-library.json from the skill directory to identify the appropriate template based on category tags and example case IDs.

3. Prompt Construction – Following the workflow defined in SKILL.md, the skill assembles prompt blocks (subject, composition, style specifications) into a cohesive GPT-Image2 prompt structure.

4. Output – The agent returns a copy-ready prompt formatted for the GPT-Image2 API, which you can use directly in the web UI or through server-side Vercel functions.

The skill maintains consistency between the web interface and agent environments by sourcing all definitions from the single style-library.json file.

Summary

• The GPT-Image2 Style Library skill resides in agents/skills/gpt-image-2-style-library/ and uses SKILL.md for agent configuration along with data/style-library.json for style definitions.
• Installation options include the npx skills add command for universal setup, Claude Code's /plugin marketplace commands for native integration, or global npm installation for development workflows.
• Directory locations for installed skills are ~/.claude/skills for Claude Code and ~/.codex/skills for Codex.
• Update workflow requires running npm run generate:style-skill followed by npm run install:skill whenever the underlying JSON data changes.
• At runtime, the skill parses natural language requests, queries the style library, and returns structured GPT-Image2 prompts ready for API consumption.

Frequently Asked Questions

Where is the GPT-Image2 skill defined in the repository?

The skill definition lives at agents/skills/gpt-image-2-style-library/SKILL.md, which contains the YAML front-matter describing the skill's capabilities, workflow, and default parameters. The actual style data is sourced from data/style-library.json at runtime, ensuring the agent always uses the current template collection.

Can I use the same skill installation for both Claude Code and Codex?

Yes. The npx skills add command with the --agent claude-code codex flag (or --all flag) copies the skill to both ~/.claude/skills and ~/.codex/skills directories simultaneously. Alternatively, running gpt-image-2-style-library install all from the global npm package handles both agents in a single operation.

What happens when I modify the style-library.json file?

Changes to data/style-library.json do not automatically update installed skills. You must run npm run generate:style-skill to rebuild the SKILL.md and related assets from the updated JSON, followed by npm run install:skill to push the regenerated files into your local Claude Code and Codex skill directories.

How does the skill construct prompts for GPT-Image2?

When invoked, the skill reads style-library.json to match your request against available categories and tags. It then follows the workflow defined in SKILL.md to assemble structured prompt blocks—including subject definition, compositional guidelines, and style specifications—into a final prompt string optimized for the GPT-Image2 API.

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