English | 한국어 | 日本語 | 中文 | Deutsch | Français

Looking for Claude Code? → my-claude — same agents as Claude Code plugin with Boss meta-orchestrator

If you are an LLM agent, run the installer entrypoint directly:

curl -fsSL https://raw.githubusercontent.com/sehoon787/my-codex/main/install.sh | bash

my-codex

All-in-one agent harness for OpenAI Codex CLI — install once, get everything.

Bundles 400+ installed agent files (auto-loaded + agent-packs) and 200+ skills from 6 upstream sources into native TOML format. The repo currently contains TOML source definitions; install-time deduplication resolves overlapping destination filenames into the final installed footprint. Codex CLI auto-discovers agents via spawn_agent and routes tasks to the optimal specialist. GitHub Actions CI syncs upstream changes weekly.

Based on the official Codex Subagents specification.

Core Principles

🎯 my-codex Originals

Features built on top of upstream sources — unique to my-codex:

Quick Start

If you are a human

git clone --depth 1 https://github.com/sehoon787/my-codex.git /tmp/my-codex
bash /tmp/my-codex/install.sh
rm -rf /tmp/my-codex

Re-running the same install command refreshes to the latest published main build, replaces only my-codex-managed files in ~/.codex/, and removes stale my-codex skill copies from ~/.agents/skills/ and ~/.claude/skills/.

Agent Packs: Domain specialist agents are installed to ~/.codex/agent-packs/. On first install, my-codex auto-activates a recommended dev set (engineering, design, testing, marketing, support) and remembers it in ~/.codex/enabled-agent-packs.txt.

Skills only (cross-platform)

npx skills add sehoon787/my-codex -y -g

Installs cross-tool skills to ~/.agents/skills/ with auto-symlinks to Codex CLI, Claude Code, Cursor, and other tools. Does not install agents, rules, or config.toml — use install.sh for the full Codex skill bundle.

If you are an LLM agent

Run the installer entrypoint directly.

curl -fsSL https://raw.githubusercontent.com/sehoon787/my-codex/main/install.sh | bash

install.sh is the only installer entrypoint. AI-INSTALL.md is reference documentation and does not install anything by itself.

Key Features

Multi-Agent Orchestration

• spawn_agent: Codex CLI auto-discovers agents from ~/.codex/agents/ and spawns them in parallel for complex tasks
• send_input: Parent-to-child agent communication for iterative workflows
• Agent Packs: Recommended domain specialists auto-activate on first install, and the active set persists across reinstalls

Model-Optimized Routing

• o3 (high reasoning): Complex architecture, deep analysis — mapped from Claude Opus equivalents
• o3 (medium): Standard implementation, code review — mapped from Claude Sonnet equivalents
• o4-mini (low): Quick lookups, exploration — mapped from Claude Haiku equivalents

All-in-One Bundle

• Install provides 400+ installed agent files and 200+ skills instantly
• Bundles 6 upstream sources via git submodules (agency-agents, everything-claude-code, oh-my-codex, awesome-codex-subagents, gstack, superpowers)
• Upstream content pinned via git submodules
• MD-to-TOML conversion handled automatically for non-native sources
• Installs a default git attribution flow so commits touched by real Codex sessions automatically receive AI-Contributed-By: Codex

Codex Attribution

• install.sh installs a codex wrapper plus global prepare-commit-msg, commit-msg, and post-commit hooks in ~/.codex/git-hooks/
• The wrapper records only files that changed during a real Codex session in the current git repository
• Commits that include recorded Codex-touched files get Generated with Codex CLI: https://github.com/openai/codex in the commit message
• The commit hook adds AI-Contributed-By: Codex only when staged files intersect that recorded change set
• my-codex does not change git user.name, git user.email, commit author, or committer identity
• To add an optional Co-authored-by: trailer as well, explicitly set both git config --global my-codex.codexContributorName '<label>' and git config --global my-codex.codexContributorEmail '<github-linked-email>'
• Local git commits cannot summon GitHub's official @codex agent identity directly; GitHub only recognizes co-authors by linked email, and no co-author is added unless you opt in with both settings
• To disable attribution entirely, set git config --global my-codex.codexAttribution false
• my-claude is a separate repository and is not version-managed or updated by my-codex

Core Agents

38 core agents from 4 source repos plus 54 Awesome Core agents are installed to ~/.codex/agents/. Overlapping filenames between sources collapse the final auto-loaded set to 80 installed files. Domain specialists are in agent-packs.

my-codex Core (1) — Boss meta-orchestrator

boss

Oh My OpenAgent / OMO (9) — Sub-orchestrators and specialists

atlas, hephaestus, librarian, metis, momus, multimodal-looker, oracle, prometheus, sisyphus

