Mehmet TURAÇ

Posted on May 18 • Originally published at github.com

eoc: Building the Operating System for OpenAI Codex — 230+ Skills, 60 Agents, Cross-Harness

#opensource #ai #codex #developerproductivity

The Problem

AI coding agents — OpenAI Codex, Cursor, Gemini CLI, Copilot — are powerful out of the box. But raw power without structure leads to:

Inconsistent output — the same task produces different results across sessions
No institutional memory — every session starts from zero
Security blind spots — no guardrails on what agents can touch
Tool lock-in — skills and workflows trapped in one harness

I built eoc (Everything OpenAI Codex) to solve this. It's an MIT-licensed workflow system that turns any AI coding agent into a repeatable engineering environment.

GitHub: mturac/everything-openai-codex

What eoc Ships

Component	Count	Purpose
Agents	60	Scoped instruction sets per role (reviewer, builder, resolver)
Skills	230	Reusable workflows across 12 language ecosystems
Rules	110	Language-specific coding standards
Hook matchers	28	Runtime gates and guards
Install modules	29	Selective component installation
Legacy shims	75	Backward-compatible command mappings

Cross-Harness Architecture

The core insight: skills shouldn't be locked to one tool. eoc includes adapters for:

OpenAI Codex (.codex/)
Cursor (.cursor/)
Gemini CLI (.gemini/)
GitHub Copilot
OpenCode (.opencode/)
Zed (.zed/)
Trae (.trae/)

Each adapter maps eoc's skill format to the target harness's native configuration. Write a skill once, use it everywhere.

# Install for Codex
eoc-install --profile core --target codex

# Same skills, different harness
eoc-install --profile core --target cursor

Quality Gates & Safety

eoc enforces safety at multiple levels:

Hook System

# Runtime gating profiles
eoc_HOOK_PROFILE=standard   # minimal | standard | strict

# Disable specific hooks
eoc_DISABLED_HOOKS=supply-chain,mcp-health

Hooks run at session start, during execution, and at stop. They check:

AGENTS.md scope — is this agent allowed to touch these files?
No-verify blocks — prevent accidental force pushes
MCP health — are connected tools responsive?
Supply-chain scanners — dependency vulnerability checks
AgentShield — security scanning for agent actions

PromptGuard

A built-in audit system for prompts. Treats every prompt as an executable contract and flags:

Vague intent
Conflicting rules
Context retention illusions
Missing verification criteria

Session Memory & State

Long-running tasks need continuity. eoc provides:

Session capture — structured recording of what happened
Compaction prompts — summarize context before it's lost
Observer memory — persistent state across agent restarts
Status snapshots — eoc status --markdown --write status.md
SQLite state store — queryable history with CLI

# Take a snapshot
eoc status --markdown --write status.md

# Query past sessions
eoc sessions list --limit 10

Skill Categories

Backend

Django patterns, Spring Boot, NestJS
REST API design, GraphQL, gRPC

Frontend

Next.js Turbopack, Bun runtime
Presentation builder (HTML slides)

Security

AgentShield integration
Security scanning pipelines

ML/Deep Learning

PyTorch patterns
MLOps workflows

Operations

Brand voice management
Social graph ranking
Customer billing ops
Google Workspace integration

Content & Media

Article writing engine
Market research automation
Manim video generation
Remotion video creation

Installation

Important: Use ONE install method only. Don't stack them.

Option 1: Plugin Install (Recommended)

For Codex plugin users:

/plugin install eoc@eoc

Option 2: Manual Installer

# Shell
./install.sh --profile minimal --target codex

# PowerShell
.\install.ps1 --profile minimal --target codex

# npm
npx eoc-install --profile minimal --target codex

Profiles

Profile	What's included
`minimal`	Core skills, no hooks
`core`	Skills + rules + hooks
`full`	Everything including operator workflows

Discover Components

# Ask the advisor
eoc consult "security reviews" --target codex

Returns matching components, related profiles, and install commands.

eoc 2.0: Rust Control Plane

The eoc2/ directory contains a Rust-based alpha control plane:

cd eoc2 && cargo build

# Available commands
eoc2 dashboard   # GUI overview
eoc2 start       # Start session
eoc2 sessions    # List sessions
eoc2 status      # Current state
eoc2 stop        # End session
eoc2 resume      # Continue session
eoc2 daemon      # Background service

Alpha stage — usable but not production-ready.

Dashboard

A Tkinter-based desktop GUI:

npm run dashboard
# or
python eoc_dashboard.py

Features dark/light theme toggle, font customization, and project branding.

Language Ecosystem Support

Language	Rules	Agents	Skills
TypeScript	✓	✓	✓
Python	✓	✓	✓
Go	✓	✓	✓
Java	✓	✓	✓
Kotlin/Android/KMP	✓	✓	✓
C++	✓	✓	✓
Rust	✓	✓	✓
PHP	✓	—	✓
Perl	✓	—	✓
Shell	✓	—	✓

Release History

Version	Date	Highlights
v2.0.0-rc.1	Apr 2026	Dashboard GUI, operator workflows, eoc 2.0 alpha
v1.9.0	Mar 2026	Selective install, 6 new agents, SQLite state store
v1.8.0	Mar 2026	Hook reliability overhaul, 997 tests
v1.7.0	Feb 2026	Codex app + CLI support, presentation builder
v1.6.0	Feb 2026	AgentShield, GitHub Marketplace launch
v1.4.0	Feb 2026	Multi-language rules, installation wizard
v1.3.0	Feb 2026	Full OpenCode integration
v1.2.0	Feb 2026	Python/Django, Java Spring Boot support

Localization

README and documentation available in:

English, Turkish, Portuguese (Brazil)
Simplified Chinese, Traditional Chinese
Japanese, Korean, Russian, Vietnamese, Thai

v1.4.0 added complete Chinese translations of all agents, commands, skills, and rules (80+ files).

Try It

# Quick start
git clone https://github.com/mturac/everything-openai-codex
cd everything-openai-codex
./install.sh --profile minimal --target codex

# Or via plugin
# /plugin install eoc@eoc

License: MIT