awesome-codex-subagents README

The project's README from github.com/VoltAgent/awesome-codex-subagents. As one markdown file: README.md.

codex

The awesome collection of 172+ Codex subagents across 13 categories.

Awesome Subagent Count Last Update Discord


Awesome Codex Subagents

This repository serves as the definitive collection of Codex Subagents, specialized AI assistants designed for specific development tasks. Written specifically for Codex and aligned with the official docs.

Installation

Use Codex custom agent directories exactly as documented:

  1. Clone this repository.
  2. Copy the .toml agent files you want into one of the directories above.
  3. Restart or refresh your Codex session if needed.
  4. Delegate explicitly in prompts (Codex does not auto-spawn custom subagents).

Examples:

mkdir -p ~/.codex/agents
cp categories/01-core-development/backend-developer.toml ~/.codex/agents/

mkdir -p .codex/agents
cp categories/04-quality-security/reviewer.toml .codex/agents/

If you use agent configuration in Codex, keep it in .codex/config.toml under [agents] as described in the official docs.

Sponsors
Crawlbase Crawlbase is web data infrastructure trusted by 70,000+ developers. Its Crawling API, MCP server, and integrations give AI agents live access to any webpage — with JavaScript rendering, proxy rotation, and anti-bot protection.
SerpApi SerpApi is a Web Search API for your AI apps. Available in Markdown and JSON for any integration.

👉 You can feature your product here and reach developers using AI coding agents like Claude Code, Codex, Gemini, and more.

Become a Sponsor

Subagent Storage Locations

Type Path Availability Precedence
Project Subagents .codex/agents/ Current project only Higher
Global Subagents ~/.codex/agents/ All projects Lower

Note: When naming conflicts occur, project-specific subagents override global ones.

Subagent Structure

Each subagent uses a Codex-native .toml format:

name = "subagent-name"
description = "When this agent should be invoked"
model = "gpt-5.6-luna"
model_reasoning_effort = "medium"
sandbox_mode = "read-only"

[instructions]
text = """
You are a [role description and expertise areas]...

[Agent-specific checklists, patterns, and guidelines]...
"""

Smart Model Routing

Each subagent includes a model field that automatically routes it to the right model -- balancing quality and cost:

Model When It's Used Examples
gpt-5.6-sol Deep reasoning -- architecture reviews, security audits, financial logic security-auditor, architect-reviewer, fintech-engineer
gpt-5.6-terra Implementation, diagnostics, evaluation, and multi-step professional analysis docker-expert, test-automator, scientific-literature-researcher
gpt-5.6-luna Bounded search, extraction, routing, drafting, and lightweight synthesis search-specialist, docs-researcher, agent-installer

Sandbox Mode Philosophy

Each subagent's sandbox_mode field controls filesystem access:

Categories

01. Core Development

Essential development subagents for everyday coding tasks.

02. Language Specialists

Language-specific experts with deep framework knowledge.

03. Infrastructure

DevOps, cloud, and deployment specialists.

04. Quality & Security — Testing, security, and code quality experts (19 agents)

04. Quality & Security

05. Data & AI — Data engineering, ML, and AI specialists (14 agents)

05. Data & AI

06. Developer Experience — Tooling and developer productivity experts (14 agents)

06. Developer Experience

awesome-codex-subagents (VoltAgent/awesome-codex-subagents)

Searching Database…

6.4K ·GitHub
Awesome-codex-subagents
这是什么？一套为 OpenAI Codex 精心整理的 170 多个专用子代理定义，涵盖编码、基础设施、安全、数据/AI 和开发者体验等开发任务。
这是一个工具，还是只是一组配置文件？它是一组 .toml 配置文件，不是可运行的应用或脚本；每个文件定义一个子代理，由 Codex 从其代理目录中加载。
怎么安装和使用这些子代理？克隆仓库，把想要的 .toml 文件复制到 ~/.codex/agents/（全局）或 .codex/agents/（项目级），然后重启或刷新 Codex 会话，并在提示词中显式委派给该子代理。
除了 Codex 本身，使用这些还需要什么？你只需要一个已有的 Codex 环境；还可以选择在 .codex/config.toml 的 [agents] 下添加代理路由配置。
Codex 会自动帮我挑选合适的子代理吗？不会。Codex 不会自动生成自定义子代理，你必须在提示词中显式引用想要使用的子代理。
项目级/全局优先级会影响运行哪个子代理吗？会。当项目级和全局子代理同名时，项目级的优先。
每个子代理的模型路由是怎么工作的？每个 .toml 文件都有一个 model 字段，将其路由到三个层级之一：gpt-5.6-sol 用于深度推理（如 security-auditor、architect-reviewer），gpt-5.6-terra 用于实现和诊断（如 docker-expert、test-automator），gpt-5.6-luna 用于轻量级搜索/提取任务（如 search-specialist、agent-installer）。
子代理是如何分类组织的？它们被划分为 13 个类别，例如核心开发、语言专家、基础设施、质量与安全、数据与 AI 和开发者体验，每个类别都有自己主题的一组代理。
sandboxmode 是什么，为什么重要？每个子代理都通过 sandboxmode 字段控制文件系统访问权限：只做分析的审查者和审计者使用 read-only（只读），需要创建或修改文件的开发者/工程师则使用 workspace-write（写入工作区）。
我可以分别控制每个项目和全局环境下哪些子代理可用吗？可以。你可以把代理放在 .codex/agents/ 中仅供单个项目使用，或放在 ~/.codex/agents/ 中让所有项目都能使用；同名冲突时，项目级代理优先。
What are similar projects?ruflo, pstack-claude, infisical, evenfire, awesome-agent-skills, static-analysis, Codex-X, coder
Which topics is it in?toml, codex-skills, terraform, subagents, code-review, chatgpt, kubernetes, testing, agents, documentation, devops, awesome-list
Creator
VoltAgent
Size
0.1 MB
Stars
6.4K
Updated
CommitOct 05
Age
7mo
Commits · 90 Days