Oh My Claude Code / OMC (18) — Specialist agents

analyst, architect, code-reviewer, code-simplifier, critic, debugger, document-specialist, executor, explore, git-master, planner, qa-tester, scientist, security-reviewer, test-engineer, tracer, verifier, writer

Awesome Core (9) — Orchestration support from awesome-codex-subagents

agent-installer, agent-organizer, code-mapper, context-manager, error-coordinator, knowledge-synthesizer, multi-agent-coordinator, task-distributor, workflow-orchestrator

Superpowers (1) — Code reviewer from obra/superpowers

superpowers-code-reviewer

Agent Packs (Domain Specialists)

installed pack files across 21 categories are written to ~/.codex/agent-packs/. On first install, my-codex writes ~/.codex/enabled-agent-packs.txt with a recommended dev set and materializes those packs into ~/.codex/agents/ as symlinks. The default favors packs that add distinct specialists beyond the core registry, so heavily overlapping packs such as security and infrastructure remain opt-in. Use the helper to inspect or change the active set:

# View current state
~/.codex/bin/my-codex-packs status

# Enable another pack immediately
~/.codex/bin/my-codex-packs enable marketing

# Or switch profiles at install time
bash /tmp/my-codex/install.sh --profile minimal
bash /tmp/my-codex/install.sh --profile full

Installed Components

The repository source inventory is larger than the install footprint because multiple upstreams ship the same destination filename. install.sh verifies the installed counts above, not the raw source totals.

4 core categories installed to ~/.codex/agents/:

01-core-development (12)
accessibility-tester, ad-security-reviewer, agent-installer, api-designer, code-documenter, code-reviewer, dependency-manager, full-stack-developer, monorepo-specialist, performance-optimizer, refactoring-specialist, tech-debt-analyzer

03-infrastructure (16)
azure-infra-engineer, cloud-architect, container-orchestrator, database-architect, disaster-recovery-planner, edge-computing-specialist, infrastructure-as-code, kubernetes-operator, load-balancer-specialist, message-queue-designer, microservices-architect, monitoring-specialist, network-engineer, serverless-architect, service-mesh-designer, storage-architect

04-quality-security (16)
api-security-tester, chaos-engineer, compliance-auditor, contract-tester, data-privacy-officer, e2e-test-architect, incident-responder, load-tester, mutation-tester, penetration-tester, regression-tester, security-scanner, soc-analyst, static-analyzer, threat-modeler, vulnerability-assessor

09-meta-orchestration (10)
agent-organizer, capability-assessor, conflict-resolver, context-manager, execution-planner, multi-agent-coordinator, priority-manager, resource-allocator, task-decomposer, workflow-orchestrator

Key skills include:

Full Architecture

┌─────────────────────────────────────────────────────────┐
│                    User Request                          │
└─────────────────────┬───────────────────────────────────┘
                      ↓
┌─────────────────────────────────────────────────────────┐
│  Codex CLI (spawn_agent / send_input / wait_agent)      │
│  Auto-discovers ~/.codex/agents/*.toml at runtime       │
│  Routes to optimal specialist based on task description │
└──┬──────────┬──────────┬──────────┬─────────────────────┘
   ↓          ↓          ↓          ↓
┌──────┐ ┌────────┐ ┌────────┐ ┌────────┐
│Single│ │Parallel│ │Parent- │ │Config  │
│Agent │ │Spawn   │ │Child   │ │Control │
│      │ │(multi) │ │Comms   │ │        │
│spawn │ │spawn × │ │send_   │ │config. │
│_agent│ │N       │ │input   │ │toml    │
└──────┘ └────────┘ └────────┘ └────────┘
┌─────────────────────────────────────────────────────────┐
│  Agent Layer (400+ installed TOML files)                  │
│    ├── Auto-loaded: final installed footprint       │
│    └── Agent Packs: final installed footprint      │
├─────────────────────────────────────────────────────────┤
│  Skills Layer (200+ from ECC + superpowers + core + gstack)        │
│    ├── tdd-workflow, security-review, autopilot         │
│    └── pdf, docx, pptx, xlsx, team                     │
└─────────────────────────────────────────────────────────┘

How Codex Multi-Agent Works

Codex Subagents Specification

Codex CLI provides a native multi-agent protocol based on the Codex Subagents specification. The protocol defines five core operations:

Codex CLI auto-discovers all .toml files in ~/.codex/agents/ at runtime. When a task requires specialist expertise, the CLI spawns the matching agent by name and passes it the relevant context.

Agent TOML Format

Every agent is defined as a native TOML file:

name = "debugger"
description = "Focused debugging specialist"
model = "o3"
model_reasoning_effort = "medium"

[developer_instructions]
content = "You are a debugging specialist..."

