Awesome Codex Subagents 🤖 Agent Open source

What it does

Awesome Codex Subagents is a curated catalogue of specialized subagent definitions written for OpenAI's Codex CLI. Each subagent is a narrow expert (an API designer, a backend developer, a security auditor, a docs researcher, and so on) that Codex can delegate part of a task to, so the main session stays focused while the specialist works in its own isolated context. The collection is aligned with the official Codex subagents documentation and organized into thirteen categories spanning core development, language specialists, infrastructure, quality and security, data and AI, developer experience, meta and orchestration, governance, platform engineering, and observability.

What is inside

The repository ships around 175 subagents as Codex-native .toml files. Each file carries a name, a description of when it should be invoked, a routing model field, a reasoning-effort setting, a sandbox mode (read-only for reviewers and auditors, workspace-write for developers), and a long instruction block with role-specific checklists and patterns. The README documents example delegation workflows for PR review, bug investigation, and repo exploration, since Codex does not spawn custom subagents automatically and expects explicit delegation prompts.

Works with

OpenAI Codex CLI. Files are copied into ~/.codex/agents/ for global use or .codex/agents/ for a single project, with project definitions taking precedence over global ones.

How to install

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

Maintenance and safety

MIT licensed and updated recently. The maintainers state plainly that subagents are contributed by the community, are not audited for security or correctness, and come with no warranty, so review a definition before adopting it. Sandbox modes limit filesystem access per agent, but the agents run whatever tools your Codex session permits.

Who should use it

Codex users who want a ready-made bench of task-specific experts to delegate to, rather than writing every subagent definition themselves.