Delegate specialized tasks to AI sub-agents with fine-grained permissions:

# Single agent execution
codex delegate code-reviewer --scope ./src --budget 40000

# Parallel execution (NEW!)
codex delegate-parallel code-reviewer,test-gen,sec-audit \
  --goals "Review code,Generate tests,Security audit" \
  --scopes ./src,./tests,./package.json \
  --budgets 40000,30000,20000

# Custom agent from prompt (NEW!)
codex agent-create "Find all TODO comments and create a summary report" \
  --budget 50000

• Parallel Execution: Run multiple agents simultaneously for 3x faster completion
• Custom Agents: Create agents from natural language prompts with LLM-powered generation
• Process Isolation: Each agent runs in a separate process for enhanced security

Quick Start: See SUBAGENTS_QUICKSTART.md and PARALLEL_CUSTOM_AGENT_GUIDE.md

Multi-source research with citation and contradiction detection:

codex research "Rust async programming best practices" --depth 3

Features:

• DuckDuckGo HTML scraping (no API key required)
• Smart sub-query generation
• Cross-source validation
• Cited reports with confidence scores

New: Codex itself as an MCP server for sub-agents!

# .codex/agents/my-agent.yaml
tools:
  mcp:
    - codex_read_file       # Full Codex file reading
    - codex_grep            # Full Codex grep
    - codex_codebase_search # Semantic search

Status: 🚧 Implementation in progress Design: _docs/2025-10-11_CodexMCP化設計書.md

This fork maintains dual compatibility with:

• ✅ OpenAI official repository
• ✅ zapabob enhancements

Development Guide: .codex/META_PROMPT_CONTINUOUS_IMPROVEMENT.md

• Continuous Integration: Automated testing on every PR

  • 3 platforms (Ubuntu, Windows, macOS)
  • Clippy lint + Rustfmt check
  • Agent definition validation
  • Integration tests (Deep Research + Sub-Agent)
  • Security audit (cargo-audit)

• Continuous Delivery: Automated releases on tag push

  • Multi-platform binaries (Linux x64, Windows x64, macOS x64/ARM64)
  • npm package generation
  • GitHub Release creation
  • Auto-generated release notes

CI/CD Guide: CI_CD_SETUP_GUIDE.md

Codex is an AI-powered coding assistant that runs in your terminal. It helps you write, understand, and improve code through natural conversation.

• Interactive Chat: Natural language conversations about your code
• Code Understanding: Analyze, explain, and refactor existing code
• File Operations: Read, write, and modify files with AI assistance
• Shell Integration: Execute commands safely with sandboxing
• MCP Support: Extensible via Model Context Protocol
• Multi-Model: Support for GPT-4, Claude, and local models

npm install -g @openai/codex

git clone https://github.com/zapabob/codex.git
cd codex/codex-rs
cargo build --release -p codex-cli
npm install -g ./codex-cli

Important: If you're building on Apple Silicon (ARM64) Macs, follow these additional steps:

Issue: The default configuration in codex-rs/.cargo/config.toml forces target-cpu=x86-64 globally, causing build errors on ARM64 Macs:

error: 'x86-64' is not a recognized processor for this target (ignoring processor)

Complete Build Steps for Apple Silicon:

1. Edit the config file: Remove or comment out the [build] section in codex-rs/.cargo/config.toml:

# Remove or comment out these lines:
# [build]
# # Use generic x86-64 instruction set to avoid ILLEGAL_INSTRUCTION errors
# rustflags = ["-C", "target-cpu=x86-64"]

2. Build the project:

cd codex/codex-rs
cargo build --release -p codex-cli

3. Copy binary to npm vendor directory:

cd ..
mkdir -p codex-cli/vendor/aarch64-apple-darwin/codex
cp codex-rs/target/release/codex codex-cli/vendor/aarch64-apple-darwin/codex/codex
chmod +x codex-cli/vendor/aarch64-apple-darwin/codex/codex

4. Install globally:

npm install -g ./codex-cli

5. Verify installation:

codex --version

# Start interactive session
codex

# Non-interactive execution
codex exec "Add error handling to main.rs"

# Deep research
codex research "Topic to research"

# Delegate to sub-agent
codex delegate code-reviewer --scope ./src

# Resume previous session
codex resume

# Login
codex login

# View status
codex login status

# Configure model
codex -c model="gpt-4" "Your prompt"

• Getting Started
• Installation Guide
• Configuration
• Advanced Usage
• FAQ

• Sub-Agents Quick Start
• Requirements Specification
• Implementation Plan
• Codex MCP Design

┌─────────────────────────────────────────┐
│           Codex CLI (Node.js)           │
│  codex, codex exec, codex delegate      │
└──────────────────┬──────────────────────┘
                   │
┌──────────────────▼──────────────────────┐
│      Codex Core (Rust)                  │
│  ┌──────────────┐  ┌─────────────────┐ │
│  │ AgentRuntime │  │ ModelClient     │ │
│  │  Sub-Agents  │  │  LLM Interface  │ │
│  └──────────────┘  └─────────────────┘ │
│  ┌──────────────┐  ┌─────────────────┐ │
│  │ Deep Research│  │ MCP Integration │ │
│  └──────────────┘  └─────────────────┘ │
└─────────────────────────────────────────┘

We welcome contributions! Please see:

• Contributing Guide
• Development Workflow

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

• OpenAI - Original Codex project
• Anthropic - Claude model support
• Contributors - All contributors to the project

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Twitter: @zapabob