Key fields:

• name — Agent identifier used by spawn_agent
• description — Used for capability matching
• model — OpenAI model to use (o3, o4-mini)
• model_reasoning_effort — Reasoning level (high, medium, low)
• [developer_instructions].content — The agent's system prompt

Configuration (config.toml)

Global multi-agent settings are defined in ~/.codex/config.toml:

[agents]
max_threads = 8
max_depth = 1

• max_threads — Maximum number of concurrent sub-agents
• max_depth — Maximum nesting depth for agent-spawns-agent chains

Multi-Agent Usage Examples

Single Agent Delegation

> Analyze the auth module for security vulnerabilities

Codex → spawn_agent("security-reviewer")
→ Agent analyzes src/auth/
→ Returns: 2 critical, 1 medium vulnerability

Parallel Spawn

> Run a multi-agent pass: refactor auth, add tests, review security

Codex → spawn_agent("executor") × refactoring
      → spawn_agent("test-engineer") × test writing
      → spawn_agent("security-reviewer") × security audit
→ All 3 run in parallel (max_threads = 8)
→ Results collected and merged

Parent-Child Communication

> Implement payment module, then have it reviewed

Codex → spawn_agent("executor")
      → executor completes implementation
      → send_input(executor, "review needed")
      → spawn_agent("code-reviewer")
      → code-reviewer reviews executor's changes

Complex Orchestration

> Plan the migration, execute it, then verify

Codex → spawn_agent("planner")
      → planner produces migration plan
      → wait_agent(planner)
      → spawn_agent("executor") with plan as context
      → executor performs migration
      → wait_agent(executor)
      → spawn_agent("verifier")
      → verifier checks all migrations applied correctly

Open-Source Tools Used

1. Agency Agents

A library of 180+ business specialist agent personas. Provides specialist perspectives across 14 categories — UX architects, data engineers, security auditors, and more. Converted from Markdown to native TOML via automated md-to-toml.sh pipeline.

2. Everything Claude Code (ECC)

A development framework originally built for Claude Code, providing 125 skills. Claude Code-specific skills were removed; the remaining skills contain generic coding guidance usable across any LLM agent. The rules/ directory is included in the repo as reference material but is not read by Codex CLI.

3. Awesome Codex Subagents

136 production-grade agents in native TOML format. Already Codex-compatible — no conversion needed. Organized across 10 categories from core development to meta-orchestration.

4. Oh My Codex (OMX)

Codex CLI multi-agent harness framework by Yeachan Heo. A Rust/TypeScript runtime providing 36 skills, hooks, HUD, and team pipelines for Codex CLI. Referenced as architectural inspiration for my-codex's harness patterns. Does not provide agent TOML files directly.

5. Oh My OpenAgent (omo)

A multi-platform agent harness by code-yeongyu. The 9 orchestration agents in this repository (atlas, hephaestus, metis, momus, oracle, prometheus, sisyphus, librarian, multimodal-looker) are adapted from omo agents, converted to Codex-native TOML format.

6. OpenAI Official Skills

The official Skills Catalog for Codex provided by OpenAI. Includes specialist skills for document processing, code generation, and development workflows. Can be installed via $skill-installer in Codex CLI.

7. gstack

• garrytan's sprint-process harness with 36 skills
• Code review, QA, debugging, benchmarking, security audit, deployment workflows
• Built-in headless Chromium browser daemon for real browser testing

8. superpowers

• Jesse Vincent's workflow skills library (v5.0.7, MIT)
• 14 skills covering brainstorming, planning, TDD, code review, debugging, and agent orchestration workflows

Troubleshooting

Skills-only recovery

If a tool reports invalid SKILL.md files under ~/.agents/skills/, the most common cause is a stale local copy or stale symlink target from an older install.

Remove the affected directories from ~/.agents/skills/ and matching entries under ~/.claude/skills/, then reinstall:

npx skills add sehoon787/my-codex -y -g

If you use the full Codex bundle, rerun install.sh once as well. The full installer refreshes ~/.codex/skills/ and removes stale my-codex-managed copies under ~/.agents/skills/ and ~/.claude/skills/.

Contributing

Issues and PRs are welcome. When adding a new agent, add a .toml file to the codex-agents/core/ or codex-agents/omo/ directory and update the agent list in SETUP.md.
See CONTRIBUTING.md for PR validation steps and Codex commit attribution behavior.

Bundled Upstream Versions

Upstream sources managed as git submodules. Metadata recorded in upstream/SOURCES.json.

Credits

This repository builds on the work of the following open-source projects:

• agency-agents — msitarzewski
• everything-claude-code — affaan-m
• oh-my-codex — Yeachan Heo
• oh-my-openagent — code-yeongyu
• awesome-codex-subagents — VoltAgent
• openai/skills — OpenAI
• gstack — garrytan
• superpowers — Jesse Vincent
• Codex Subagents Spec — OpenAI

License

MIT License. See the LICENSE file for details.