This repository offers a comprehensive collection of over 130 specialized Codex subagents, designed to enhance AI-assisted development workflows. It targets developers using the Codex environment, providing task-specific AI assistants that improve accuracy, efficiency, and consistency across a wide array of programming and operational tasks.

How It Works

The project curates Codex subagents defined in .toml files, each tailored for specific development use cases. These subagents leverage Codex's custom agent directory structure. Key design elements include smart model routing, which directs tasks to appropriate models (e.g., gpt-5.4 for deep reasoning, gpt-5.3-codex-spark for lighter tasks) to balance performance and cost. Additionally, a configurable sandbox_mode (read-only or workspace-write) controls filesystem access, ensuring agents operate safely. Subagents are not auto-spawned; explicit delegation in prompts is required.

Quick Start & Requirements

To install, clone this repository and copy the desired .toml agent definition files into either ~/.codex/agents/ for global availability or .codex/agents/ for project-specific use. A Codex environment is required. Subagents must be explicitly invoked in prompts.

Highlighted Details

• Features 136+ specialized Codex subagents organized into 10 categories, including Core Development, Language Specialists, Infrastructure, Quality & Security, Data & AI, Developer Experience, Specialized Domains, Business & Product, Meta & Orchestration, and Research & Analysis.
• Supports smart model routing to optimize for quality and cost based on task complexity.
• Includes configurable sandbox modes (read-only, workspace-write) for controlled filesystem access.
• Agents are designed for explicit delegation, requiring users to specify which subagents to use in their prompts.

Maintenance & Community

Contributions are welcomed via pull requests. A Discord community is available at https://s.voltagent.dev/discord. The repository also links to related collections for AI agent skills, Claude subagents, OpenClaw skills, and AI agent papers.

Licensing & Compatibility

The collection is distributed under the MIT License, which is permissive for commercial use and integration. Compatibility is specific to the Codex environment.

Limitations & Caveats

All subagents are provided "as is" without warranty, and the maintainers do not audit or guarantee their security or correctness. Users are responsible for reviewing subagents before use and accept all liability for any issues arising from their deployment. Explicit delegation is necessary for subagent activation.