opencode-openai-codex-auth

opencode-openai-codex-auth

Skill by ara.so — Codex Skills collection.

Overview

opencode-openai-codex-auth is a TypeScript-based OAuth authentication plugin that connects OpenCode to OpenAI's GPT-5.x and Codex models using your existing ChatGPT Plus/Pro subscription. It provides 22 pre-configured model presets with variant support, eliminating complex API setup while maintaining full model access.

Key Features:

• Official ChatGPT OAuth flow (no API keys required)
• 22 model presets across GPT-5.2, GPT-5.2-Codex, and GPT-5.1 families
• Variant system (none/low/medium/high/xhigh) for fine-grained control
• Automatic token refresh and usage-aware error handling
• Multimodal input support (text, images, files)
• Legacy OpenCode (v1.0.209-) and modern (v1.0.210+) compatibility

Installation

Quick Install (Recommended)

# Modern OpenCode (v1.0.210+)
npx -y opencode-openai-codex-auth@latest

# Legacy OpenCode (v1.0.209 and below)
npx -y opencode-openai-codex-auth@latest --legacy

Manual Installation

# Clone or download the config
git clone https://github.com/numman-ali/opencode-openai-codex-auth.git
cd opencode-openai-codex-auth

# Copy config to OpenCode directory
# Modern:
cp config/opencode-modern.json ~/.opencode/config.json

# Legacy:
cp config/opencode-legacy.json ~/.opencode/config.json

Uninstallation

# Remove plugin only
npx -y opencode-openai-codex-auth@latest --uninstall

# Remove plugin and all configs
npx -y opencode-openai-codex-auth@latest --uninstall --all

Authentication

Initial Login

# Authenticate with ChatGPT account
opencode auth login

This opens your browser to complete OAuth flow. Follow the prompts to authorize OpenCode access.

Verify Authentication

# Check current auth status
opencode auth status

# Refresh token if needed
opencode auth refresh

Logout

opencode auth logout

Available Models

GPT-5.2 Family

gpt-5.2 (General purpose, latest flagship)

• Variants: none, low, medium, high, xhigh
• Use case: General coding, documentation, architecture

gpt-5.2-codex (Code-specialized)

• Variants: low, medium, high, xhigh
• Use case: Code generation, refactoring, debugging

GPT-5.1 Family

gpt-5.1-codex-max (Maximum code performance)

• Variants: low, medium, high, xhigh
• Use case: Complex algorithms, large refactors

gpt-5.1-codex (Balanced code model)

• Variants: low, medium, high
• Use case: Standard coding tasks

gpt-5.1-codex-mini (Faster, lighter)

• Variants: medium, high
• Use case: Quick fixes, snippets

gpt-5.1 (General purpose)

• Variants: none, low, medium, high
• Use case: Mixed coding and documentation

Usage Examples

Modern OpenCode (v1.0.210+)

# Basic usage with model and variant
opencode run "write hello world to test.txt" \
  --model=openai/gpt-5.2 \
  --variant=medium

# Use Codex-specialized model
opencode run "refactor this function for performance" \
  --model=openai/gpt-5.2-codex \
  --variant=high

# File operations with multimodal input
opencode run "analyze this image and extract data" \
  --model=openai/gpt-5.2 \
  --variant=medium \
  --input=screenshot.png

# Maximum quality for complex tasks
opencode run "implement OAuth2 flow with PKCE" \
  --model=openai/gpt-5.1-codex-max \
  --variant=xhigh

Legacy OpenCode (v1.0.209 and below)

# Model name includes variant suffix
opencode run "write hello world to test.txt" \
  --model=openai/gpt-5.2-medium

# High-quality code generation
opencode run "create REST API with Express" \
  --model=openai/gpt-5.2-codex-high

# Fast iterations
opencode run "fix syntax error in app.ts" \
  --model=openai/gpt-5.1-codex-mini-medium

Interactive Mode

# Start interactive session
opencode chat --model=openai/gpt-5.2-codex --variant=medium

# In chat:
> create a TypeScript interface for a User model
> add validation methods
> generate unit tests

Configuration

Modern Config Structure (v1.0.210+)

{
  "models": {
    "openai/gpt-5.2": {
      "provider": "openai",
      "model": "gpt-5.2-base",
      "apiKeyEnv": "OPENAI_SESSION_TOKEN",
      "endpoint": "https://chat.openai.com/backend-api/conversation",
      "variants": {
        "none": {"temperature": 0.3},
        "low": {"temperature": 0.5},
        "medium": {"temperature": 0.7},
        "high": {"temperature": 0.85},
        "xhigh": {"temperature": 0.95}
      },
      "capabilities": ["code", "multimodal"],
      "maxTokens": 8192
    }
  }
}

Legacy Config Structure (v1.0.209-)

{
  "models": {
    "openai/gpt-5.2-medium": {
      "provider": "openai",
      "model": "gpt-5.2-base",
      "apiKeyEnv": "OPENAI_SESSION_TOKEN",
      "endpoint": "https://chat.openai.com/backend-api/conversation",
      "temperature": 0.7,
      "capabilities": ["code", "multimodal"],
      "maxTokens": 8192
    }
  }
}

Environment Variables

The plugin stores the OAuth session token automatically. No manual environment configuration needed.

