oh-my-customcodex (baekenough) — Summary

oh-my-customcodex is a Codex-native (GPT-based) port of oh-my-customcode, maintaining the same agent compilation metaphor but running on top of GPT Codex + OMX stack instead of Claude Code. It ships 49 agents, 123 skills, and 22 rules accessible via a single install command (omcustomcodex init). The distinctive philosophy: "agent systems are compiled, not configured" — skills are source code, agents are compiled output, and mgr-sauron acts as the compiler/verifier. The reasoning-sandwich pattern (Opus for pre-analysis, Sonnet for implementation, Haiku for post-verification) and mgr-creator (dynamic agent creation when no specialist exists) are the two most novel components. The npm package ships a dedicated CLI binary omcustomcodex and maintains multi-model routing (opus/sonnet/haiku per role).

Differs from seeds: Closest to BMAD-METHOD (large specialist agent roster with routing) and claude-flow (CLI binary + npm package), but uniquely frames agent systems as a compilation problem (skills=source, agents=compiled artifacts, sauron=compiler), implements dynamic agent creation (mgr-creator) for missing domain expertise, and runs on Codex/OMX rather than Claude Code — a provider boundary that distinguishes it from all 11 seeds.

oh-my-customcodex (baekenough) — Overview

Origin

A child package / port of oh-my-customcode, converting the Claude Code native harness to run on GPT Codex + OMX. The AGENTS.md explicitly states:

"oh-my-customcodex is a child package of oh-my-customcode. The core porting purpose of this repository is: take the Claude Code native harness from oh-my-customcode and make it run on top of a GPT Codex + OMX stack."

Version 0.5.4, MIT license, 3 stars.

Philosophy — "Compiled, Not Configured"

The most distinctive framing in this batch:

"Agent systems are compiled, not configured."

Dynamic Agent Creation

"If it can't be done, make it work. When no specialist exists for a task, oh-my-customcodex does not fail. It creates one."

User: "Review this Terraform module"
  → Routing: no terraform expert found
  → mgr-creator discovers: infra-aws-expert skills + docker-best-practices guide
  → Creates: infra-terraform-expert.md
  → Executes the review immediately
  → Agent persists for future use

Reasoning-Sandwich Pattern

Opus (pre-analysis) → Sonnet (implementation) → Haiku (post-verification)

Response Contract

Every response starts with an agent identification block:

┌─ Agent: Codex (gpt-5.4)
├─ Skill: <active-skill-or-routing-surface>
└─ Status: <current-action-or-verdict>

oh-my-customcodex (baekenough) — Architecture

Distribution

• Type: npm package + CLI binary (omcustomcodex)
• Version: 0.5.4
• Default branch: develop

Install

npm install -g oh-my-customcodex && cd your-project && omcustomcodex init

Directory Tree

oh-my-customcodex/
├── .claude-plugin/       # Claude Code plugin compatibility
├── .codex/               # Codex-specific config
├── .clawhip/             # OMX integration
├── .omx/                 # OMX config files
├── .mcp.json             # MCP server config
├── agents/               # 49 compiled agent definitions
├── skills/               # 123 skill definitions (+ routing)
├── src/
│   └── cli/
│       └── index.ts      # CLI entry point
├── templates/            # Project templates
├── package.json
│   └── bin.omcustomcodex: "./dist/cli/index.js"
├── missions/             # Mission definitions
├── seminar/              # Seminar-style learning
├── shellmark/            # Shell bookmarks
├── dist/                 # Build output
└── README.*.md           # 12-language READMEs

Multilingual Documentation

README available in 12 languages (German, Spanish, French, Italian, Japanese, Korean, Portuguese, Russian, Turkish, Vietnamese, Chinese + English) — most localized framework in this batch.

Required Runtime

• Node.js (npm package)
• Bun (build tooling)
• GPT Codex + OMX (runtime agents)

Target Tools

• Codex (primary)
• Claude Code (via .claude-plugin compatibility)
• OMX orchestrator

oh-my-customcodex (baekenough) — Components

Agents (49)

Each agent uses YAML frontmatter declaring tools, model, memory scope, and limitations.

Skills (123)

Routing Skills (4)

Key Manager Agents

• mgr-creator: Creates new specialized agents when no expert exists for a task
• mgr-sauron: Structural verification and integrity compiler
• mgr-gitnerd: Git workflow management
• mgr-updater: Updates existing agents with new skills
• mgr-supplier: Supplies context and materials to agents

Model Assignment

oh-my-customcodex (baekenough) — Uniqueness

Differs from Seeds

Closest to BMAD-METHOD (49 specialist agents + routing) and claude-flow (npm package + CLI binary + TypeScript), but differs on three axes: (1) "compiled not configured" framing — skills as source code, agents as compiled artifacts, sauron as compiler/verifier, mgr-creator for dynamic agent compilation on demand — no seed has this architecture metaphor; (2) runs on GPT Codex + OMX rather than Claude Code, making it the Codex-native counterpart to the Claude Code-native frameworks; (3) the reasoning-sandwich pattern (opus pre-analysis → sonnet implementation → haiku post-verification) as an explicit named workflow primitive, different from BMAD's persona chain or claude-flow's MCP routing.

The visible response format (┌─ Agent: ... ├─ Skill: ... └─ Status: ...) is also unique: no other seed or batch framework mandates an agent identification header on every response.

Positioning

• A Codex port of oh-my-customcode (parent package), targeting the "same UX on a different provider" use case
• 49 agents covers the broadest domain roster in this batch (languages, backends, frontends, data, DB, infra, QA, security, managers, system agents)
• At 0.5.4, it is early-stage but architecturally ambitious

Observable Failure Modes

1. Dynamic agent quality: mgr-creator creates agents from combining existing skills + guides — the quality of created agents depends on the available skills and the composition logic; poorly assembled agents may execute with incorrect constraints
2. mgr-sauron integrity: If the sauron verifier has incomplete rules or false positives, it may block valid agent configurations
3. Codex-first, Claude-second: Claude Code compatibility is a .claude-plugin/ wrapper around a Codex-native design — feature parity with native Claude Code frameworks is not guaranteed
4. Rule proliferation: 22 rules in .codex/rules/ + 123 skills creates a large knowledge surface; agents may receive contradictory or overlapping instructions
5. Parallel agent conflicts: Up to 4 agents running in parallel on the same codebase may create edit conflicts (no worktree isolation mentioned)