Best practices for wrangling OpenAI’s Codex CLI

What it does This repo is a living best-practice manual for OpenAI’s Codex CLI. It pairs conceptual explanations with working TOML configs, custom agents, reusable skills, and hooks meant to live in a .codex/ directory. The goal is to give teams a reference architecture for consistent, multi-agent coding sessions instead of ad-hoc prompting.

The interesting bit The author treats Codex CLI as a platform that needs governance—complete with orchestration diagrams, Starlark execution rules, and a frank inventory of what actually ships versus what is still experimental. A working Agent → Skill weather demo is included, alongside an honest note that custom commands are documented but not yet publicly available, so the full Command → Agent → Skill chain remains blocked.

Key highlights

• Working implementations of subagents, skills, MCP servers, hooks, and marketplace configs
• Explicit version tracking against Codex CLI releases (currently v0.125.0)
• Coverage of beta features like memories, hooks, and parallel MCP calls with feature-flag caveats
• Comparative survey of other agentic workflows (Superpowers, Spec Kit, gstack)
• AGENTS.md project-context patterns and sandbox approval policies

Caveats

• Requires an existing Codex CLI setup; this is structured documentation and config templates, not a standalone tool
• Several showcased features (hooks, memories, marketplace) are gated behind beta flags or pending release

Verdict Bookmark this if you are already using Codex CLI and want to stop reinventing prompts. Look elsewhere if you need a drop-in framework—this is curated knowledge, not a product.