Token storage location:

~/.opencode/auth/session.json

Custom Configuration

Create or modify ~/.opencode/config.json:

// Add custom model preset
{
  "models": {
    "openai/my-custom-preset": {
      "provider": "openai",
      "model": "gpt-5.2-codex",
      "temperature": 0.6,
      "maxTokens": 4096,
      "topP": 0.9,
      "capabilities": ["code"]
    }
  }
}

Common Patterns

Selecting the Right Model

# Quick tasks, fast iteration → codex-mini or low variant
opencode run "add comments to this function" \
  --model=openai/gpt-5.1-codex-mini --variant=medium

# Standard development → gpt-5.2-codex medium
opencode run "implement user authentication" \
  --model=openai/gpt-5.2-codex --variant=medium

# Complex architecture → codex-max high/xhigh
opencode run "design microservices architecture" \
  --model=openai/gpt-5.1-codex-max --variant=xhigh

# Documentation/mixed → gpt-5.2 or gpt-5.1
opencode run "write API documentation" \
  --model=openai/gpt-5.2 --variant=medium

Batch Operations

# Process multiple files
for file in src/*.ts; do
  opencode run "add TypeScript strict type checks to $file" \
    --model=openai/gpt-5.2-codex \
    --variant=medium
done

Project Scaffolding

# Generate project structure
opencode run "create Next.js 14 project with TypeScript, Tailwind, and shadcn/ui" \
  --model=openai/gpt-5.2-codex \
  --variant=high \
  --output=./new-project

Code Review

# Analyze code quality
opencode run "review this PR for security issues and best practices" \
  --model=openai/gpt-5.2-codex \
  --variant=high \
  --input=git-diff.txt

TypeScript Integration

If you're building tools on top of this plugin:

import { OpenCodeAuth } from 'opencode-openai-codex-auth';

// Initialize auth client
const auth = new OpenCodeAuth({
  configPath: '~/.opencode/config.json'
});

// Authenticate
await auth.login();

// Get session token
const token = await auth.getSessionToken();

// Make authenticated request
const response = await auth.request({
  model: 'gpt-5.2-codex',
  variant: 'medium',
  messages: [
    { role: 'user', content: 'Write a TypeScript function' }
  ]
});

// Refresh token if expired
if (auth.isTokenExpired()) {
  await auth.refresh();
}

// Logout
await auth.logout();

Troubleshooting

Authentication Issues

Problem: opencode auth login fails or times out

# Clear auth cache
rm -rf ~/.opencode/auth/

# Re-authenticate
opencode auth login

# Check for browser issues
opencode auth login --debug

Problem: "Invalid session token" error

# Refresh token
opencode auth refresh

# If refresh fails, re-login
opencode auth logout
opencode auth login

Model Access Issues

Problem: "Model not available" error

• Verify ChatGPT Plus/Pro subscription is active
• Check model name matches exactly (case-sensitive)
• Ensure using correct variant syntax for your OpenCode version

# Modern (v1.0.210+)
--model=openai/gpt-5.2 --variant=medium

# Legacy (v1.0.209-)
--model=openai/gpt-5.2-medium

Problem: Rate limiting or quota errors

• ChatGPT subscriptions have usage limits
• Wait 1-2 hours and retry
• Use lower variants (low, medium) for less resource-intensive tasks
• Switch to codex-mini for quick iterations

Configuration Issues

Problem: Config not loading

# Verify config exists
cat ~/.opencode/config.json

# Reinstall with correct version flag
npx -y opencode-openai-codex-auth@latest  # modern
npx -y opencode-openai-codex-auth@latest --legacy  # legacy

# Check OpenCode version
opencode --version

Problem: Wrong config format

# Check OpenCode version
opencode --version

# If v1.0.210+ but using legacy config:
npx -y opencode-openai-codex-auth@latest

# If v1.0.209- but using modern config:
npx -y opencode-openai-codex-auth@latest --legacy

Performance Issues

Problem: Slow responses

• Use lower variants (low, medium) instead of xhigh
• Switch to codex-mini for simpler tasks
• Check network connection
• Verify ChatGPT service status

Problem: Timeouts

# Increase timeout in config
{
  "timeout": 120000,  // 2 minutes
  "retries": 3
}

Debug Mode

# Enable verbose logging
export OPENCODE_DEBUG=1
opencode run "test command" --model=openai/gpt-5.2 --variant=medium

# Check logs
tail -f ~/.opencode/logs/debug.log

Best Practices

1. Use appropriate variants: Start with medium, scale up only when needed
2. Model selection: Use codex variants for pure code tasks, general models for mixed content
3. Token management: Auth tokens auto-refresh, but monitor session expiry
4. Rate limiting: Respect ChatGPT usage limits; avoid rapid-fire requests
5. Version compatibility: Match plugin config to your OpenCode version
6. Security: Never commit ~/.opencode/auth/ directory; tokens are sensitive

Additional Resources

• Official Docs: https://numman-ali.github.io/opencode-openai-codex-auth/
• Configuration Guide: docs/configuration.md
• Architecture: docs/development/ARCHITECTURE.md
• Troubleshooting: docs/troubleshooting.md
• GitHub Issues: https://github.com/numman-ali/opencode-openai-codex-auth/issues